• TABLE OF CONTENTS
HIDE
 Front Cover
 Foreword
 Acknowledgement
 Table of Contents
 Main






Group Title: Technical report - U.S. Geological Survey, Biological Resources Division. Florida Cooperative Fish and Wildlife Research Unit ; 56
Title: Demography and movements of snail kites in Florida
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00000125/00001
 Material Information
Title: Demography and movements of snail kites in Florida
Series Title: Technical report - U.S. Geological Survey, Biological Resources Division. Florida Cooperative Fish and Wildlife Research Unit ; 56
Physical Description: Book
Creator: Bennetts, Robert E.
Kitchens, Wiley M.
Publisher: University of Florida
United States Geological Survey
Place of Publication: Gainesville, Fla.
Publication Date: 1997
 Record Information
Bibliographic ID: UF00000125
Volume ID: VID00001
Source Institution: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: aleph - 003507177
ltqf - AAA0280
ltuf - AME1632
 Related Items
Other version: Alternate version (PALMM)
PALMM Version

Table of Contents
    Front Cover
        Front Cover 1
        Front Cover 2
        Front Cover 3
        Front Cover 4
    Foreword
        Foreword 1
        Foreword 2
        Foreword 3
    Acknowledgement
        Acknowledgement
    Table of Contents
        Table of Contents 1
        Table of Contents 2
        Table of Contents 3
        Table of Contents 4
    Main
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
        Page 80
        Page 81
        Page 82
        Page 83
        Page 84
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
        Page 91
        Page 92
        Page 93
        Page 94
        Page 95
        Page 96
        Page 97
        Page 98
        Page 99
        Page 100
        Page 101
        Page 102
        Page 103
        Page 104
        Page 105
        Page 106
        Page 107
        Page 108
        Page 109
        Page 110
        Page 111
        Page 112
        Page 113
        Page 114
        Page 115
        Page 116
        Page 117
        Page 118
        Page 119
        Page 120
        Page 121
        Page 122
        Page 123
        Page 124
        Page 125
        Page 126
        Page 127
        Page 128
        Page 129
        Page 130
        Page 131
        Page 132
        Page 133
        Page 134
        Page 135
        Page 136
        Page 137
        Page 138
        Page 139
        Page 140
        Page 141
        Page 142
        Page 143
        Page 144
        Page 145
        Page 146
        Page 147
        Page 148
        Page 149
        Page 150
        Page 151
        Page 152
        Page 153
        Page 154
        Page 155
        Page 156
        Page 157
        Page 158
        Page 159
        Page 160
        Page 161
        Page 162
        Page 163
        Page 164
        Page 165
        Page 166
        Page 167
        Page 168
        Page 169
        Page 170
        Page 171
Full Text






of Sni KienFoida


















I~./I
EN.*' x


RECOMMENDED CITATION
Bennetts, RE and WM. Kitchens. 1997 The Demography and Movements of Snail Kites in Florida. US G.S.
Biological Resources Division, Florida Cooperative Fish & Wildlife Research Unit Tech. Rep. No. 56. 169 pp


~::
-.










THE DEMOGRAPHY AND MOVEMENTS

OF SNAIL KITES IN FLORIDA



Robert E. Bennetts
and
Wiley M. Kitchens



U.S. Geological Survey/Biological Resources Division
Florida Cooperative Fish & Wildlife Research Unit

Technical Report No. 56
1997



Prepared In Cooperation with:

U.S. Fish & Wildlife Service
Jacksonville Ecological Services Field Office
and
South Florida Ecosystem Field Office

National Park Service
Everglades National Park
and
Big Cypress National Preserve

U.S Army Corps of Engineers


U S Geological Survey


Soulh Florida Waler Managemeni Dislrcl

SI. Johns River Waler Managemenl DIslirci



QX~



















Foreword


In preparing this report, we recognized that our audience consisted
of both scientists and managers. We also recognized that there often is a
significant gap in the information exchange between these disciplines.
Consequently, the format and style of this report represent an attempt to
bridge that gap. To this end, we have tried to provide sufficient detail of our
methodology and results to enable the scientific community to evaluate the
validity of our research or to incorporate our findings into a wide variety of
potential models. At the same time we have also tried to provide "down to
earth" explanations, and to frame our research in such a way as to enable
managers to incorporate our findings into realistic management scenarios
and planning. For those with a strong quantitative science background, we
hope that our explanations do not border on being offensive. For those with
a background of more applied biology, we hope that we have retained the
essence of the biology within a potential myriad of statistics. Our goal is to
make this report usable to all and that requires a balance in presentation.
We hope that we have succeeded in our attempt.
We also realize that there often is some reluctance (and rightly so)
on the part of scientists to make conclusions when they know that the
temporal or spatial scale of their research precludes valid inferences beyond
the scope of the conditions in which their research was conducted. At the
same time, management agencies urgently need answers to questions that
may be a long way from being answered. Hence, we are often forced to
balance the need for answers with the validity of inferences derived from
research conducted within a limited set of conditions. Again, we have
attempted to bridge this gap through our interpretation and conclusions of the
patterns we have observed. This occurs primarily in the Management and
Conservation section of this report and we have tried to be very clear about
what is speculative and what we have actually observed. We anticipate and
welcome challenges to our ideas. That is part of the scientific process and
how we advance in our knowledge.





UNIVERSITY OF

FLORIDA







Executive Summary


Florida's wetlands have undergone extensive
anthropogenic change over the past century including
drainage, impoundment, changes in water flow regimes,
increased nutrient loadings, and invasion of exotic plants
and animals. The Snail Kite (Rostrhamus sociabilis),
like many other species, is potentially influenced by these
environmental changes. Snail Kite populations during
this century have changed considerably in number and
distribution and several authors have suggested that
changes in kite populations correspond with changes in
environmental conditions, particularly hydrology. Our
knowledge, however, of demographic processes and
their influences has been far from complete. In addition
to demographic parameters, movements of Snail Kites
are poorly understood and have been the subject of
recent controversy during the planning of wetland
restoration (e.g., the Everglades) within Central and
South Florida. The underlying purpose of this study was
to better understand snail kite population and spatial
dynamics and how they are influenced by environmental
conditions. This understanding combined with reliable
estimates of demographic parameters would enable a
wide range of demographic modeling (e.g., viability
analyses and risk assessments) with a higher degree of
confidence. It also increases our predictive capability
regarding the response of snail kites to changes in water
management. The primary objectives of this study were
to (1) estimate demographic parameters with an emphasis
on survival, (2) evaluate the influences of environmental
conditions (e.g., hydrology) on survival, (3) evaluate the
movement patterns of snail kites in Florida including
rates, locations, and what environmental conditions are
correlated with these movements, and (4) develop a
protocol for future monitoring of snail kites.
Radio telemetry and mark-resighting (banding)
were the two primary field methods we used to obtain
estimates of survival and movement probabilities. The
combination of these two methods provided a
comprehensive assessment of these parameters and was
complimentary to each other. Radio telemetry enabled
assessment of "within-year" patterns of both survival and
movement This allowed us to determine such things as
the cause of death of an animal or what the
environmental conditions were at the time an animal
moved. Mark-resighting is intended to assess "between-
year" patterns of these parameters. Our goal was to
annually (for three consecutive years) capture and radio
tag 100 snail kites of which 60% were adults and 40%
juveniles. Our targeted ratio of adults to juveniles was
intended to emphasize adult survival because


demography of long-lived avian species (e.g., snail kites)
tends to be more sensitive to adult rather than juvenile
survival. Additionally, we targeted a 50:50 sex ratio of
adults to keep our sample balanced. Our annual sample
size of 100 was based on estimates of the statistical
power to distinguish survival differences among groups
(e.g., age or sex) or time periods.
We attached 282 radio transmitters on 271
individual Snail Kites. Eleven birds were recaptured and
re-tagged in a subsequent year. We were short (82%) of
our targeted sample size of 100 birds during 1992, but
fully attained our targeted sample sizes in 1993 and
1994. We were very close to our targeted age and sex
ratios for all years. Our total sample of individual
banded birds for our mark-resighting analyses of survival
was 913. Of this sample, 191 were adults at the time of
addition to our sample and 722 were juveniles.
We estimated adult survival to be an average of
approximately 0.90 using radio telemetry and 0.92 mark-
resighting. Estimates of juvenile survival were not as
consistent between methods and averaged 0.71 using
radio telemetry and 0.50 using mark-resighting. We
present evidence that the radio telemetry estimates may
be biased high. Both sources of data indicated that
survival was age-dependent and that survival differed
among years for juveniles but not adults. Survival was
high enough that within-year patterns would have been
difficult to detect without enormous effort (i.e., too few
individuals died to enable quantitative comparisons with
those that survived) and probably not have been very
insightful. Based on our results, we believe that
between-year differences in survival would be more
detectable and more insightful. From the outset of this
study we realized that our scope of inference would be
limited to those conditions encountered during our study.
We encountered relatively high water conditions
throughout this study. Consequently, we were unable to
derive inferences about drought conditions. Fortunately,
the mark-resighting program should enable a long-term
evaluation of environmental correlates (e.g., hydrologic
conditions) as variable conditions occur in the future.
Our data on movement indicated that Snail Kites
frequently move among wetlands throughout their range.
Overall probabilities of movement (per month) averaged
approximately 0.25 for adults and 0.20 for juveniles.
Our results also indicated that the probability of
movement is influenced by age, time (yearly and
seasonal differences), and location. Although relatively
high water conditions persisted throughout this study,
water levels (independently of location) did not appear to







influence monthly movement probabilities. This does not
imply that such an influence would not occur under low-
water conditions. Some shifting of birds among regions
within the state also may have been a result of a drought
that preceded our study. Dispersal is generally thought to
be favored when local resources (e.g., food) are low or
better conditions exist elsewhere. In contrast, our results
from both within-year and between year comparisons
suggest that higher probabilities of movement occur
when food resources are high. We also found that natal
dispersal of juveniles was lower in areas where food
resources were likely depressed. We suggest the
hypothesis that this may be a reasonable strategy given
the dynamic and unpredictable nature of a kite's
environment. In years that food is not limiting, which
for kites may be most years, high food resources may
enable kites to "explore" their potential habitats with little
risk. The resulting experience may then help kites to
locate food resources faster during times (e.g., droughts)
when food is limiting.
Prior to 1969 the statewide Snail Kite population
was monitored only through sporadic and haphazard
surveys. Numbers of Snail Kites in Florida since 1969
have been monitored via a quasi-systematic annual
survey. Since these surveys began there have been
numerous biological interpretations, often with little or
no regard for the inherent sources of variation in these
data that could influence the validity of subsequent
interpretations. We examined several sources of
variation inherent in the annual survey and discuss how
this variation could influence the validity of data
interpretations. Based on our results and on logic of valid
scientific inference, we suggest the annual survey is not
a valid estimator of population size; nor should year to
year variation in the count be used to estimate
demographic parameters (i.e., survival or recruitment).
We do, however, believe that the annual count has some
value for examining long term population trends
provided that the sources of variation be incorporated
into any analysis. We explore alternatives to the annual
count and provide recommendations that are dependent
on what parameters are being estimated.
There has been considerable discussion in the
literature about the influence of drought on Snail Kite
populations; however, few authors have even defined a
drought sufficiently to enable an independent observer to
designate a given year as a "drought year" based on
objective criteria. We suggest that there are three
essential characteristics of droughts (i.e., intensity,
spatial extent, and temporal extent) that should be
operationally defined for effective evaluation of
droughts. We further suggest ways that these
components can be measured such that they may be


incorporated into demographic analyses.
Our data are consistent with previous views that
the habitats used by Snail Kites in Florida are
considerably more extensive than the currently-
designated-critical habitat. We also believe that the
protection of only the currently designated critical habitat
would be insufficient to maintain viable populations of
Snail Kites over the long term. We suggest that the use
of habitats can be characterized as an extensive network
and present a hypothesis of how the spatial and temporal
patterns of this network might influence viability of Snail
Kites in Florida. We believe that the general directions
and goals of the South Florida Ecosystem restoration
process are not in conflict with maintaining a viable
population of Snail Kites; however, a broader spatial
extent of habitat protection (i.e., outside of the
Everglades and Okeechobee watersheds) probably is
necessary for long-term viability of the Florida
population.













ACKNOWLEDGMENTS

We are very grateful to the many people who helped us during this study.
Financial support was provided by the U.S. Fish and Wildlife Service (USFWS),
National Park Service (NPS), U.S. Army Corps of Engineers (USACOE), South
Florida Water Management District (SFWMD), St. Johns River Water
Management District (SJRWMD), and the Biological Resources Division (BRD) of
the U.S. Geological Service. John Ogden (SFWMD) and David Wesley (USFWS)
were largely responsible for getting this project started and continued to provide
strong support throughout its duration. We are also especially grateful to Reid
Goforth (BRD), Steve Miller (SJRWMD), Ed Lowe (SJRWMD), Mary Ann Lee
(SJRWMD), Jon Mouldling (USACOE), Lewis Hornung (USACOE), Peter David
(SFWMD), Dale Gawlik (SFWMD), Paul Warner (SFWMD), and Jim Brown
(USFWS). We greatly appreciate the help of our field biologists Phil Darby, Patty
Valentine-Darby, Katie Golden, Steve McGehee, Scott Severs, Hilary Maier, David
Boyd, James Conner, and Lynn Bjork. Their ability to work independently for long
hours, and get the job done made our job much easier. We also appreciate the
volunteer assistance from Bob Dill and Theresa Johnson.
This project has been a cooperative effort among biologists and agencies
from the outset. For their help in the field and/or logistic support we are grateful
to Brian Toland (USFWS), Tim Towles (GFC), Laura Brandt (UF), Peter Frederick
(UF), Marilyn Spaulding (UF), Mary Beth Mihalik (West Palm Beach Solid Waste
Authority), Al Vasquez (West Palm Beach Solid Waste Authority), Deborah Jansen
(NPS), Mike Wilson (NPS), Sue McDonald (NPS), Vivie Thue (NPS), Fred
Broerman (Arthur R. Marshall Loxahatchee National Wildlife Refuge), Angela
Chong (SFWMD), Vicky Dreitz (Univ. Miami), F.K. Jones (Miccosukee Tribe of
Indians), and Steve Terry (Miccosukee Tribe of Indians). The banding of Snail
Kites was conducted in cooperation with the GFC. In this effort, we appreciate the
cooperation of James Rodgers Jr. (GFC), Jon Buntz (GFC), and Brian Toland
(USFWS). We greatly appreciate the effort of Charlie Shaiffer (Mingo National
Wildlife Refuge) who took the time to travel to Florida to share his knowledge of
trapping and handling birds. We are grateful to Patuxent Wildlife Research Center,
particularly Jim Nichols and Jim Hines, for housing and assistance during the
analysis phase of this project.
We appreciate the helpful comments on drafts of this report by Jim Nichols
(BRD), Steve Miller (SJRWMD), Don DeAngelis (BRD), Dale Gawlik (SFWMD),
Jim Rodgers Jr. (GFC), and Vicky Dreitz (UM).
For allowing us access to areas used by kites, which were closed to public
access, we are grateful to the Miccosukee Tribe of Indians, the City of West Palm
Beach, and the Arthur R. Marshall Loxahatchee National Wildlife Refuge.
We are grateful to our pilots Karen Dunne and Morton Sund of Wyatt
Aviation. We appreciate their patience, competence, and many hours of safe flying.
We are grateful to the employees of the Florida Cooperative Fish and
Wildlife Research Unit, particularly Barbara Fesler and Debra Hughes, for their
help with administration of this study.








Table of Contents



FORWARD ........................................................................................................................ i

EXECUTIVE SUMMARY ............................................................... .................................... ii

ACKNOW LEDGMENTS........................................................................................................... iv

Chapter 1. INTRODUCTION ................................................................................................. 1

Objectives .............. ..................................................................................... ................... 2

Chapter 2. STUDY AREA ........................................................................................................ 5

Study Population ......................................................................... ................................... 5

Spatial Scales ............................................ .............................................. ................ 5

Regions ............................................................................................................... 5

Habitat Types .................................... .......... ........................... 6

Chapter 3. METHODS ............................................................................................................. 9

Overview of Field Methods .................................... ......................................... 9

Capture and Marking of Animals ...................... ........................................ 10

Sampling Protocols ....................................................... .............................................. 10

Monitoring Protocols ..................................................................... ............................. 11

Estimation of Survival ................................................................. ............ ................... 11

The Kaplan-Meier Estimator ............................................................. ........................ 11

The Cormack-Jolly-Seber Model ............................................................ 12

Mean Life Span .................................................................................. 13

Estimation of Natal Dispersal of Juveniles ................................................ 14

Estimation of Movement Probabilities ........................ .................... ................ 14

Model Selection as a Basis for Data Analysis ..................... .................. .................. 14

Analysis Philosophy ..................... ..................................... 14

The Likelihood Ratio Test ................................................ ........................................ 15

Akaike's Information Criteria ..................... .......................................... 15

Estimating Power .................................................. .............................................. 16

Logistic Regression and Log-linear Models ........................................... ................... 17

The Analysis of Residuals from Cross-classification Models ................................ ........ 17

Multiple Comparisons ................................... ......................................... 17







Chapter 4. SURVIVAL AND MORTALITY ................................................................................. 18

Longevity ........................................... ... .. ....... .......................... ..................... 18

Mean Life Span ................................................................................................................... 19

Estimation of Survival from Radio Telemetry ............................................................................ 19

Effects of Age on Survival ............................. ............................. ........................ 21

Effects of Sex on Survival ......................... .......................................... 22

Temporal Effects of Survival ........................... ...................................... 22

Spatial Effects of Survival ........................ .......................................... 24

Model Selection for Effects on Survival .................................... ............................... 30

Effects of Hydrology ................................................. ........................................ 30

Estimation of Survival from Banding Data .................................................... ........................ 31

Effects of Age and Time ............................................ .................................. 31

Effects of Sex ............................................................ ............................................. 34

Regional Effects ........................................................................ 34

Conclusions About the Effects of Survival from Banding Data ................................ ...... 38

A Synthesis of the Effects of Survival .................................... .................................... 38

Effects of Age ............................................................ ............................................ 38

Effects of sex ........................................ .................................. .. ..................... 38

Temporal Effects ............................................... .................................................. 38

Assumptions, Bias, and Sources of Error ....... ............ ......... ..................... 38

Assumptions Inherent in the Study Design for Valid Inferences From Survival Analyses ............... 38

Assumptions of the Kaplan-Meier Estimator ..................................... .................. 39

Assumptions of the Cormack-Jolly-Seber Models ...................... ................ ......... .......... 43

Causes of Mortality ........................................ ........................................ ................... 44

Previous Estimates of Survival ............................................... ............................................ 46

Chapter 5. REPRODUCTION ........................................................... ........... ...................... 48

Semantics .......................................................... ............................................... 48

The Breeding Season .................................................... .............................................. 50

The Breeding Population .................................................. ............................................... 51

Age of First Reproduction .............. .......................... .. ............................. 51

Proportion of Birds Attempting to Breed ........................ ........ ........ ..................... 51

Nest Success ................................................................................. .................................. 52

Areas of Disagreement Regarding Estimates of Nest Success ................................. ....... 52








Estimates of Nest Success and its Process Variance ............................................................ 54

Influences of Nest Success ............................... ........................................ ... 55

Number of Young Per Successful Nest ........................................ ... ................... 58

Number of Nesting Attempts Per Year ........................ ..... ......... ................................. 58

Chapter 6. MOVEMENTS ................................................................................................. 64

Natal D ispersal ......................................................................... ...................................... 64

Temporal Patterns of Natal Dispersal ............................................................ 65

Differences in Natal Dispersal Between Northern and Southern Regions ................................... 65

Differences in Natal Dispersal Between Lake and Marsh Habitats ......................................... 66

Discussion of Natal Dispersal ......................... ............................ 66

Movement Probabilities ..................................................................................................... 67

The Effect of Age and Sex on Movement Probabilities ............................................... 67

Temporal Effects on Movement Probabilities ............................... ..................... .... 68

Spatial Effects on the Probability of Movement .............................................. 69

Hydrologic Effects on the Probability of Movement .................................. ............... 72

The Effect of Food Resources on Movement ....................................... ................... 74

Model Selection and Synthesis of Effects on the Probability of Movement ................................ 75

Spatial Patterns of Movement ................................... ..................................................... 78

Effect of Distance ............................................................... ............. .................. 78

The Effect of Age, Sex, and Time on Movements between Specific Locations ........................... 80

Seasonal Shifts in Latitude ...................................................... 80

Shifts in Regional Use ....................................................................................... 81

Seasonal Shifts in Habitat Use .................................... ...................................... 82

Natal Philopatry and Site Fidelity ..................... .................................. 84

Natal Philopatry ............................................................................. ................... 84

Site Fidelity ........................................................... 85

Assumptions, Bias, and Sources of Error ..................................... .......................... 89

Effects of Radio Transmitters on Movement of Snail Kites ........................................... 89

Conditional Independence ................................................................. ................... 91

M monthly Time Steps ............................................................ .............. ................ 91

Chapter 7. MONITORING THE FLORIDA SNAIL KITE POPULATION ...................................... 92

The Annual Survey ......................................................... ........................................ .... 92

Sources of Variation ........................................................... .............. ................. 93


vii








Some Alternative Field Methods to the Annual Survey ................................... ........................ 97

Radio Telemetry ................................................................................ 97

Capture-Recapture (Mark-Resighting) Data From Banding .................................................. 98

Distance Sampling ............................................................... .....98

Discussion and Recommendations ................................................................ .. 100

Population Size .......................................................................... .................... 100

Population Indices .......................................................... ............ ................... 101

Population Change and Viability ..................................................... .......................... 101

Estimates of Demographic Parameters .............................................................................. 103

External Influences on the Population ............................ ................... ........ ............... .... 105

Conclusions ................................................................................ .......... ............ 105

Chapter 8. MANAGEMENT AND CONSERVATION .................................................................... 106

W ater Management and Snail Kites ................................................... 106

Drought Semantics ............................................. .................................. .......... 106

Hydrologic Regimes of Snail Kite Habitat ........................................ ................ 110

The Hydrologic Window ............................................... ........................................... 113

Critical Habitat .......................................................................................... ........... .......... 116

Current Designation ............................................... ........................................... 116

The Habitat Network .............................................................................................. 116

Meta-habitats: a Hypothesis about the Relationship Between the Habitat Network and Meta-
population Structure ........................................ ......... .............................. 118

Protection of Habitat ........................................... ......... 123

South Florida Ecosystem Restoration and Snail Kites ...................................... ................ 125

LITERATURE CITED .................................................................... .................... ............. 126

APPENDICES ................................................................................ ................................. 135




































Chapter 1. INTRODUCTION


Florida' weldands have undergone extensive
anthropogemr change over the past century Icludmg
drainage, impoundmen, changes I water flow re-
gmes, increased nutrient loadlngs, and inasion by
exoic plants d and ams TheSnadlKile(Roprrhamus
ouabtasa), Itke many other species, is poaenlially in-
fluenced by Ihese environmental changes. Snail Kite
population dunng th century have changed consid-
erably in number and distribution and several authors
(e g, Sykes 1984, Belssinger 1988; Bennetts et al



knowledge however of demographic processes and
their sllduencss s far rom complete (Bennetts and
Kitchens 1994)
Changes the size of all populations are
sum of births and immigration mlnu deaths and em-
grallon The Florda populatan ofSnal Kaes, how-


ion and emigration Snai Kates n lorida have long


ng to our previous suggestion that the Florida popu-
lation is not compr sed of d\etesubpopulatlons, but
instead, is one population tha frequently shifts n dis


change between population of the United Staes and


pothess hasemerged Thus, fom a demographic per-
spcctive, we are concerned priaanly with birth and
death process and the nencesonthose process
(Fig 1-1)
The bth and death process can be concept






wilh a higher degree of confidence It also would
















































increase our predictive capability regarding the response
of Snail Kites to changes in water management. Given
the scope of projects currently being planned or
implemented (e.g., the Central and South Florida
Project, the South Florida Ecosystem Restoration
Initiative, Kissimmee River Restoration, Upper St. Johns
River Basin Project, Kissimmee chain of Lakes Fishery
Restoration) an improved predictive capability would be
highly beneficial and would greatly reduce controversies.
The goal of this study was to better understand
Snail Kite population and spatial dynamics and how they
are affected by both natural and anthropogenic
processes. We believe that demographic models play an
important role in the refinement of our understanding of
these dynamics. However, it is also our belief that
reliable parameter estimates, particularly if a model is
sensitive to those parameters, are an essential basis for


reliable model outputs. Lastly, we believe that our
models, as well as our knowledge, should be an iterative
and adaptive process (Walters 1986). As we acquire
new information or better parameter estimates, or if our
predictions are falsified, we need to adjust our models,
as well as our thinking, to adapt to new information (Fig.
1-3).




Conservation
Strategy

Parameter Population
Estimation Modelg Projection

Risk Assessment



Figure 1-3. Conceptual framework for this study. Reliable
estimation ofparameters is the first step toward the
development of a wide variety of demographic models.


Objectives

Most previous research on the demography of
Snail Kites has focused on reproduction. Nesting
success, in particular, has received considerable attention
in recent years (e.g., Sykes 1987b, 1987c, Bennetts et
al. 1988, 1994; Snyder et al. 1989a). There remains
considerable debate about the what factors influences
nesting success (Bennetts et al. 1994, Sykes et al. 1995);
however, compared to other species, the relationship
between nesting success of Snail Kites and environmental
conditions is relatively well understood. Other
reproductive parameters are less well known.
Unsubstantiated estimates or speculation have been made
regarding the proportion of birds attempting to breed
each year and the number of nesting attempts per year
(e.g., Snyder et al. 1989a, Beissinger 1995); however,
reliable estimates for these parameters have been
lacking.
In as much as there appears to be general
agreement that changes in Snail Kite populations are
more sensitive to survival than to reproduction (Nichols
et al. 1980, Beissinger 1995, Sykes et al. 1995), data to
estimate survival are very limited (Snyder et al. 1989a)
and as a result, reliable estimates of survival are sorely
lacking (Beissinger 1995). Consequently, the first
objective of this study was to estimate adult and juvenile


Figure 1-2. The demographic cycle of Snail Kites showing
three age classes (Juveniles =J, Subadults=S, and
Adults=A). Parameters for survival (0) and fecundity () are
shown for each age class. Adapted from Caswell (1989),
Beissinger (1995) and Legendre and Clobert (1995).







survival and to evaluate the influences of environmental
conditions (e.g., hydrology) on survival. In addition to
this primary goal, we also recorded supplementary
information on reproductive parameters to the extent that
it did not conflict with accomplishing our primary goals
and in areas where such information was not already
being collected.
In addition to demographic parameters,
movements of Snail Kites also are poorly understood and
have been the subject of recent controversy during the
planning of marsh restoration within Central and South
Florida. While long term changes in Snail Kite
distribution tend to coincide with changes in hydrologic
regimes, shorter term (e.g., annual and seasonal) shifts
do not always coincide with local hydrologic conditions
(Bennetts et al. 1994). It has been hypothesized that
dispersal of kites may be in response to hydrologic
conditions (Takekawa and Beissinger 1989), localized
food depletion (Bennetts et al. 1988), or localized
environmental conditions (e.g., dissolved oxygen in the
water) that may influence apple snail (Pomacea
paludosa) availability (Bennetts et al. 1994). To what
extent movements reflect long-term changes in habitat
quality versus short-term environmental dynamics is
poorly understood, as is their ability to locate and re-
colonize wetlands that have been, or will be, restored.
Thus, movements are critical to understanding Snail Kite
population dynamics leading to our second primary
objective to evaluate the movement patterns of Snail
Kites in Florida including rates, locations, and what
environmental conditions are correlated with movements.
Because of the kites' endangered status and
because there are a multitude of projects than have and
continue to alter hydrologic regimes of southern Florida,
there has been considerable interest in monitoring the
Florida population of Snail Kites. Since 1969, Snail
Kites in Florida have been monitored using an annual
statewide survey. Unfortunately, there are serious
questions regarding the inferences that can be reliably
made from this data source. Consequently, our third
primary objective was to evaluate the validity of existing
monitoring and to provide recommendations for future
monitoring of Snail Kites in Florida.














' *








Chapter 2. STUDY AREA


Study Population

Snail Kites within the United States occur only
in Florida (Sykes 1984). It has been suggested (Bennetts
and Kitchens 1992, 1993, 1994, Beissinger 1995) that
Snail Kites comprise one population that shifts in
distribution throughout the state, rather than there being
separate subpopulations within the state. Data from
studies on movements (this study) and genetics (Rodgers
and Stangel 1996) support that there is considerable
interchange of birds among wetlands in Florida.
Consequently, it was deemed essential for the scope of
this study to include the entire population of Snail Kites
in Florida and our study area comprised a network of
wetlands throughout Central and South Florida within the
entire documented range of Snail Kites (Fig. 2-1).


Spatial Scales

Because the scale of our study is statewide, we
did not focus on movements within individual wetlands.
For the purpose of this study, we considered wetlands to
be distinct if they were separated by a physical barrier
(e.g., ridge or levee) and/or were under a different
hydrologic regime either through natural or managed
control. Thus, adjacent wetlands, which were once
hydrologically continuous (e.g., WCA-2A and WCA-
2B), were considered separate units if they were under
different water regulation schedules.
Although we recorded locations of animals by
specific wetland, for many analyses we had insufficient
data to consider wetlands individually. For example, in
the highly fragmented agricultural areas, there were
more than 50 wetlands used by kites during this study.
Consequently, some pooling of locations was required.
For most analyses agricultural areas were pooled into a
single class of wetlands. It was not uncommon for kites
to frequent several such wetlands in immediate
proximity, and we seldom (if ever) would have had
sufficient data to support estimating parameters (i.e.,
survival or movement probabilities) for each of these
wetlands. Other cases of pooling are described below,
or are reported on a case specific basis based on model
selection criteria (see methods).


T ToH0 ET'OO Atlantic
ss Ocean






WPB. -

'Lox
HOLEY_
2A
Gulf of I, :;2-
Mexico BIC







Figure 2-1. Major wetlands of South Florida referred to in
this report. Wetlands are Everglades National Park (ENP),
Big Cypress National Preserve (BICY), Water Conservation
Areas 3A, 3B, 2B, 2A, Loxahatchee National Wldlife Refuge
(LOX), Holey Land Wildlife Management Area (HOLEY),
West Palm Beach Water Catchment Area (WPB), Lake
Okeechobee (OKEE), Upper St. Johns [Blue Cypress] Marsh
(SJM), Lake Kissimmee (KISS), Lake Tohopekaliga (TOHO),
and East Lake Tohopekaliga (ETOHO).


REGIONS

For some analyses (e.g., survival) we treated
location at a regional scale because it was infeasible to
estimate separate parameters for all wetlands. Based
primarily on watersheds, climatic factors, physiography,
and management regimes we assigned each location to
one of five primary regions (Fig. 2-2). Locations not
included in these five regions (e.g., agricultural areas
and isolated peripheral wetlands) were assigned to a sixth
region we call the peripheral region. Undoubtedly, there
are differences in the quantity and quality of habitats
within this sixth "catch all" region (and within the 5
primary regions as well); however, the amount of data
required to partition the effects of this within-region
variability would be enormous and require significantly
more effort that the scope of this study. However,




































whenever the data supported partitioning beyond a
regional scale we did so.
The Everglades and Big Cypress Region is
comprised of Water Conservation Areas 1,2, and 3,
Everglades National Park, and Big Cypress National
Preserve. The Loxahatchee Slough Region is comprised
of wetlands in the drainage system of the Loxahatchee
Slough and vicinity including the Corbitt Wildlife
Management Area, Pal-Mar Water Control District,
private wetlands owned by Pratt-Whitney Corp., and
wetlands within the Loxahatchee Slough owned by the
City of West Palm Beach (i.e., the West Palm Beach
Water Catchment Area and vicinity). The Okeechobee
Region is comprised of Lake Okeechobee within the
Herbert Hoover Dike. The Kissimmee Chain-of-Lakes
Region was comprised of all lakes within this chain
including Lakes Kissimmee, Tohopekaliga, East
Tohopekaliga, Marion, Marian, Tiger, Pierce, Jackson,
and Hatchineha. The Upper St Johns Region includes
wetlands within the Upper St. Johns River Basin, but
most Snail Kites used the Blue Cypress Marsh Water
Conservation Area, Blue Cypress Water Management
Area, and surrounding wetlands in private ownership.
Agricultural areas (e.g., citrus groves, canals,
agricultural fields, or agricultural retention ponds) within
each of these regions, as well as all other areas not


included in one of the above regions, were assigned to
the peripheral region.


HABITAT TYPES

Snail Kites inhabit freshwater wetlands
throughout central and south Florida. There is
considerable variation in the physiographic
characteristics and specific plant communities that
comprise Snail Kite habitat (reviewed by Sykes et al.
1995). Our objectives did not warrant documentation of
micro-habitat use by kites, nor was our sampling (often
by aircraft) conducive to recording such data. However,
for some analyses we wanted to incorporate the effects
of at least a broad classification of habitats being used by
kites. This classification had to be broad enough to
enable assignment of locations obtained from aircraft to
a given habitat type and sufficiently broad such that
micro-habitat variation did not confound the assignment
given normal daily movements of foraging birds.
Consequently we assigned each location to one of five
habitat types: (1) graminoid marsh, (2) cypress prairie,
(3) Okeechobee, (4) northern lakes, and (5)
miscellaneous peripheral. Graminoid marshes (Fig. 2-3)
were generally slough and wet prairie communities
(Loveless 1959). We distinguished cypress prairies (Fig.
2-4) in that a dominant feature of the landscape profile
was comprised of cypress. This habitat occurred
primarily in western WCA3A, and portions of the Big
Cypress National Preserve and Loxahatchee Slough.
The littoral zone of Lake Okeechobee is an extensive
system of diverse marsh habitats, and consequently had
elements of at least three of our other habitat types (i.e.,
graminoid marsh, northern lake, and highly disturbed).
Because of this high local diversity we were unable to
assign locations to a particular type without extensive
ground verification. Even then, birds often used more
than one of these habitat types within a given day. Thus,
we assigned locations at Lake Okeechobee (Fig. 2-5) to
its own habitat type. The northern lake habitat type (Fig.
2-6) consisted primarily of lakes within the Kissimmee
Chain-of-Lakes, but also included a few lakes along the
Lake Wales Ridge. In contrast to Lake Okeechobee, this
habitat type generally was comprised of a narrow littoral
zone (usually < 200 m) on the periphery of these lakes.
This littoral zone had a relatively steep elevation gradient
compared to other habitat types; the zone used by
foraging kites often was a band of < 100 m usually
dominated by maidencane (Panicum spp) interspersed
with patches of bulrush (Scirpus spp) or cattail (Typha
spp). Primary nesting areas were often a zone of cattail







and/or willow (Sahx spp) in the shallower zone adja-
cent to foraging areas. The peripheral habitat type (Fig.
2-7) was comprised primarily of agricultural areas.
These included retention ponds for citrus groves, ag-
ricultural ditches, and other miscellaneous, usually
highly disturbed, habitats. Larger canals, not neces-
sarily associated with agriculture, were also included
in this habitat type.
For some analyses we had insufficient data to














Figure 2-3 Graminoid marsh habitat type.














Figure 2-4. Cypress prairie habitat type.


partition locations into each of these habitat types. Con-
sequently, for some analyses we assigned locations to
an even broader category of lakes (i.e., Lake
Okeechobee, the northern lake habitat type, and per-
manently flooded canals [<0.01% of our locations])
and marshes (any non-lake habitat). This was intended
to distinguish habitats that had a permanent water
source component available (even if it was not used)
with those that dried periodically.












Fgure 2-5 Lake Okeechobee habitat pe.












Figure 2-6. Not themlkake habitua type.












Figue 2-7. Peripheral habi/,tat txe.


~. ~ .1r






































Chapter3. METHODS


To provide sufficient detail of our methods and
still keep it readable for general audiences we have
split up some of the information between the text and
appendices. In the text, we have tried to provide a
more general description of our methods including the
appropriate citations for analytical procedures. How-
ever, for some analyses we have provided a more de-
tailed description, including the corresponding fmnnu-
lae in appendices. Because we also make use of exten-
sive notation, which for readers not familiar with such
notation, can be confusing, we have also provided a
summary (Appendix 3-1) of notation used through-
out this report.

Overview of Field Methods

Radio telemetry and mark-resighting (band-
ing) were two primary field methods used to estimate
survival and movement probabilities. The combina-
tion of these two methods provided a comprehensive


assessment of these parameters and were complemen-
tary to each other. Radio telemetry enabled assess-
ment of "within-year" patterns of both survival and
movement. This allowed us to determine such things
as the cause of death of an animal or what the environ-
mental conditions were at the time an animal moved.
Mark-resighting is intended to assess "between-year"
patterns of these parameters. This approach was a
much more cost efficient method for estimating an-
nual survival. We are also exploring its potential to
estimate population size of Snail Kites. Within-year
patterns could not be determined using the mark-
resighting approach; however, we had hypothesized
that the primary factor that regulates Snail Kite
populations is periodic drought (Takekawa and
Beissinger 1989, Belssinger 1995), which is a between-
year phenomenon. Thus, the ability to detect differ-
ences and to estimate survival for drought phenomena
can therefore be assessed using mark-resightmg
methodology.


9








Capture and Marking of Animals

We captured adult snail kites primarily using a
net gun (Mechlin and Shaiffer 1979). We initially tried
to capture kites using a variety of noose carpets (Snyder
et al. 1989b) and traps. These methods proved ineffi-
cient and indicated a relatively strong heterogeneity
(catchability) among individuals that could have biased
our sample. Some individuals would actively avoid
perches or food that were associated with traps, while
others showed no apparent avoidance. We decided to
use the net gun after careful evaluation of the potential
risks (i.e., injury, death, or nest abandonment) and ap-
parent poor success and potential for bias in other tech-
niques.
The net gun propels a 10-foot triangular nylon
net using 22 caliber blank cartridges. The projectiles of
the net gun were encased in foam rubber to reduce the
chance of injury. Most captures were of adults while
flying close during nest defense. We also captured some
foraging birds although this was less effective. To fur-
ther reduce the potential for injury, we did not shoot at
birds that were: (1) closer than 7-8 m, (2) in a defense
dive toward us, or (3) in a position where the vegetation
below the bird posed a risk (e.g, too high to retrieve the
bird). We almost always were able to retrieve the bird
within 60 s of capture. In order to reduce the risk of
nest abandonment, we did not attempt to capture birds
in the early stages of nesting (e.g., courtship or egg lay-
ing). We attempted to capture birds primarily after eggs


group defense of an adjacent nest.
All adults captured and most nestlings encoun-
tered at fledging were handed with U.S. Fish & Wild-
lile Service bands as well as a uniquely-numbered-an-
odized aluminum band. The anodized-aluminum bands
enabled an individual to be identified from up to 40 m
(a distance not difficult to obtain with snail kites) using
a spotting scope The banding of nestlngs was con-
ducted in a cooperative effort with the Florida Game
and Fresh Water Fish Commission (GFC) Biologists
with the GFC were banding Snail Kites as part of other
ongoing studies using colors to designate natal wetlands
Consequently, we provided GFC biologists our anod-
i/ed numbered bands which enabled us to identify indi-
viduals in the field using colors needed for their respec-
tive studies.
In addition to leg bands, all adults and some
juveniles (below) were equipped with radio transmit-
ters Radio tagging more than one juvenile per nest
would have allowed us to explore some questions about


the tendency for siblings to disperse together; however,
it would have violated the assumption of independence
required for our survival analysis. Consequently, a
maximum of one juvenile per nest was equipped with a
radio transmitter.
All radio transmitters were equipped with a
mortality switch that upon prolonged lack of motion (=
6 h) altered the pulse rate such that we could remotely
determine if a bird were dead or had dropped its radio.
Radio transmitters were attached using a backpack har-
ness Harnesses were made of four separate pieces of
i'" Teflon ribbon which were securely fastened at the
transmitter, but were sewn together at the breast with a
single cotton stnng. This cotton string attachment was
intended as a "weak link" which would allow the trans-
mitter to fall off cleanly as the string weakened and broke
over time The 15 g transmitters were approximately
3.5% of the body weight of adult snail kites (X, =427
g) and 3.9% of the body weight of juveniles at the time
of fledging (x,-v =384 g) (Darby et al., in press), which
were within the 5% required by our permits.

Sampling Protocols

For our sample to be representative of the state-
wide population we targeted the sample from each lo-
cation to be proportional to the preceding annual count
made approximately 3 months prior to trapping. Our
goal was to annually (flor three consecutive years) cap-
ture and radio tag 100 snail kites of which 60% were
adults and 40% juveniles. Our targeted ratio of adults to







juveniles was intended to emphasize adult survival
because demography of long-lived avian species (e.g.,
snail kites) tends to be more sensitive to adult rather than
juvenile survival (Young 1968, Grier 1980, Beissinger
1995). Additionally, we targeted a 50:50 sex ratio of
adults to keep our sample balanced. Our annual sample
size of 100 was based on estimates of the statistical
power to distinguish survival differences among groups
(e.g., age or sex) or time periods (Fig. 3-1). We
estimated power assuming a binomial distribution
because we intended to use a binomial estimator for
movement probabilities (below) and the Kaplan-Meier
estimator for survival analyses (below) which is an
extension of a binomial estimator that enables censoring
and staggered entry of animals (White and Garrott
1990). Consequently, a binomial estimator was our best
starting point from which to estimate power prior to
actual collection of data. We used an initial survival
estimate of 0.90 as a null hypothesis (H.) based on
previous speculation (Nichols et al. 1980, Snyder et al.
1989a). We then computed power for the alternative
hypotheses (Ha) of 0.80, 0.70 and 0.60 using a range of
potential sample sizes. This analysis indicated that
substantially smaller sample sizes than those we targeted
would have had little power to detect even large
differences in annual survival.
Sample sizes needed for capture-recapture data
are highly dependent on estimates of re-sighting
probability for which we had no preliminary indication.
Consequently, our goal was to band as many birds as
logistically possible to evaluate whether this method had
sufficient power given the sample sizes of banded birds
and resightings we were able to obtain.


Monitoring Protocols

We attempted to locate all radio-tagged kites bi-
weekly. At this frequency we were likely to detect most
major movements and had a reasonable probability of
finding most dead birds before they were too
decomposed to determine cause of death. Given the size
of our study area, more frequent locations would have
required considerably more effort and expense. Also
because of the size of our study area, searches were
conducted most efficiently by aircraft. Even by aircraft,
only a small portion of the potential habitat could be
adequately covered during any given flight.
Consequently, we flew approximately two times per
week (for = 4-5 h each flight) to obtain our bi-weekly
locations. We also verified the status (i.e. alive or dead)
and location of birds on the ground (i.e., by airboat or
from levees) whenever we visited each area.
Mark-resighting was conducted from March
through June of each year. Using this period enabled
simultaneous estimates of survival along with recruitment
through nest monitoring and created additional
advantages. For example, the tendency for a bird to
remain in proximity of its nest increased our ability to
read band numbers (i.e., a resighting) of an individual.
Nest visits were also required for the marking of
individuals (i.e., fledglings), making for an efficient use
of our effort by simultaneously monitoring nesting
success. During nest monitoring we checked the adults
at each nest for leg bands, in addition to other non-
nesting birds observed, and attempted to read the bands
of all birds encountered.


Estimation of Survival

THE KAPLAN-MEIER ESTIMATOR

We estimated survival (0) of radio-tagged kites
using a staggered entry design (Pollock et al. 1989) of
the Kaplan-Meier product limit estimator (Kaplan and
Meier 1958). Detailed descriptions of the estimator and
its properties can be found in Kaplan and Meier (1958),
Cox and Oakes (1984), Pollock et al. (1989a), and
White and Garrott (1990); formulae are provided in
Appendix 3-2.
This estimator is preferred to other estimators
because (1) no assumptions are required about the hazard
function (e.g., constant survival over intervals)(White
and Garrott 1990), (2) the estimator allows for staggered
entry of radios (i.e., radios can be added to the
population at any time)(Pollock et al. 1989), and (3) the


0.4 -d


.... Difference in Surviva of 0.1
--- Diference n Survival of0.2
-- ierence h Surviv of 0.3


20 40 60 80 100 120 140 160 180
Sample Size


Figure 3-1. The probability (power) of detecting differences
in survival based on using different sample sizes. We
assumed a binomial distribution and an initial survival
estimate of 0.9.





------------


... ... ... ... ... ... ... ... ..


..... .... .....






estimator allows censoring (i.e., radio failure or
loss)(White and Garrott 1990). We used an arbitrary
starting date of 15 April for our survival estimates. At
this time we had a reasonable sample (n = 16 during our
1st year) to allow estimation of survival. Birds captured
after 15 April were included in the analysis in
accordance with the staggered entry procedures
described by Pollock et al. (1989) and annual survival
each year was estimated from April 15 to April 14. All
calculations were conducted in SAS (SAS Inc. 1988)
using variations (for our specific data) of the program
provided in White and Garrott (1990).

Comparison among groups- For comparison
among survivorship curves generated by the Kaplan-
Meier estimator we used the log-rank tests (Savage 1956,
Cox and Oakes 1984). For a log-rank test, each time
step is treated as a 2 x 2 contingency table. Detailed
descriptions of the test and corresponding formulae are
provided in Appendix 3-3. While alternative tests are
possible, the log-rank test is easily generalized to the
staggered entry design (Pollock et al. 1989). Cox and
Oakes (1984) describe three variations of the log-rank
test based on slightly different procedures for estimating
the variance for the number of deaths (dy). The tests
differ slightly in their Type I error rate and
corresponding power. Unless otherwise stated, we have
reported only the results from the most conservative test
(i.e., less likely to make a Type I error, but having
lower power). However, when the significance levels
were questionable (i.e., at or near a=0.05) we reported
all three test results for comparison. All of the variations
assume conditional independence and asymptotic
normality of dy (Cox and Oakes 1984, Pollock et al.
1989, White and Garrott 1990).

THE CORMACK-JOLLY-SEBER MODEL

We estimated annual survival from banding data
using the capture-recapture (resighting) models originally
developed by Cormack (1964), Jolly (1965) and Seber
(1965). The basic Cormack-Jolly-Seber (CJS) approach
has undergone extensive advancement in recent years to
become an extremely flexible unified framework capable
of handling a variety of models ranging from simple to
complex models of survival including effects of
individual characteristics (e.g., age and sex),
environmental variables (e.g., weather), and the ability
to incorporate transition probabilities and multiple states
(e.g., exchange among geographically stratified
populations)(Lebreton et al. 1992, Nichols 1992,
Brownie et al. 1993, Nichols et al. 1993). This
increased flexibility has shifted the emphasis of CJS


models from primarily parameter estimation to a
powerful tool for testing biological hypothesis about the
life history parameters being estimated (Lebreton et al.
1992, Nichols 1992).
Unlike radio telemetry data, where survival is
estimated on a relatively continuous-time basis
(estimation is discrete at the interval of obtaining radio
locations), survival from banding data is estimated for
discrete time intervals, which are usually yearly. Our
capture and resighting period corresponded with the peak
time of fledging (March-June). Thus, survival estimates
are not strictly annual, but rather can be roughly
interpreted as survival from one breeding season to the
next, regardless of whether a given animal was breeding.
The basic probabilistic framework for estimating
survival lies in the estimation of two parameters. Let <
denote the probability that an animal is alive and in the
population at time t + 1 given that the animal was alive
at time t, and let p denote the probability that an animal
alive and in the population at time t is seen (i.e.,
recaptured or resighted). As such, the probabilities of
survival (0) for each sampling occasion (for this study
each occasion equals 1 year) can be described (Fig. 3-2).
Unfortunately, without radio telemetry, we usually do
not know if an animal survived over an interval; we only
know whether or not it was observed. We summarize
this knowledge in the form of capture (resighting)
histories for each individual marked during the study
(Fig. 3-3). For example, a capture history of {1 1 1}
indicates that an animal was seen on each of 3 occasions
for a 3-occasion study. Similarly, a capture history of {1
0 1} represents an animal that was seen on occasions 1
and 3, but not seen on occasion 2.
We can then represent each capture history in
the form of a probablistic model using the parameters 0P
and p (Table 3-1). From this probabilistic framework,
parameters were estimated using maximum likelihood
estimation. Maximum likelihood estimates (MLE) in the


Time 4


Time 3


Time 2


Time 1


Arinal
AlAve
ANove<3An"
Not Akive


Ajkrdn WlWi y Alve
cas^ mykeo
and rIasedA101
Animal
Not Alve


Figure 3-2. The probabilistic framework for estimating
survival over 4 time periods. Adapted from Nichols (1992).







Capture
Time 1 Time 2 Time 3 History

Animal 111
Seen
Animalnimal

Animal Initially Not Seen
caught marked,
and released
Animal
Se /^101
< Animal Seen 10
Not Seen Animal
Animal
NotSeen 100
Figure 3-3. Diagram of events that result in each of 4
capture historiesfor animals released at time 1 in a 3-
occasion Cormack-Jolly-Seber (CJS) model. Adapted from
Nichols (1992).


context of survival analyses are described in considerable
detail elsewhere (e.g., White 1983, Brownie et al. 1985,
Pollock et al. 1990, Lebreton et al 1992); however, for
our purposes, it is important to recognize that an MLE is
asymptotically unbiased, normally distributed, and has
minimum variance (White 1982, Brownie et al. 1985,
Lebreton et al. 1992). These properties make MLE well
suited for our purposes compared to other estimators. An
additional advantage of MLE is that it enables the
resulting models to be evaluated in the context of
likelihood-ratio framework for hypothesis testing (White
1983, Lebreton et al. 1992)(see Likelihood Ratio Tests
below). All parameter estimation of CJS models was
conducted in either Program SURVIV (White 1983,
White and Garrott 1990) or MSSURVIV (Hines 1994).


Table 3-1. Possible capture historiesfor animals
released at time 1 and corresponding probabilities
for a Cormack-Jolly-Seber model with 4
occasions. Capture histories and probabilities for
animals released at times 2 and 3 of this model
are not shown..

Capture
Capture Probability
History

1111 12 P, 1P3
1110 01p,2 1101 1P2 02 (1 p3)IP
1100 01p, [l- p ( 1-p3,)3pl
1011 1, (1 -p)02p, A
1010 (1 p2) p, (1 ApO)
1001 (1 p2)2 (1 -p) Op.
1000 1 (all terms above)


MEAN LIFE SPAN


We estimated mean life span (MLS) for adults
using the estimator introduced by Cormack (1964):

MLS. = -



This approach assumes constant survival among years
(Cormack 1964); however, when the above estimator is
considered as an approximation, Brownie et al. (1985)
suggest that this estimate is useful when computed from
mean annual survival (6.). Our data also do not indicate
a violation of the assumption of constancy for adults.
Our data do indicate that juvenile survival is lower than
adults for approximately the first 100 days post fledging
after which it is similar to adult survival. Thus, our
estimate of adult MLS, is applicable to birds conditional
upon their survival of the first 100 days. It has also
been suggested by numerous authors (e.g., Sykes 1979,
Beissinger 1986, Rodgers et al. 1988, Snyder et al.
1989a) that there is increased mortality associated with
droughts. Beissinger (1995) estimated adult survival
during droughts to be 0.60. This was based on the
average difference in the annual count for drought years
after 1973 (Beissinger 1995)(but see discussion in
Monitoring Snail Kite Populations in Florida). Based on
ancillary evidence (e.g., band sightings from known-age
birds), we believe Beissinger's survival estimate for
drought years is substantially low. To our knowledge,
no reliable data exist for estimating survival during
droughts (although we have a mark-resighting program
in place to derive such estimates when future droughts
occur). Consequently, we estimated MLS using a range
of hypothesized estimates for drought-year survival
ranging from Beissinger's (1995) estimate of 0.60 to
estimates that we believe are more reasonable.
Beissinger (1995) also suggested that survival in lag
years (i.e., one-year after a drought) was 0.90 based on
radio telemetry of eight adult Snail Kites that were
radioed during the 1981 drought (7 of 8 survived and the
fate of the other was unknown)(Snyder et al. 1989a,
1989b). We are unclear as to what extent this estimate
applies to lag years versus drought years (the radios were
attached in May and June when water levels were
lowest) and the actual survival of these birds was
0.885 < < 1.00. However, for lack of a better
estimate, we also included this estimate as a hypothesized
value for lag years.
Brownie et al. (1985) derived an analogous
estimator for the expected life time E(T) of juveniles as:






1 i -ln(:) -In(o )


+ JU


Thus, for our application this estimates the expected life
time of a juvenile conditional upon it surviving to
fledging (i.e., when birds are banded). It includes the
first 100 days post fledging, but does not include the pre-
fledging nestling period.

Estimation of Natal Dispersal of
Juveniles

Natal dispersal of birds has been defined as the
dispersal of juveniles from their natal site to the site of
their first reproduction or potential reproduction
(Greenwood and Harvey 1982). Unlike most species for
which natal dispersal has been described, Snail Kites
exhibit nomadic tendencies and may breed colonially or
solitarily throughout their range (Sykes et al. 1995).
Consequently, the assessment of whether or not a site is
a potential breeding site is rather ambiguous. Thus, for
the purposes of this study, we have defined natal
dispersal as the initial dispersal of a bird from its natal
wetland regardless of whether or not there was potential
for breeding at the site to which it dispersed.
We estimated the cumulative probability of natal
dispersal of radio-transmitted birds using the Kaplan-
Meier product-limit estimator (see above). For this
analysis, a juvenile was considered to have dispersed
when it was located alive outside its natal wetland. The
time of dispersal was estimated as the midpoint between
its previous location (in its natal wetland) and the first
location outside of its natal wetland. This analysis only
relates to the initial dispersal from its natal wetland. Any
movements subsequent to initial dispersal were ignored
for this analysis, but were included in our estimation of
movement probabilities (below). Birds were censored if
either we were unable to locate their radio signal or if
they were known to have died prior to dispersal from
their natal area. For clarity, we report these results as
the complement of our estimate (the actual estimate is the
probability of not dispersing). Log-rank tests were used
for comparisons among dispersal functions.

Estimation of Movement
Probabilities

We estimated the conditional probability of
radio-transmittered kites moving (0f) or staying (1 i,)


at a given location over a finite time period t to t + 1
(one month) as a simple binomial parameter, conditional
upon the animal being alive and its location known at t
and t + 1 (Appendix 3-4). A one-month time interval
was based on our sampling frequencies. Although the
average time between consecutive locations was 13.5 days
(our targeted interval was 14 d), there was considerable
variability in this parameter. We estimated the upper limit
of a 95% confidence interval for the time between
locations as 29 days. Thus, we were reasonably certain to
have located all birds within our study area every 29 days.
For the ease of interpreting our results we used calendar
months as the time step for these analyses.
There were alternative approaches to estimating
movement probabilities that did not rely upon a fixed
time interval (i.e., variations of the Kaplan Meier);
however we chose a fixed time interval because it allows
more straightforward interpretations of our results. The
estimates derived from a variable-time estimator would
have been confusing because each estimate would be
associated with a different length of time. One problem
with using a fixed time interval is that some birds may
have moved more than once during our interval. However,
this problem is inherent in our data regardless of what
analysis approach was used because our sampling of radio
locations was not continuous.

Model Selection as a Basis for
Data Analysis

ANALYSIS PHILOSOPHY

Although the philosophy and statistical methods
we used for much of our data analysis have become
increasingly more common in ecological literature and
are described in considerable detail elsewhere (e.g.,
Burnham and Anderson 1992, Lebreton et al. 1992,
Brownie et al. 1993, Nichols et al. 1993, Anderson et al.
1994, Spendelow et al. 1995), some of the statistical
tools we used may not be familiar to some readers.
Consequently, we present a brief general discussion of
our approach. A critical concept to our approach to
data analysis is that a given amount of data will support
only limited inference (Burnham and Anderson 1992).
Given that premise, we treat data analysis as a problem
of optimization, rather than strict elimination of
alternative hypotheses based on arbitrary criteria (e.g.,
c level). We also recognize that data often will suggest
more than one model as being appropriate (McCullagh
and Nelder 1989), and that we must integrate biological
common sense into our statistical inference.







The Principle of Parsimony- We use the
principle of parsimony as a basis for model selection.
We begin with the recognition that all models are wrong
(McCullagh and Nelder 1989). Given this recognition,
our goal is to be able to make generalizations about the
system we are trying to model that are supportable by the
data. If we incorporate all possible effects (including
their interactions) into our model, we will have the best
fit of the data. The problem is that we will not have
learned anything. As the number of parameters in a
model is increased, the fit of the model to data will
improve. The cost, however, for increasing the number
of parameters in a model is that precision of the
parameter estimates decreases (i.e., the confidence
intervals increase)(Burnham and Anderson 1992,
Lebreton et al. 1992)(Fig. 3-4). Thus, the goal of our
analysis is to find a good balance between having enough
parameters in the model to adequately describe the
underlying patterns, but not so many so as to defeat the
purpose of making generalizations. Thus, a
parsimonious model is one which retains only those
parameters that are warranted (judged using the tools
such as likelihood ratio tests and Akaike's Information
Criterion [AIC] described below) and eliminates
excessive parameters (Lebreton et al. 1992).


Few Number of Parameters Many

Figure 3-4. Conceptual tradeoff between precision (low
variance)(doned line) and bias (solid line) as function of the
number ofparameters in a model (After Burnham and
Anderson 1992).



Starting Point-- Whenever possible, we tried to
evaluate a full spectrum of models. However, in some
cases the number of parameters in more saturated
models (i.e., with many main effects and interactions)
was prohibitively large (e.g., >200). These more
complex models also usually contained some parameters
that were not identifiable or estimable (see discussion in


parameter identifiability in Lebreton et al. 1992).
Consequently, when we were unable to evaluate a full
suite of models, we used the starting procedures
described by Hosmer and Lemeshow (1989). Their
approach begins with a univariate assessment of each
main effect. Effects that are significant at a liberal
a =0.25 are retained for further evaluation. This liberal
rejection criterion retains main effects that may have an
influence in an interaction that might have been masked
during initial tests. The next step was to evaluate a
multivariate model with all of the retained main effects,
but lacking interactions. Effects that were retained at
this stage were further evaluated including their
interaction effects. This approach enabled us to reduce
the parameter space for more complex models down to
a more manageable level, while retaining the integrity of
the analysis.

THE LIKELIHOOD RATIO TEST

One of the advantages of maximum likelihood
estimation (MLE) is that it enables a likelihood ratio
testing framework. Likelihood ratio tests (LRT) allow
testing hypotheses about whether the addition of
parameters to a model significantly improves the fit of
that model. The test statistic is:

-21n = -2 [ln(,) ln(2)]



where 3 is the likelihood for a given model evaluated at
its maximum, and is distributed as X2 with np, np2
degrees of freedom, where np is the number of
estimable parameters. The null hypothesis (Ho) is that
the reduced model (i.e., the model with fewer
parameters) fits equally well as the more general model
(i.e., the model with more parameters). Thus, a failure
to reject H, indicates that the additional parameters of the
more general model are not warranted. A limitation of
LRTs is that they are limited to pairwise comparisons
and that the models being compared must be nested (i.e.,
one is a reduced subset of the other). Consequently, for
non-nested models and for multiple comparisons we used
Akaike's Information Criteria (AIC)(below) to compare
models.

AKAIKE'S INFORMATION CRITERION


In contrast to using a LRT for hypothesis testing,
where the intent (in the context used for our analysis) is
to test whether the addition of a particular parameters)
improves the fit of a model, the use of Akaike's






Information Criterion (AIC)(Akaike 1973, Sakamoto et
al. 1986, Shibata 1989, Anderson et al. 1994) provides
a systematic approach to the analysis of complex
multidimensional data, and it removes the limitation of
nested models required by LRTs (Burnham and
Anderson 1992, Lebreton et al. 1992, Anderson et al.
1994).
Akaike's Information Criterion is defined as:

-2 In(Ef) + 2np



where In(2) is the log-likelihood function evaluated at
the maximum likelihood estimates (Akaike 1973). The
first term of AIC (i.e., -2 In(2), called the relative
deviance (Lebreton et al. 1992), is a measure of
goodness-of-fit of the model and the second term (i.e.,
2np) can be viewed as a cost for adding excessive
parameters (Lebreton et al. 1992). The properties and
benefits of using AIC as a model selection tool have been
extensively documented (Shibata 1989, Burnham and
Anderson 1992, Anderson et al. 1994, Burnham et al.
1994).
We emphasize that the notion of finding one
"true" model for complex biological data is unlikely, and
often more than one model may be indicated as being
appropriate for a given data set (Burnham and Anderson
1992). Anderson et al. (1994) suggest viewing model
selection in the analogous context of confidence
intervals, rather than point estimation, such that the
selection process often indicates a range of models that
are appropriate, rather than a single model. Thus, as the
differences in AIC become smaller, so does the statistical
basis for distinguishing among alternative models
(Burnham and Anderson 1992). Sakamoto et al. (1986)
suggests that AIC differences of > 1-2 should be
considered as statistically significant. Models within this
range were consequently judged not to be more or less
suitable based on AIC criteria alone.

ESTIMATING POWER

In the context of hypothesis testing, the results of
a given statistical test can have only one of four possible
outcomes with respect to "truth"; two that are correct
(i.e., are consistent with truth) and two that are
erroneous (Table 3-2). The Type I error rate (i.e., a)
is reported as a standard protocol for reporting the
outcome of statistical tests. Unfortunately, the Type II
error rate (i.e., /) is often overlooked.
Power is defined as 1 f, or the probability of
not committing a Type I error. Estimation of power


Table 3-2. Four outcomes with respect to truth of
the null hypothesis H, and their corresponding
errors (After White et al. 1982).
Decision Based
on Test Statistic

Reject H, Type IError No Error
(a)
Type II Error
Fail to Reject H, No Error Err






requires specifying the Type I error rate (a) and an
alternative hypothesis (HJ to the null hypothesis (Ij ).
We have attempted to report power for tests where we
fail to reject a specified alternative hypothesis. If
otherwise unspecified, all Type I error rates for power
estimation were a = 0.05. In most cases we have used
more than one alternative hypothesis H, and express
these hypotheses in the form of a systematic departure
(A) from the null hypothesis H,, so that we may evaluate
power over a range of hypothesized differences. We
usually express these alternative hypotheses as:

H(0) =A*0

where 0 is a given parameter (e.g., survival or
movement probability), and A is the departure from H,
imposed on that parameter. For example, if 6=0.8
under our null hypothesis and A = 0.95, then 6 under
our alternative hypothesis would be 0.95*0.8 = 0.76
We evaluated the power of a given test by
simulating an effect on the actual data (Lebreton et al.
1992). We generated expected count data under an
alternative model H. (as above) that included the
differences (A) we intended to evaluate. These expected
values were then analyzed as if they were real data
(Drost et al. 1989, Lebreton et al. 1992). The resulting
test statistic asymptotically has a non-central chi-square
distribution and can be used to approximate the non-
centrality parameter for an alternative hypothesis K to
the null hypothesis H of no effect. The probchi function
(SAS Inc. 1988) enables us to then estimate f (i.e., the
Type II error rate) using this non-centrality parameter
for a specified a (in this case a=0.05). Power is then
estimated as 1 8.






LOGISTIC REGRESSION AND LOG-LINEAR
MODELS

For some analyses of the influences of survival
and movement probabilities, we used a conditional
logistic regression or log-linear models based on one
month time intervals (see discussion of time interval
under Estimation of Movement Probabilities). We used
logistic regression in cases where there was a binary
response variable (i.e., survival or movement)(Cox
1970, Cox and Oakes 1984, Hosmer and Lemeshow
1989). Because individual animals can move more than
once, we are assuming conditional independence (e.g.,
that the probability of moving between times t and t + 1
is not dependent on whether that animal moved during
the previous interval).
We used log-linear models for cases where there
was no designated response variable, but rather, for
exploration of interactions in multi-way cross tabulation
(e.g., the probabilities of movements between specific
locations)(Agresti 1990, Everitt 1992). We used MLE
for the parameter estimation from both logistic and log-
linear models (SAS inc. 1988) and consequently, could
apply all of the model selection tools described above.

THE ANALYSIS OF RESIDUALS FROM CROSS-
CLASSIFICATION MODELS

In some cases in our analysis of cross-classified
models (e.g., contingency tables or log linear analyses)
we make use of the residuals for exploring the
contributions of individual cells to the overall test
significance. We do not use this approach in a
hypothesis testing or model selection capacity as
described for LRTs or AIC; but rather, to further
describe patterns already detected in our analysis. This
is often done informally by inspection of deviations from
expected values. A problem, however, with this
approach is that a fixed deviation does not have the same
importance across all sample sizes (Everitt 1992).
Consequently we used standardized residuals (Agresti
1990, Everitt 1992) for evaluating the contribution of
individual cells to the overall model significance from
models in which the expected values were derived from
an a priori hypothesis. Where expected values were
derived from the marginal totals, we evaluated cell
contributions using adjusted residuals (Haberman 1973).
Adjusted residuals are normally distributed with a mean
of zero and a standard deviation of one such that a
residual of 1.96 corresponds with an a of 0.05.


MULTIPLE COMPARISONS

Throughout this report we have conducted
numerous statistical analyses, often involving multiple
comparisons of the same parameter. Multiple
comparisons can result in an inflated alpha (a) level (Day
and Quinn 1989, Fowler 1990); however, the complexity
of many of our analyses would enable several approaches
to correct for this inflation. Consequently, unless
otherwise stated, we did not attempt to adjust the a level.
Thus, the reader is cautioned to consider the number of
multiple comparisons when interpreting our results.



















































('Impur -4. IR 1I IX)IN I/


LomtI iitr'






.. ... .. .. I .






. .. -h... .. h



r, i I I I I. ,


I ,,, I, Ih ,I I I ,,. 1 I h .... .. h






I I ,, I. I I .. II ,,
, ., ,, 1 I I ... ... 1 I ,. 1 I, I


L_





































Mean Life Span (MLS)

Beissinger (1986) suggested that the average
adult life span of Snail Kites in the wild is 5-8 years.
This suggestion was based on presumed mortality
associated with droughts, based on differences in
consecutive counts from the annual Snail Kite survey.
Our results, based on Cormack's (1964)
estimator, suggests that, given a bird has survived 100
days post-fledging, mean life span (MLS) for adults
ranges from 6.0 to 10.3 years (depending on the
assumptions made about drought intervals and survival
during droughts), and the expected life time E(T), based
on the estimator of Brownie et al. (1985), for a newly-
fledged juvenile ranges from 4.2 to 6.7 years (Table 4-
1). Improvements on these estimates should be possible
as data on drought-year and lag-year survival become
available (a monitoring program is in place to obtain
these data, provided that funding is available to maintain
this effort). These estimates also do not take into
account the potential for senescence (i.e., declines in
survival rates as age increases).


Estimation of Survival from Radio
Telemetry

We attached a total of 282 radio transmitters,
representing 271 individual Snail Kites with 11 birds
having been recaptured in a subsequent year and their
radios replaced. We were slightly short (82%) of our
targeted sample size of 100 birds during 1992 (Table 4-
2); but fully attained our targeted sample sizes in 1993
(Table 4-3) and 1994 (Table 44). We were very close to
our targets of age and sex ratios of our sample. The
overall age distribution of our sample was 59% (165 of
282) adults and 39% (117 of 282) juveniles; our target
was 60% adults and 40% juveniles. We had a total of 82
males (49.7%) and 83 females (50.3%); our target was
50% of each. We had a few minor discrepancies
between the distribution of our samples each year and
the preceding annual counts; however, most of these
resulted from shifts in the distribution of birds, rather
than an inability to sample from particular areas.


Table 4-1. Mean life span (MLS)for adult Snail Kites and expected life time E(T)forjuvenile
Snail Kites; given estimated parameter values for adult (ad) and juvenile (juv) survival (0)
during high water and hypothesized parameter values for the drought interval (years), Pad
(drought), and 4a (lag year).
Hypothesized Hypothesized Hypothesized
Drought -d ~d ad 4 MLSad E(T)j,
Interval (High Water) (Drought) (Lag Year)
5 0.91 0.6 0.9 0.58 6.0 4.2
5 0.91 0.7 0.9 0.58 7.0 4.8
5 0.91 0.8 0.9 0.58 8.3 5.6
5 0.91 0.9 0.9 0.58 10.1 6.6
6 0.91 0.6 0.9 0.58 6.5 4.5
6 0.91 0.7 0.9 0.58 7.4 5.1
6 0.91 0.8 0.9 0.58 8.6 5.7
6 0.91 0.9 0.9 0.58 10.2 6.7
7 0.91 0.6 0.9 0.58 6.9 4.7
7 0.91 0.7 0.9 0.58 7.7 5.3
7 0.91 0.8 0.9 0.58 8.8 5.9
7 0.91 0.9 0.9 0.58 10.3 6.7










Table 4-2. The number of snail kites captured during 1992 and equipped with radio transmitters at each location
of each age class and sex.
Total 1992 % of % of 1991
Location Adult Juvenile Male' Female' Sample Sample Count'
WCA-2A 2 7 0 2 9 12 4
WCA-2B 1 0 1 0 1 1 3
WCA-3A 2 0 1 1 2 2 1
Holey Land W.M.A. 1 1 1 0 2 2 0
Loxahatchee Slough 2 2 1 1 4 5 7
Lake Okeechobee 19 14 12 8 34 41 40
Lake Kissimmee 7 0 5 2 7 9 13
Lake Tohopekaliga 3 7 1 2 10 12 1
St Johns Marsh 7 6 2 5 13 16 22
Total 45 37 24 21 82 100 913
'Applies to adults only, juveniles cannot readily be sexed in the field
2 Percentage of kites in each area based on the 1992 annual survey
'Some peripheral areas included in the annual count were not sampled due to absence of birds at time of sampling.




Table 4-3. The number of snail kites captured during 1993 and equipped with radio transmitters at each location of
each age class and sex.

Total 1993 % of % of 1992
Location Adult Juvenile Male' Female' Sample Sample Count2

Everglades N.P.' 0 2 0 0 2 2 9
WCA-2A 5 2 3 2 7 7 6
WCA-2B 1 1 0 1 2 2 0
WCA-3A 6 9 4 2 15 15 15
WCA-3B 2 2 1 1 4 4 2
Holey Land W.M.A. 2 1 0 2 3 3 2
Loxahatchee Slough 7 2 1 6 9 9 2
Lake Okeechobee 20 11 10 10 31 31 29
Lake Kissimmee 10 3 7 3 13 13 5
Lake Tohopekaliga 2 0 0 2 2 2 3
E. Lake Tohopekaliga 1 2 0 1 3 3 1
St. Johns Marsh 4 5 4 0 9 9 11
Total 60 40 30 30 100 100 854
'Applies to adults only, juveniles cannot readily be sexed in the field
SPercentage of kites in each area based on the 1992 annual survey
'Including the North East Shark River Slough addition lands.
'Some peripheral areas included in the annual count were not sampled due to absence of birds at time of sampling.













































EFFECTS OF AGE ON SURVIVAL

Our estimates of survival for adults were
generally higher than for juveniles in each year of this
study (Fig. 4-1)(Appendix 4-1). Based on log-rank
statistics, adult and juvenile survival estimates differed
for study years (SY) 1992 and 1994, but not in SY 1993
(Table 4-5). In both years where they differed, estimates
of adult survival were higher than estimates of juvenile
survival (Table 4-6). However, in 1993 estimates of
juvenile survival were slightly (but not significantly, P =
0.869) higher than adult survival. Ancillary evidence
suggests that our estimates for juvenile survival during
1992 and 1993 were biased high (see discussion of
Assumptions, Bias, and Sources of Error). Our data
from banding (i.e., mark-resighting) also indicates an
age effect on survival (below). Consequently, we
believe that real differences exist between adult and
juvenile survival- and that differences were
underestimated for 1992 and 1993.


Table 4-4. The number of snail kites captured during 1994 and equipped with radio transmitters at each
location of each age class and sex.

Total 1994 % of % of 1993
Location Adult Juvenile Male' Female' Sample Sample Count2

Everglades N.P.3 0 1 0 0 1 1 3
Loxahatchee N.W.R. 0 1 0 0 1 1 0
WCA-2A 0 0 0 0 0 0 0
WCA-2B 18 10 8 10 28 28 4
WCA-3A 11 11 4 7 22 22 41
WCA-3B 2 0 1 1 2 2 12
Lake Jackson 1 1 0 1 2 2 0
Loxahatchee Slough 0 0 0 0 0 0 4
Lake Okeechobee 10 4 5 5 14 14 13
Lake Kissimmee 9 6 5 4 15 15 5
Lake Tohopekaliga 4 1 1 3 5 5 0
E. Lake Tohopekaliga 2 4 2 0 6 6 3
St. Johns Marsh 3 1 2 1 4 4 1
Total 60 40 28 32 100 100 86'
'Applies to adults only, juveniles cannot readily be sexed in the field
2Percentage of kites in each area based on the 1992 annual survey
'Including the North East Shark River Slough addition lands.
4 Some peripheral areas included in the annual count were not sampled due to absence of birds at time of sampling.


Table 4-5. Results of log-rank tests
between survivorship functions of adult
and juvenile Snail Kites during each
study year (April 15 April 14).

Year X2 df Prob

1992 4.611 1 0.032
1993 0.027 1 0.869
1994 29.520 1 <0.001










1992 1993
1.0 ---1.0
0.8 ......................... ...... ....... 0.8
a0.6 0.6
0.4 30.4
0.2 l 0.2 I ene
0.0 0.0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Day Since 15 April Day Since 15 April


1.0-------
0.8
I0.6


0.2
0.0
0 50 100 150 200 250 300 350
Day Since 15 April


Figure 4-1. Estimates of survivorship functions of radio-transmittered adult and juvenile Snail Kites during study
year (SY) 1992, 1993, and 1994. Estimates were derived using a Kaplan-Meier estimator. Confidence intervals
for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 4-1.


EFFECTS OF SEX ON SURVIVAL

In contrast to age, survivorship functions did not differ
between adult male and female kites at a = 0.05 (Table
4-7); although there was a slight divergence (significant
at a = 0.1) in these functions during SY 1994 (Fig.
4-2)(Appendix 4-2). We were unable to determine the
sex of juveniles in the field and consequently have no
information regarding survival differences between sexes
for this age class.

TEMPORAL EFFECTS OF SURVIVAL

Annual Effects- Our estimates of survivorship
functions of adult Snail Kites did not differ among years
at a = 0.05; although, SY 1992 differed from SY 1993
at a = 0.10 (Fig. 4-3)(Table 4-8). In contrast, our
estimates of survivorship of juveniles during SY 1994
differed from both SY 1992 and SY 1993 (Table 4-9).


Table 4-6. Annual survival estimates () for adult
and juvenile Snail Kitesfor 1992, 1993, and 1994.
Survival is estimated from 15 April to 14 April (e.g.,
Q for 1992 is estimated from 15 April, 1992 to 14
April, 1993). Estimates are derived from Kaplan
Meier estimator.

Age Year 4 SE(<) 95% CI(~
Class

Adult 1992 0.962 0.038 0.888 1.000
Adult 1993 0.858 0.063 0.734 0.982
Adult 1994 0.883 0.042 0.801 0.965
Juvenile 1992 0.825 0.080 0.668 -0.981
Juvenile 1993 0.867 0.088 0.695 1.000
Juvenile 1994 0.439 0.090 0.263 0.615


1992 1993 1994
1.0 1.0.... --- 1.0.......... ......
0.8 0.8 0.8 ...........
S0.6 0.6 0.6
S04 0.4 0.4
tn -F (0 --- Fema: e c -- F-emale
0.2 ae 0.2 Male 0.2 e
0.0 0.0 0.0
0 50 100 150 200 250 300 350 0 50 0 1 150 200 250 300 350 0 50 100 150 200 250 300 350
Day Since 15 April Day Since 15 April Day Since 15 April

Figure 4-2. Estimates ofsurvivorship functions ofradio-transmittered adult male and female Snail Kites during
study years (SY) 1992, 1993, and 1994. Estimates were derived using a Kaplan-Meier estimator. Confidence
intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 4-2.










Table 4-7. Results of log-rank tests between
survivorshipfunctions of adult female and
male Snail Kites during each study year
(April 15 April 14).

Year X df Prob

1992 1.667 1 0.280
1993 0.243 1 0.622
1994 2.753 1 0.097'

SUsing the alternative tests described by Cox and
Oakes (1984)(Appendix 3-3) that are slightly less
conservative (i.e., have greater power, but
higher risk of Type I error) we estimated
Z=2.77, P=0.095 and f=2.75, P=0.097for
alternative tests I and 2, respectively.




1.0"
1.0- -.--- ..... .. '- ---..

0.8-

S0.6-

O8 0.4
119943
0.2-


0 50 100 150 200 250 300 350
Day Since 15 April

Figure 4-3. Kaplan-Meier estimates ofsurvivorship
functions of radio-transmittered adult Snail Kites
during each study year (SY). Confidence intervals
for estimates are not shown to minimize cluttering,
but are provided in detail in Appendix 4-1.



Table 4-8. Results of log-rank tests between
survivorshipfunctions of adult Snail Kites during
each study year (SY)(April 15-April 14).

Comparison X2 df Prob

1992 vs. 1993 2.836 1 0.092'
1992 vs. 1994 1.762 1 0.184
1993 vs. 1994 0.480 1 0.486
SUsing the alternative tests described by Cox and
Oakes (1984) (Appendix 3-3) that are slightly less
conservative (i.e., have greater power, but higher risk
of Type I error) we estimated x2=2.93, P=0.087for
both tests.


Survival estimates for juveniles during SYs 1992 and
1993 were very similar, but were markedly higher than
the estimate for 1994 (Fig. 4-4) (but see discussion of
censoring below in section on Assumptions, Bias, and
Sources of Error).


Table 4-9. Results of log-rank tests between
survivorship functions ofjuvenile Snail Kites
during each study year (SY)(April 15-April 14).

Comparison X2 df Prob

1992 vs. 1993 1.432 1 0.231


1992 vs. 1994 6.156
1993 vs. 1994 12.412


1 0.013
1 <0.001


1.0

0.8-

S0.6-

o 0.4- -1

0.2

0.0 ..... .. ... ... ........ ....... ....... .,
0 50 100 150 200 250 300 350
Day Since 15 April

Figure 4-4. Kaplan-Meier estimates ofsurvivorshp
functions of radio-transmittered juvenile Snail Kites
during each study year (SY). Confidence intervals
for estimates are not shown to minimize cluttering,
but are provided in detail in Appendix 4-1.



Seasonal Effects-- It was apparent from
survivorship functions that the risk of mortality (i.e.,
hazard function) was not constant over time. Compared
to an expected value for the number of deaths being
equal in each month, mortality of adults tended to be
highest during the winter months (x= 18.00, df=ll,
P=0.08), and juveniles during late spring and summer
(X2=39.54, df= 11, P<0.001)(Fig. 4-5). The seasonal
pattern of juveniles corresponds with the first few
months post fledging (Fig. 4-6). Juveniles are becoming
independent of their parents, beginning to forage on their
own, and disperse into unfamiliar areas. Juveniles that
survived the first few months post fledging appeared to
be most vulnerable at the same time as peak mortality for
adults (i.e., January). It is less clear why adult









10
Adults
S8-


0
06


E
4 -

S

APR MAY JUN JUL AUG SEP OCT NOV DEC JAN FEB MAR


APR MAY JUN JUL AUG SEP OCT NOV DEC JAN FEB MAR
Month


Figure 4-5. The number of radio-transmittered adult
and juvenile Snail Kites found dead during each month.



10
Adults
8-

06-

4-

2 l

0 30 90 120 150180 210240 270 300 330 30 420 450 480


mortality was highest in winter; however, during this
time cold temperatures tend to lower the availability of
existing food resources (Cary 1983, 1985), and leaves
are absent from willows, which is the most commonly
used species for communal roosting (Sykes et al. 1995,
Darby et al. 1996a). We offer this latter suggestion
because predation was the most common cause of death
of adults. Ancillary evidence suggests that Great-horned
Owls (Bubo virginianus), which forage at night, were the
most common predator. We emphasize, however, that
our suggestions for why adult mortality tended to occur
during winter is purely speculative. Other anecdotal
evidence (i.e., approximately 6 to 10 birds without
radios found dead at nests) suggested that adults can also
be vulnerable to owl predation during nesting while
incubating at night.

SPATIAL EFFECTS OF SURVIVAL

Regional Effects (By Region of Initial
Capture)- We approached regional differences in
survival two ways. First we tested the hypothesis that
differences in survival were attributable to differences in
the region of initial capture. For this analysis, a bird
was assigned to the region of its initial capture,
regardless of whether it moved subsequent to capture.
For juveniles, the region of capture represents their natal
region; however, in most cases we do not know the natal
origin of adults or their history of locations prior to
capture. Consequently, this hypothesis is biologically
more intuitive for juveniles than adults. Differences in
survival could reflect several aspects of regional quality
(e.g., food abundance).
Overall there were few differences in survival
among regions of capture. Survival of adult birds
captured in different regions did not differ among any
regions during 1992 or 1993 (Table 4-10)(Fig. 4-7).
During 1994, adult survival differed between the
Everglades and Okeechobee regions and between the
Everglades and Kissimmee regions but not among any
other pairwise combinations. Juvenile survival differed
between the Everglades and Okeechobee regions during
1992, but no other differences were detected in any year
(Table 4-1 )(Fig. 4-8).
We urge caution in interpreting these results.
Partitioning survival into two age classes and five
different regions (no animals were captured in the
peripheral region) often resulted in the number of
animals at risk for a given group (r5) being small.
Although this does not impose bias, it can present
misleading appearances (see discussion of small sample
size in Assumptions, Bias, and Sources ofError).


0 30 60 90 120150180210 240270320 0003 604 204 50 4
Day Since Fledging


Figure 4-6. The number of radio-transmittered adult
and juvenile Snail Kites found dead during each 30-day
interval after fledging (uveniles) or capture (adults).









10
Adults
0 a
6-

AT

Z2


APR MAY JUN JUL AUG SEP OCT NOV DEC JAN FE MAR


Month


Figure 4-5. The number of radio-transmittered adult
and juvenile Snail Kites found dead during each month.




10
Adults
8-



04

2
Z 2

0 30 60 9 0 1 150 180210240270 300 0360390420450 48











10 3 111 240 27-00 330 300 3003420-450 40



Figure 4-6. The number of radio-transmittered adult
and juvenile Snail Kites found dead during each 30-day
interval after fledging uveniles) or capture (ults).
j28-








0 S0 0 120W15 I 180210240 270 303W0360 340450440
Day Since Fledging


Figure 4-6. he number of radio-transmittered adult
andjuvenile Snail Kites found dead during each 30-day
interval afterfledging (juveniles) or capture (adults).


mortality was highest in winter; however, during this
time cold temperatures tend to lower the availability of
existing food resources (Cary 1983, 1985), and leaves
are absent from willows, which is the most commonly
used species for communal roosting (Sykes et al. 1995,
Darby et al. 1996a). We offer this latter suggestion
because predation was the most common cause of death
of adults. Ancillary evidence suggests that Great-horned
Owls (Bubo virginianus), which forage at night, were the
most common predator. We emphasize, however, that
our suggestions for why adult mortality tended to occur
during winter is purely speculative. Other anecdotal
evidence (i.e., approximately 6 to 10 birds without
radios found dead at nests) suggested that adults can also
be vulnerable to owl predation during nesting while
incubating at night.

SPATIAL EFFECTS OF SURVIVAL

Regional Effects (By Region of Initial
Capture)- We approached regional differences in
survival two ways. First we tested the hypothesis that
differences in survival were attributable to differences in
the region of initial capture. For this analysis, a bird
was assigned to the region of its initial capture,
regardless of whether it moved subsequent to capture.
For juveniles, the region of capture represents their natal
region; however, in most cases we do not know the natal
origin of adults or their history of locations prior to
capture. Consequently, this hypothesis is biologically
more intuitive for juveniles than adults. Differences in
survival could reflect several aspects of regional quality
(e.g., food abundance).
Overall there were few differences in survival
among regions of capture. Survival of adult birds
captured in different regions did not differ among any
regions during 1992 or 1993 (Table 4-10)(Fig. 4-7).
During 1994, adult survival differed between the
Everglades and Okeechobee regions and between the
Everglades and Kissimmee regions but not among any
other pairwise combinations. Juvenile survival differed
between the Everglades and Okeechobee regions during
1992, but no other differences were detected in any year
(Table 4-11)(Fig. 4-8).
We urge caution in interpreting these results.
Partitioning survival into two age classes and five
different regions (no animals were captured in the
peripheral region) often resulted in the number of
animals at risk for a given group (r,) being small.
Although this does not impose bias, it can present
misleading appearances (see discussion of small sample
size in Assumptions, Bias, and Sources of Error).











































Regional Effects (By Region of Current
Location)- The second approach we used for regional
differences in survival reflected actual time spent in each
region, rather than just focusing on the region of capture.
Thus, we test the hypothesis that the survival of a bird is
affected by its current location (e.g., predation risk).
For this analysis, a bird that moved from a given region
to another was censored from the number of animals at
risk for that region (rj) at the midpoint of the interval,
and added to the number of animals at risk in the region
to which it moved. All deaths were assigned to the


1.0
0.8

> -Evr.gld.
|04 --.-.:S S
-- Upper StJohn
02

0.2. =-JPoP m
0.0
0 50 100 150 200 250 300 350
Date


region where the dead bird was actually found.
Similar to region of initial capture, there was
little indication of regional differences in adult survival.
We found no differences during 1992 or 1993 although,
as above, the number of birds at risk (rj) was often low
(Fig. 4-9)(Table 4-12). Survival did differ between the
Everglades and Okeechobee regions during 1994 at a =
0.05, and between the Everglades and both the Upper St.
Johns and Peripheral regions at a = 0.10. We found no
significant regional differences for juveniles among any
regions during any year (Fig. 4-10)(Table 4-13).


1993 1994
1.0 ---------- 1.0


0.4a D0.4
o.s ........... o.
^' -- -4 *..
22 0. 2 =.*fl0.

0.0 0.0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Date Date


Figure 4-9. Estimates of survivorship functions of radio-transmittered adult Snail Kites while they were present in each
region during each study year (SY). Estimates were derived using a Kaplan-Meier estimator. Confidence intervals for
estimates are not shown to minimize cluttering, but can be easily estimated from information in Appendix 4-5.


Table 4-11. Results ofpairwise log-rank tests between survivorship functions of radio-transmittered juvenile Snail
Kites fledged from each region during each study year (SY).

SY1992 SY1993 SY 1994

Comparisoni X2 df Prob 72 df Prob X2 df Prob
EVER vs. OKEE 4.582 1 0.032 0.300 1 0.584 0.095 1 0.758
EVER vs. KISS 1.138 1 0.286 2 1.121 1 0.290
EVER vs. USJ 0.873 1 0.350 2 _2 --
EVER vs. LOXSL 3 _3 .3 _
OKEE vs. KISS 1.333 1 0.248 0.655 1 0.418
OKEE vs. USJ 1.750 1 0.186 2 -3 .3
OKEE vs. LOXSL 3 3 3
KISS vs. USJ 0.050 1 0.824 2 3
KISS vs. LOXSL -- ---3
USJ vs. LOXSL --3 3
' Regions ofcomparison are Everglades (EVER), Okeechobee (OKEE), Kissimmee (KISS), Upper St. Johns (USJ), and
Loxahatchee Slough (LOXSL).

SNo deaths occurred of birds from this region during this year; however some sample sizes were quite small (Appendix
4-4).

s There were insufficient data to estimate survival and to conduct corresponding log-rank test.










Table 4-12. Results ofpairwise log-rank tests between survivorship functions of radio-transmittered adult Snail Kites
while they were present in each region during each study year (SY).

SY1992 SY1993 SY 1994

Comparison' Y2 df Prob k2 df Prob 92 df Prob
EVER vs. OKEE 0.545 1 0.460 -2 4.061 1 0.044'
EVER vs. KISS 0.377 1 0.539 1.358 1 0.244
EVER vs. USJ 2 -. 2.842 1 0.092'
EVER vs. LOXSL 4 4 _
EVER vs. PERI 2.768 1 0.096'
OKEE vs. KISS 0.364 1 0.546 0.1787 1 0.673 0.055 1 0.814
OKEE vs. USJ 0.136 1 0.712 0.035 1 0.851
OKEE vs. LOXSL _4 -.
OKEE vs. PERI 0.045 1 0.831 0.099 1 0.753
KISS vs. USJ 2 -4 0.042 1 0.838
KISS vs. LOXSL 4 -- -
KISS vs. PERI- -4 -2 _
USJ vs. LOXSL 4 -. 4 -
UJS vs. PERI 2 0.020 1 0.887
LOXSL vs. PERI __4 _4 __4

SRegions of comparison are Everglades (EVER), Okeechobee (OKEE), Kissimmee (KISS), Upper St. Johns (USJ),
Loxahatchee Slough (LOXSL), and Peripheral Wetlands periI).

SNo deaths occurred while in this region during this year; however some sample sizes were quite small (Appendix 4-5).

SUsing the alternative tests described by Cax and Oakes (1984)(Appendix 3-3) that are slightly less conservative (i.e., have
greater power, but higher risk of Type I error) we estimated f =4.210, P=0.040 and eJ=4.149, P=0.042for alternative
tests 1 and 2, respectively.

There were insufficient data to estimate survival for one or both groups and to conduct corresponding log-rank test.

s Using the alternative tests described by Cox and Oakes (1984)(Appendix 3-3) that are slightly less conservative (i.e., have
greater power, but higher risk of Type I error) we estimated j=2.847, P=0.092for both alternative tests I and 2.

6 Using the altenative tests described by Cox and Oakes (1984)(Appendix 3-3) that are slightly less conservative (i.e., have
greater power, but higher risk of Type I error) we estimated =2.829, P=0.093 for both alternative tests 1 and 2.



1992 1993 1994
1.0 --------------------- 1.0 ---------------- 1.0
0.8 0.8 0.8
-- ---------- g .,
0.6 > .B. 0.6 0.6
-i 0 --i0.4 |
0.2 UppSWtJoSto 02 P- 0.2 ier
0.0 -- 0.0 0.0
0 50 100 150 250 300 350 0 50 100 150 200 250 300 350 50 100 150 200 250 300 350
Date Date Date

Figure 4-10. Estimates of survivorshipfunctions of radio-transmittered juvenile Snail Kites while they were present in each
region during each study year (SY). Estimates were derived using a Kaplan-Meier estimator. Confidence intervals for
estimates are not shown to minimize cluttering, but can be easily estimated from information in Appendix 4-6.














































Habitat Effects-Because most of our radio
locations were obtained from aircraft, we were
sometimes unable to record the habitat type of a given
animal's location. This, in combination with the
intermittent use of some habitat types (e.g., cypress and
disturbed habitats) precluded a meaningful analysis using
a Kaplan-Meier estimator (i.e., the number of animals at
risk [rJ for these habitats was often too small, or zero).
However, we did conduct a more cursory examination
by testing whether the number of deaths that occurred in
each habitat was proportional to the overall use of that
habitat. This test gave us no indication of disparity
between mortality.and habitat type (x2=4.68, 4 df,
P=0.68)(Fig. 4-11).
We were unable to detect any difference in
survival of adults between lake and marsh habitats for
any year (Table 4-14)(Fig. 4-12). In contrast, juvenile
survival differed between these habitats during 1992
(Table 4-15), but not during 1993 or 1994 (Fig. 4-13).
Although differences were not statistically significant
when regions were compared, this is consistent with the


survival estimates from different regions (i.e., estimates
from the Everglades and Upper St. Johns regions [marsh
habitats] were lower than from Okeechobee or
Kissimmee regions [lakes]. This may, at least partially,
reflect the conditions in the Everglades following an
extended dry period from 1989 to 1991.


% Habitat Use % Deaths
36.7% 30
35,0%
15.205


.Sa ia -150%

SOmiao M-n 0 Olkhob. 0 N~m ,ort5,m e LW Cypn Pr*l


Figure 4-11. Percentage of total use by adult Snail Kites
of each habitat type and percentage of the total number of
birds that died in each habitat type.


Table 4-13. Results ofpairwise log-rank tests between survivorship functions of radio-transmittered juvenile Snail
Kites while they were present in each region during each study year (SY).
SY 1992 SY 1993 SY 1994

Comparison' X2 df Prob Y2 df Prob Z2 df Prob
EVER vs. OKEE 1.750 1 0.186 0.400 1 0.527 2
EVER vs. KISS 1.000 1 0.317 1.452 1 0.228
EVER vs. USJ 0.002 1 0.968 -2 -2 --
EVERvs.LOXSL --2 2 _
EVER vs. PERI 2 0.200 1 0.655 -2
OKEE vs. KISS -j 2 _
OKEE vs. USJ 2.333 1 0.127 2
OKEE vs. LOXSL -2 -- 2
OKEE vs. PERI 2 2
KISS vs. USJ 1.333 1 0.2482 2
KISS vs. LOXSL -2 --2 2_ .2
KISS vs. PERI 2 _
USJ vs. LOXSL _2 2 _2
UJS vs. PERI _2 2 2
LOXSL vs. PERI _2 -2 .2 -
'Regions of comparison are Everglades (EVER), Okeechobee (OKEE), Kissimmee (KISS), Upper St. Johns (USJ),
Loxahatchee Slough (LOXSL), and Peripheral Wetlands periI).

2 There were insufficient data to estimate survival for one or both groups and to conduct corresponding log-rank test.
SNo deaths occurred while in one or both of these regions during this year; however some sample sizes were quite small
(Appendix 4-6).










1992 1993 1994
1.0 1.0 771--\ 1.0
0.8 0.8 ........ 0.8
00.6 0.6 0.6
S0.4 0.4 04
)) U)
0.2 0.2 0.2 i Ii
0.0 0.0 0.0
00 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Date Date Date

Figure 4-12. Estimates ofsurvivorshipfunctions of radio-transmittered adult Snail Kites while they were present in lake
and marsh habitats during each study year (SY). Estimates were derived using a Kaplan-Meier estimator. Confidence
intervalsfor estimates are not shown to minimize cluttering, but are provided in detail in Appendix 4-7.


Table 4-14. Results of log-rank tests between
survivorship functions of adult Snail Kites while
they were in lake and marsh habitats during each
study year (SY).

Year 2 df Prob

1992 0.633 1 0.426
1993 1.367 1 0.242
1994' 2.028 1 0.154
SUsing the altenative tests described by Cox and
Oakes (1984)(Appendix 3-3) that are slightly less
conservative (i.e., have greater power, but higher
risk of Type I error) we estimated X'=2.075,
P=0.150 and e=2.046, P=O.153 for alternative
tests 1 and 2, respectively.


Table 4-15. Results of log-rank tests between
survivorship functions ofjuvenile Snail Kites while
they were in lake and marsh habitats during each
study year (SY).

Year X2 df Prob

1992' 4.353 1 0.037

1993 0.903 1 0.342

1994 0.559 1 0.455

SUsing the alternative tests described by Cox and Oakes
(1984)(Appendix 3-3) that are slightly less conservative
(i.e., have greater power, but higher risk of Type I
error) we estimated J=4.383, P=0.036for alternative
tests I and 2, respectively.


1992 1993 1994
0 ------- ------------- 1.0 ------------- 1.0
0.8 0. 0.8
0 6 0.6 -0.6-
0.4 =0.4 0.4
0.2 L02 2-------
0.0 0.0 l .l l i ii l 11 111. .1..1. 0.0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Date Date Date
Figure 4-13. Estimates of survivorship functions of radio-transmittered juvenile Snail Kites while they were present in lake
and marsh habitats during each study year (SY). Estimates were derived using a Kaplan-Meier estimator. Confidence
intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 4-8.










MODEL SELECTION FOR EFFECTS ON SURVIVAL

Above we provided evidence for individual
effects on survival based on relatively continuous time
(data were discrete at the interval of telemetry locations)
using a Kaplan-Meier estimator. While this approach
has many advantages (e.g., no assumptions about the
hazard function), it is not a convenient method for
exploration of more complex models including the
relative effects of multiple factors. Here we use a
discrete-time approach (conditional logistic regression)
to evaluate the probability that a bird survives to time t
+ 1, conditional on it being alive at time t. This
approach enables us to better evaluate effects on survival
in combination to determine effects most appropriately
supported by our data in an overall survival model.
We began our analysis with a univariate
examination of the sources of variation (Table 4-16).
Because the potential for interaction effects to be masked
at this preliminary stage, we followed the
recommendation of Hosmer and Lemeshow (1989) and
used 0.25 as a rejection criteria (a) for inclusion in
further models. The effects due to region were the only
source eliminated at this phase of our analysis. We also
found little evidence for regional effects on survival
using a Kaplan-Meier estimator. We were unable to
estimate the effects of month because no deaths occurred
during some months. Consequently, we retained month


Table 4-16. Summary statisticsfor the univariate
analyses (using conditional logistic regression) of
effects on the conditional probability of an animal
surviving to time t + 1, given that it was alive at
time t. Z' is based on a Wald statistic (SAS Inc.
1988).

Source df X2 P

Age 1 18.29 <0.001
Sex 2 18.32 <0.001
Season 2 5.17 0.075
Month' -- --
Year 2 8.59 <0.001
Region 5 3.06 0.691
'Because deaths did not occur in some months,
estimation of these statistics were not reliable. We
evaluated this source in a more comprehensive
analysis (below) using model selection procedures.


as a potential source of variation at this stage of the
analysis.
Some of the parameters at this stage of the
analysis were subsets of other parameters (e.g., season
represents combinations of months). Next we evaluated
which of these potentially redundant parameters to use in
further models. For this analysis, we used 3 classes of
sex (male, female, and unknown). Because we were
unable to determine the sex of juveniles in the field, sex
represents a combination of age and sex (i.e., all cases
where the sex class is unknown are juveniles). Based on
AIC (Table 4-17) sex provided no substantial gain in
model fit at the cost of an additional parameter.
Consequently, we used age, but not sex, in all
subsequent models. Similarly, individual months
provided no substantial gain in model fit over season at
the cost of 9 additional parameters. Thus, we used
season, but not month, in all subsequent models.
The remainder of our model selection invloved
the comparison of models using different combinations
of age, season, and year effects. Based on AIC, any of
the models which did not include some combination of
all three of these effects generally were less adequate
than models which included all three effects. Although
model selection often indicates that more than one model
is suitable (McCullagh and Nelder 1989) for a given data
set, our analysis indicated (by a substantially lower AIC)
that a model including the main effects of age, season,
year, and the interaction effects of age*year, and
age*season was the most appropriate model for our data.
This conclusion is further supported by the goodness-of-
fit (P=0.806) of this model compared to the fully
saturated model.

EFFECTS OF HYDROLOGY

One objective of this study was to determine the
environmental correlates of survival, particularly
hydrology. However, as previously suggested (Bennetts
and Kitchens 1993, 1994), survival was sufficiently high
that within-year patterns would have been difficult to
detect without enormous effort (i.e., too few individuals
died to enable quantitative comparisons with those that
survived) and probably would not have been very
insightful. Rather, we now believe that between-year
differences in survival will be more appropriate. Even
though we encountered relatively high water conditions
throughout this study, the mark-resighting phase of our
study (below) was intended to detect these differences
and will enable a long-term evaluation of environmental
correlates (e.g., hydrologic conditions) as variable
conditions occur.










Table 4-17. Summary statistics for conditional logistic regression modelfor the factors effecting the probability
ofsurvival to time t + 1 (at monthly time steps), given that an animal was alive at time t. Shown are the model
description, number of estamable parameters (np), relative deviance (-21n[]), Akiake's Information Criteria
(AIC), and a measure ofgoodness-of-fit (GOF)for models with relatively low AIC values. GOF was derived
using the probability of a LRT comparing a given model with afully saturated model. The null hypotheses (HI) of
the GOF statistic is that the reduced model (with fewer parameters) fits tha data as well as the more general
model (with more parameters). Thus, failure to reject H, indicates a fit of the reduced model. The model with
the lowest AIC.(bold) would be selected if based only on this criteria.

Model' np -21n(f) AIC GOF

Age 2 369.333 373.333
Sex 3 368.243 374.243
Seas 3 382.272 388.272
Mon 12 366.496 390.496
Yr 3 379.712 385.712
Reg 6 385.931 397.931
Age Seas 4 363.063 371.063
Age Seas Age*Seas 6 346.849 358.802
Age Yr 4 359.446 367.446
Age Yr Age*Yr 6 349.633 361.633
Seas Yr 5 372.513 382.513
Seas Yr Seas*Yr 9 363.101 381.101
Age Seas Yr Age*Seas Age*Yr 10 324.142 344.142 0.806
Age Seas Yr Age*Seas Seas*Yr 12 327.212 351.212 0.240
Age Seas Yr Age*Yr Seas*Yr 12 333.876 357.876 0.045
Age Seas Yr Age*Seas Age*Yr Seas*Yr 14 322.120 350.120 0.665
Age Seas Yr Age*Seas Age*Yr Seas*Yr Age*Seas*Yr 18 318.039 354.039 1.000
' Term abbreviations: Season (Seas), Month (Mon), Year (Yr), and Region (Reg).


Estimation of Survival from
Banding Data

Our sample of banded birds for survival analyses
was obtained through a cooperative banding effort with
the Florida Game and Fresh Water Fish Commission
(GFC). This sample was also supplemented by
resighting of individually marked birds banded on
previous studies that were observed during this study
(REB, J. Rodgers unpubl. data). Our total sample of
individual banded birds for analyses of survival was 913
(Table 4-18); although birds banded in 1995 will only
contribute to estimates following subsequent resighting
periods (i.e., in future years). Of this sample, 191 were


adults at the time of addition to our sample and 722 were
juveniles; however, resightings of juveniles after their
fledging year then re-enter our sample as adults
(although resightings were not included in our sample
size reported above).

EFFECTS OF AGE AND TIME

Estimates of survival from banding data differ
from those obtained via radio telemetry in that they are
discrete at the interval of sampling periods. In this case
we measured survival from one breeding season to the
next (i.e., approximately annually, see methods). This
precludes the ability for testing some of the effects that
we tested using radio telemetry data. For example, we















































are unable to test seasonal differences in survival without
having had a sampling (i.e., capture and/or resighting)
period in each season of interest.
We began our assessment using the testing
sequence of models described by Pollock et al. (1990) as
Models A, B, and D. Model A assumes that both
survival and resighting probabilities are time dependent
(i.e., that separate estimates for each year are
warranted). Model B assumes that survival is constant
over time (i.e., that a single estimate can be applied to
all years), but that resighting probability is time
dependent. Model D assumes that both survival and
resighting probabilities are constant over time. Each of
these models was generated with and without separate
parameters for each age class.
In this set of models juveniles are assumed to
become adults at the beginning of their first resighting
period after their fledging year (this assumption is tested
below). Juveniles are capable of breeding at nine months


of age (Snyder et al. 1989a) and our radio telemetry
results suggest that, for our data, survival of juveniles
was similar to adults following an initial period of about
3-4 months of high post-fledging mortality. Thus, in
these models, one resighting probability is used for both
juveniles and adults (because juveniles become adults at
the beginning of their first resighting period); but, a
separate estimate of juvenile survival is initially
generated (because juvenile survival is measured from
the time of fledging to their first resighting period), but
may be later deemed unnecessary through model
selection procedures.
The results from this initial sequence (Table 4-
19) indicated that model A with 2 age classes was the
most suitable. The AIC (114.811) from this model was
substantially lower than the alternative models.
Likelihood ratio tests (LRTs) reinforced the selection of
Model A as being the most appropriate with regard to
time dependency. LRTs between models B and A with


Table 4-18. Number of banded birds from each age class, location, and year obtained through a cooperative
banding effort with the Florida Game and Fresh Water Fish Commission or through resighting of birds banded on
previous studies. Individual birds are only shown at the time and location that they entered our sample.
1992 1993 1994 1995'
Location AD JUV AD JUV AD JUV AD JUV
Everglades N.P.2 2 -- 1 -- 8
Big Cypress N.P. -- -- -- 18
Loxahatchee N.W.R. 1 2 2 -
WCA-2A 3 13 6 15 4
WCA-2B 1 1 2 16 45 -- 87
WCA-3A 4 1 8 24 11 23 2 40
WCA-3B 1 -- 1 6 3 -- --
Holey Land W.M.A. 2 2 2 6 --
Loxahatchee Slough 3 3 6 10 1 --
Lake Okeechobee 27 61 20 96 9 6 -- 15
Lake Kissimmee' 8 14 9 31 11 19
Lake Tohopekaliga 4 21 1 7 4 8 1 15
East Lake Tohopekaliga 1 1 1 18 1 14 1 -
Upper St. Johns Marsh 12 43 6 33 3 1 5
Total 66 159 62 252 59 119 4 192

' Birds newly marked in 1995 do not contribute to survival analyses until subsequent resighting periods are completed.

' Includes the Northeast Shark River Slough addition lands

' Includes nearby Lake Jackson




























and without age effects both strongly rejected the more
reduced model (Table 4-20), as did the comparisons
between models A and D with and without age effects.
This indicates that constraining survival (0) and/or
resighting probability (p) to be constant among years was
not justified (i.e., these data support a year effect).
Similarly, a LRT between models A with and without an
age effect (Models A and A.) strongly rejected (x2 =
44.846, 3 df, P < 0.001) suggesting that constraining
parameter estimates to be equal for adults and juveniles
also was not justified (i.e., these data also support an
age effect). The goodness-of-fit (GOF) test also failed to
reject the null hypothesis (Ho) that this model adequately
fits the data (GOF tests for all of the alternative models
strongly rejected Ho.
In order to test the assumption (above) that a
separate estimate for juvenile resighting probability was
not warranted, we generated an alternative model



Table 4-20. Resulting statistics from Likelihood
Ratio Tests (LRT) between Cormack-Jolly-Seber
(CJS) models "A ", "B and 'D"with and
without separate estimates for each age class.

General Reduced > 2
Model Model df P >

A B 10.687 2 0.005
A, B, 19.350 4 <0.001
A D 12.756 4 0.013
A, D,. 21.783 6 0.001


structure in which we assumed that juveniles became
adults at the end (rather than the beginning) of their first
resighting period in the post fledging year. This enabled
separate estimation for juvenile resighting, as well as
survival, probabilities. We then tested the assumption
using an analogous model to Model A, from above by
comparison of models where resighting probability is
constrained and unconstrained to be equal for juveniles
and adults. The AIC of the model in which resighting
probability was constrained to be equal (which is
equivalent to Model A, from above) for the two age
classes was lower than for the unconstrained model
(Table 4-21). A LRT between these models also failed
to reject the more reduced model (x2 = 1.808, 3 df, P
=0.613) indicating that the additional parameters for a
separate resighting probability for each age class were
not warranted for these data.
Based on the results from our radio telemetry,
which indicated greater differences in survival among
years for juveniles than adults, we generated one
additional model to test this hypothesis (i.e., that survival
was time dependent for juveniles, but not for adults).
This model (Model Aj,.) had a lower AIC than model
A,, (Table 4-22). A LRT between these models also
failed to reject the null hypothesis (Ho) that the reduced
model (Model A,,. ) fit the data equally well as the
more general model (Model Ay) (X2 = 0.290, 2 df, P
=0.865). The goodness-of-fit for this model also failed
to reject the null hypothesis (Ho) that this model
adequately fits the data.
Thus, these data support a model that survival
differs among years for juveniles, but not adults.
Parameter estimates for the most parsimonious model
(Model Aju.) are provided in Table 4-23.


Table 4-19. Model selection statistics for initial set of Cormack-Jolly-Seber (CIS) models with and without age
and time dependency. The model with the lowest AIC (bold) would be selected if based only on this criteria.
Model No. Age
Model Description Classes -21n(S) np AIC GOF

A A Pi 1 141.657 6 153.657 <0.001
A., p, 2 96.811 9 114.811 0.167
B 9. p, 1 152.344 4 160.344 <0.001
B,. 0. p, 2 116.161 5 126.161 0.004
D P. p. 1 154.412 2 158.412 <0.001
D,, 0. p. 2 118.595 3 124.595 0.004
























































EFFECTS OF SEX

Most banded birds in this study were banded as
juveniles, which cannot be sexed in the field. Thus, a
large segment of our sample was of unknown sex.
Consequently, we did not attempt any analysis from
banding data on survival differences attributable to sex.

REGIONAL EFFECTS

To test for regional effects of survival and


resighting probabilities, we generated a suite of multi-
state models (Brownie et al. 1993, Nichols et al. 1993)
analogous to the base models described above as models
A, B, and D by Pollock et al. (1990) except that they
include a multi-state component that enables parameters
to be estimated for multiple strata (Brownie et al. 1993,
Nichols et al. 1993)(in this case strata= 5 of the 6
regions of capture no captures occurred in the
peripheral region). Regional effects were tested based
only on the region of capture since banding data do not
provide information regarding where a given bird has


Table 4-21. Model selection statistics for model A4g in which the assumption of equal resighting probabilities
between adults and juveniles is relaxed or constrained. The model with the lowest AIC (bold) would be selected if
based only on this criteria.

Model Resighting probability
Model Description ) e lfor age -21n(E) np AIC GOF
classes

A., p, No 95.004 12 119.004 0.106
A,, A p, Yes 96.811 9 114.811 0.167


Table 4-22. Model selection statistics for variations of models in which survival
is time dependent for both adults and juveniles (Model A,), and is constrained
to be constant for adults, but not juveniles (Model A,,,. The model with the
lowest AIC (bold) would be selected if based only on this criteria.

S Model
Model
Description -21n(S) np AIC GOF

A. p, 96.811 9 114.811 0.167
Aj.. 4,,, pA 97.101 7 111.101 0.253


Table 4-23. Parameter estimates for the selected Cormack-Jolly-Seber (CIS) Model (ModelIA4), in which adults
and juveniles differ in survival, survival is constant among years for adults, and survival is variable among years
for juveniles. Resighting probability in this model differs among years, but is the same for adults and juveniles
(i.e., juveniles are considered adults at the beginning of their first resighting period after their fledging year).

Age Year i 95% C.I.(O) pi 95% C.I.(p)

Adult 1992 0.840 0.637 1.000 0.142 0.073 0.210
Adult 1993 0.840 0.637 1.000 0.255 0.150 0.359
Adult 1994 0.840 0.637 1.000 0.217 0.095 0.336
Juvenile 1992 0.451 0.281 0.620 0.142 0.073 0.210
Juvenile 1993 0.163 0.071 0.256 0.255 0.150 0.359
Juvenile 1994 0.620 0.177 1.000 0.217 0.095 0.336








been between resightings. These models are also the
default models generated by program MSSURVIV
(Hines 1994) used for this analysis. As above, we
generated models with and without age dependency,
enabling us to test hypotheses that survival ( ) and/or
resighting probabilities (p) were affected by age, time,
and region, simultaneously.
For this set of models, the AIC was lowest for
Model B with no age effect; however, Model B. with
an age effect was sufficiently similar (a difference of
1.36- less than our criteria for exclusion) to be retained
as a reasonable alternative (Table 4-24). LRTs
reinforced the selection of Model B as being the most
appropriate with regard to time dependency. LRTs
between models B and A with and without age effects
both failed to reject (Table 4-25), indicating that separate
estimates of survival (0) for each year were not
warranted. Similarly, LRTs between models B and D
were highly significant, indicating that constraining
resighting probability (pi) to be constant among years
also was not justified.
Because AIC was similar for models B and B,,
we used a LRT to specifically test the null hypothesis that
an age effect for Model B was not warranted (i.e., that
the fit of the 1-age model was equal to the fit of 2-age
model). This test (Model B vs. Model B,.) strongly
rejected the null hypothesis (x2 =48.638, 25 df,
P=0.003) indicating that separate parameter estimates
for each age class were warranted (i.e., there was an age
effect). Thus, from this set of models we selected Model


B with 2 age classes (Model B.) as the most appropriate
model for our data.
Using Model B, as a base model, we generated
another set of models in which (1) survival was
constrained to be constant for all regions (Model Ba,_,.),
(2) resighting probability was constrained to be constant
for all regions (Model B ,,P), and (3) both survival and
resighting probability were constrained to be constant for
all regions (Model B_,,.p.).
Based on AIC, these results indicated that Model
B ,, was an improvement (i.e., had a lower AIC) over
the base model but the other two regional models were
worse than the base model (Table 4-26). Thus, our data
do not support regional differences in survival but do
support differences in resighting probability among
regions.
As above we also wished to test the hypothesis
that juvenile survival was time dependent but adult
survival was not. To test this hypothesis, we used model
Bm, _.p. as our base model and relaxed juvenile survival
to be time specific. The resulting model (Model B-.,r p)
had a lower AIC than alternative models (Table 4-27)
indicating a time effect for juvenile, but not adult,
survival. A LRT between Model B.,_,.p (the reduced
model) and Model B. ra, (the more general model) also
strongly rejected (2 = 10.204, 2 df, P=0.006),
indicating that the additional parameters for year-specific
survival of juveniles were warranted.
Parameter estimates using the model selected
from above (Model Ba.,,,) are provided in Table 4-28.


Table 4-24. Model selection statisticsfor initial set of multi-state (regional) models with and without age and time
dependency. The models) in bold text were selected as being the most appropriate (parsimonious) for these data.
Model No. Age
Model Description Classes -21n(g) np AIC
A p, 1 420.679 89 598.679
A.. 2 338.447 164 666.447
B 4. p; 1 448.420 42 532.420
B.. 0. p, 2 399.781 67 533.781
D 0. p. 1 473.808 32 537.808
D, 0. p. 2 429.037 57 543.037











Table 4-25. Resulting statistics from Likelihood Ratio Tests (LRT) between multi-state (regional) models "A", '"B
and D" with and without separate estimatesfor each age class.

General Model Reduced Model X2 df P > x2

A B 27.740 47 0.989
A. B,, 61.335 97 0.998
B D 25.389 10 0.005
B. D, 29.256 10 0.001





Table 4-26. Model selection statistics for set of models which constrained survival and/or resighting probability to
be constant for all regions. The base modelfor comparison was Model B,,. Model(s) in bold text were selected
as being the most appropriate (parsimonious) for these data.

Model Parameters constrained
Model Description to be constant across np AIC
for time dependency regions,

B. 4. p, 399.781 67 533.781
B0_. 0. p; 0 409.749 59 527.749
B ,0 p, p 490.202 55 600.202
B. ,0. p, O& p 455.231 47 549.231

' One parameter estimate is used for all regions. For example Model B,_. has one estimate ofsurvivalfor each age class
that is used for all regions, but a separate estimate of resighting probability for each age class and each region.

SIncludes parameters for transition probabilities (i. e., the probability of being in a different strata [region] between times t
and t + 1. These probabilities are better estimated using radio telemetry and consequently not reported [see methods]).




Table 4-27. Model selection statistics for set of models which constrained survival to be constant for adults,
variable forjuveniles. The base model for comparison was Model B,. Model(s) in bold text were selected as
being the most appropriate (parsimonious) for these data.

Model Parameters constrained
Model Description to be constant across -21n() n AIC
for time dependency regions,

B 0. pP -- 399.781 67 533.781

B 0 p, 409.749 59 527.749
B _,,p .. p, 4& p 455.231 47 549.231

Bd. rO 0.(< pi O&p 389.717 61 511.717

1 Oneparameter estimate is usedfor all regions. For example Model B._,. has one estimate of survivalfor each age class
that is used for all regions, but a separate estimate of resightingprobabilityfor each age class and each region.

'Includesparametersfor transition probabilities (i.e., the probability of being in a different strata [region] between times t and
t + 1. These probabilities are better estimated using radio telemetry and consequently not reported [see methods]).










Table 4-28. Parameter estimatesfor the selected multi-state model (Model B~,,), in which adults and juveniles
differ in survival, survival is constant among yearsfor adults, and survival is variable among years for juveniles.
Resighting probability in this model differs among years and regions, but is the same for adults and juveniles (i.e.,
juveniles are considered adults at the beginning of theirfirst resighting period after their fledging year).

Age Year Region A 95% C.I.() Pi 95% C.I.(p)


Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Adult
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile
Juvenile


EVER
OKEE
KISS
USJ
LOXSL
EVER
OKEE
KISS
USJ
LOXSL
EVER
OKEE
KISS
USJ
LOXSL
EVER
OKEE
KISS
USJ
LOXSL
EVER
OKEE
KISS
USJ
LOXSL
EVER
OKEE
KISS
USJ
LOXSL


0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.996
0.709
0.709
0.709
0.709
0.709
0.302
0.302
0.302
0.302
0.302
0.731
0.731
0.731
0.731
0.731


0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.715 1.000
0.228 1.000
0.228 1.000
0.228 1.000
0.228 1.000
0.228 1.000
0.036 0.568
0.036 0.568
0.036 0.568
0.036 0.568
0.036 0.568
0.000 1.000
0.000- 1.000
0.000 1.000
0.000 1.000
0.000- 1.000


0.126
0.099
0.293
0.766
<0.001
1.000
0.119
0.416
0.298
0.005
0.967
0.082
0.170
1.000
<0.001
0.126
0.099
0.293
0.766
<0.001
1.000
0.119
0.416
0.298
0.005
0.967
0.082
0.170
1.000
<0.001


0.000 0.377
0.000 0.213
0.076 0.510
0.000 1.000
0.000 0.321
0.999 1.000
0.000 0.250
0.156 0.678
0.000 1.000
0.000 0.018
0.378 1.000
0.000 0.199
0.026 0.314
0.000 1.000
0.000 0.237
0.000 0.377
0.000 -0.213
0.076 0.510
0.000 1.000
0.000 0.321
0.999 1.000
0.000 0.250
0.156 -0.678
0.000 1.000
0.000 0.018
0.378 1.000
0.000 0.199
0.026 0.314
0.000 1.000
0.000 0.237


'Region abbreviations: Everglades (EVER), Okeechobee (OKEE),
Loxahatchee Slough (LOXSL).


Kissimmee (KISS), Upper St. Johns (USJ), and









CONCLUSIONS ABOUT THE EFFECTS OF
SURVIVAL FROM BANDING DATA

Our overall conclusion from both the CJS
models and the extension to multi-state models support
(1) an age effect on survival, (2) annual differences in
survival of juveniles, but not adults, and (3) regional
differences in resighting probability, but not survival.
Both the CJS models and the multi-state models ended up
with the same model with respect to age and time
dependency; although they arrived there in different
ways. The CJS model indicated Model Ag, (i.e., age
and time dependency for both survival and resighting
probabilities) of the initial set of models was most
parsimonious. However, when survival was constrained
to be constant among years for adults (but not juveniles),
it improved the model. In contrast, the multi-state
models initially resulted in the selection of Model B,.
(i.e., survival was constant among years but resighting
probabilities differed among years). However, when
juvenile (but not adult) survival was relaxed to vary
among years, it likewise improved the model. Thus,
both the CJS and multi-state models ended up with
selection of models in which (1) adult and juvenile
survival differed, (2) adult survival was constant among
years, (3) juvenile survival differed among years, and (4)
resighting probability differed among years. The multi-
state models also supported the hypothesis that resighting
probabilities differed among regions but survival
probabilities did not.

A Synthesis of the Effects of
Survival

EFFECTS OF AGE

The results from radio telemetry and banding
data were generally consistent. Both sources of data
indicated that survival was dependent on age. Log-rank
tests between survivorship functions of radio-
transmittered adults and juveniles were significant for 2
of 3 years. Model selection from a conditional logistic
model of survival of radio-transmittered birds also
indicated an age effect on survival; as did LRTs of
banding from both (CJS) and multi-state models. In all
cases, where significant differences occurred adult
survival was higher than juvenile survival. The
exception to that pattern was the radio telemetry data for
1993 in which juvenile survival (0.87) was estimated to
be slightly (and not significantly, X2=0.243, 1 df, P =
0.622) higher than adult survival (0.86).


EFFECTS OF SEX

As explained above, we did not attempt any
analysis from banding data to evaluate differences in
survival that were attributable to the sex of the bird. Our
Kaplan-Meier estimates and our logistic regression
analyses of radio-telemetry data generally did not support
differences in survival attributable to sex of the bird.
However; survivorship functions from Kaplan-Meier
estimates did indicate a difference between males and
females for one of three years (1994) at a = 0.10.

TEMPORAL EFFECTS

Survival estimates from both radio telemetry and
banding data generally indicated annual differences in
survival among juveniles but not adults. However,
survivorship functions of adults did differ between 1992
and 1993 at a = 0.10. Although both sources of data
indicated differences among years for juvenile survival,
the parameter estimates from these two data sources
were not consistent in their relative ranking among
years. The Kaplan-Meier estimates for juveniles were
0.83, 0.87, and 0.44 for 1992, 1993, and 1994,
respectively. In contrast the CJS estimates were 0.45,
0.16, and 0.62 for 1992, 1993 and 1994, respectively.


Assumptions, Bias, and Sources of
Error for Survival Estimators

Although it is often not explicitly stated, virtually
all estimators require making assumptions. Estimators
are often robust to violation of some assumptions (i.e.,
violation of the assumption does not strongly affect its
performance) but not to others. Some of the readers of
this report may not be familiar with particular estimators
we have used or their corresponding assumptions. We
have attempted to summarize here what we believe are
the major factors that could produce spurious or
misleading results regarding our estimates of survival
and discuss what influence these factors may or may not
have on interpreting our results.

ASSUMPTIONS INHERENT IN THE STUDY
DESIGN FOR VALID INFERENCES FROM
SURVIVAL ANALYSES

Study Animals are Representative of
Population-- Random sampling within a population is
usually intended to assure that the scope of inference
from a study can validly be applied to the population of









interest. Our design was intended to systematically
accomplish that same goal. Our study area encompassed
the entire range of Snail Kites in Florida. We then
stratified our sample to be proportional to the annual
count and balanced our sample with respect to age and
sex. Thus, we believe that we have a representative
sample from the entire Florida population. We cannot
be sure that there was an equal probability of capturing
any given individual; however, we believe that
heterogeneity of catchability played only a minor role in
determining our final sample.

Study Conditions are Representative of the
Conditions of Interest- There is considerable variability
in the environmental conditions that may influence Snail
Kite populations. We recognized from the outset of this
project that our inference would be limited to the
conditions encountered during our study (Bennett and
Kitchens 1992, 1993, 1994). Most notably, drought has
been reported as a major influence on kite populations
(Sykes 1983b, Takekawa and Beissinger 1989,
Beissinger 1995), and we did not encounter drought
conditions during this study. Consequently, valid
inference from our study cannot be extended to drought
conditions. Although speculation has often been made
concerning survival during drought conditions (e.g.,
Snyder et al. 1989a, Beissinger 1995), we know of no
valid estimation of survival during such conditions.

Survival Is Independent of Other Animals-- An
assumption shared by both radio telemetry and banding
studies is that the fate of one marked animal is not
influenced by the fate of another. This assumption is
likely to be violated when being in a group of animals
results in exposure to some common risk (e.g., predation
at a roost site). Although we were unable to explicitly
test for independence, violation of this assumption does
not cause bias in the estimates of survival. Rather, it
will artificially reduce the variance of the resulting
estimates (Burnham et al. 1987, Pollock et al. 1989).
We do not believe that any severe violation of this
assumption occurred. Only once during this study did we
find mortality of more than one individual at the same
time and place. Two banded siblings, one of which was
radio transmittered, were found dead at the same site in
Loxahatchee National Wildlife Refuge at what was
believed to be the feeding site of a Great-horned Owl.


ASSUMPTIONS OF THE KAPLAN-MEIER
ESTIMATOR

Carrying a Radio Transmitter Does Not Affect
Survival-- A critical assumption of studies using radio
telemetry to estimate survival is that the radio transmitter
does not affect survival. There has been substantial
evidence in recent years to suggest that, for some
species, radio transmitters may influence the behavior,
weight gain, reproduction, or survival of study animals
(e.g., Marks and Marks 1987, Hooge 1991, Paton et al.
1991, Foster et al. 1992).
We tested for the effect of carrying a radio
transmitter on survival using both CJS and multi-state
CJS models from our banding data. These data consisted
of animals both with and without radio transmitters, thus
allowing us to specifically test for radio effects. Because
our previous analyses using CJS models had indicated
Model Aj. (see above) as the most parsimonious, we
used it as a base model for testing radio effects. We
maintained the constraints of this model for constant
adult survival among years but variable juvenile survival;
however, we derived separate parameter estimates (for
survival, resighting probability, or both) for birds with
and without radio transmitters. The resulting models all
had a higher AIC than the previously selected model
(Table 4-29). This indicates that using separate
parameters for birds with and without radio transmitters
was not supported by our model selection procedures
(i.e., there is no radio effect). This conclusion was
further supported by LRTs between models with separate
estimates for birds with and without radio transmitters
and our base model in which parameters among these
groups were equal (Table 4-29). In addition to model
selection and LRTs, the actual parameter estimates from
these models were higher for birds with radio
transmitters than for birds without radio transmitters. Of
course, we do not mean to imply that radio transmitters
somehow improved the chances for survival (our
procedures above suggest survival of birds with and
without radio transmitters does not differ); but rather,
that our parameter estimates also support the conclusion
that radio transmitters did not lower the probability of
survival.
We also explored the effects of radio
transmitters on survival using a link function (Lebreton
et al. 1992) in our multi-state models. Rather than use
a separate parameter for survival of birds with and
without radios for each group of animals (e.g., age
class), a link function (t) links the parameters to an
external variable (e.g., radios) via a linear formula.





























In this case we modeled survival of birds with and
without radios as:

0 d- 0*fdit*y(o)
and
P-ad Pdi** APdio)


Thus, the values of the link functionsAj~ and ftp) are
the only two additional parameters estimated and a test
of a radio effect is whether or not the confidence
intervals of these link functions cover 1 (i.e., a value of
1 shows no effect from the parameter).
As above, we used the most parsimonious model
from our previous analysis of multi-state CJS models
(Model B, ,,y) as a base model for testing radio effects.
For both survival and resighting probability the
confidence intervals of the link functions covered 1
(Table 4-30). In both cases the estimated link-function
values were < 1, which would imply that our survival
estimate for birds with radio transmitters was higher than
for those without radios. We concluded this assessment
by simulating a radio effect (A[4]) on the actual data and
estimating the power to detect differences in survival
attributable to that effect. This analysis showed that we
had relatively low power to detect small differences
attributable to radio transmitters, but had reasonable
power to detect substantial radio effects (Fig. 4-14).
Although our tests for an effect of radios on survival
were not completely conclusive, all evidence that we had
indicated that radios had no negative effect on survival
during this study.


Table 4-30. Parameter estimates, standard error
(SE), and 95% confidence intervals (C..) for link
functions f(O) of radio effects on survival and
resighting probability from the most parsimonious
multi-state CJS model (Model B ,a).

Parameter Estimate SE 95% C.I.

f() 0.977 0.053 0.873 1.081

ftp) 0.848 0.240 0.377 1.319






_0.8-

S0.6

b 0.4-

0.2

0.0
0.5 0.6 0.7 0.8 0.9 1.0
A



Figure 4-14. Estimated powerfor detecting differences
(A) in survival of birds with and without radio
transmitters. Differences in survival were simulated
using Model B, as a base modelfrom which
differences were imposed.


Table 4-29. Variations of base (most parsimonious) model (Model Aj, in which separate estimatesfor birds with
and without radio transmitters were derivedfor survival (4), resighting probability (p), or both.

Parameter(s)' -21n(?g) np AIC LRT2 df P> X2

none 97.101 7 111.101 -- -
0 92.697 11 114.697 4.404 4 0.354

p 89.755 13 115.755 7.345 6 0.290
0and p 87.478 14 115.478 9.623 7 0.211

Parameter(s) for which separate estimates are derivedfor birds with and without radio transmitters.

2 Compared to base model where parameter estimates are equalfor birds with and without radio transmitters.









Censoring Is Random- Censoring is the
removal of radio-transmittered animals from a sample
when the radio transmitter signal can no longer be
detected. An important assumption of the Kaplan-Meier
estimator is that the censoring mechanism is random
(Pollock et al. 1989). This means that the probability of
a bird being censored is not related to its fate (i.e.,
censored and uncensored animals have the same survival
probability). In the case of simple radio failure (a
common reason for censoring) this assumption probably
is valid. In cases where a radio is destroyed in the
process of an animal dying (e.g., a radio is destroyed
during predation or scavenging), or a radio-transmittered
animal leaves the study area (e.g., through migration, or
having been hunted or poached) this assumption may not
be valid. When the probability of a censored animal
dying is higher than an uncensored animal the resulting
survival estimate will be biased high (i.e., survival will
be over estimated).
We believe that there are 3 likely reasons
animals were censored on this study: (1) simple radio
failure, (2) when Snail Kites die they likely end up in
water which reduces the life of the radio and/or
decreases its range (i.e., they were harder to detect), and
(3) some birds left the study area.
We defined simple radio failure as the failure of
a radio transmitter that resulted from manufacturer
defects, exhausted batteries, or electronic deterioration
resulting from normal exposure to environmental
elements. We did not include in this category, radios
that had been damaged as a result of traumatic
encounters (e.g., predation or vehicle collisions) or
radios that had been exposed to conditions not normally
encountered by living birds (e.g., prolonged submersion
in water). Given the above definition, we had no reason
to suspect that the rate of simple radio failure should
differ between adults and juveniles. Consequently, the
expected rate of censoring of radio-transmittered adult
and juvenile Snail Kites due to simple radio failure also
should not have differed. An examination of the
distribution of "times to censoring" revealed a departure
from this expectation (t=3.77, df=179, P<0.001)(Fig.
4-15). Juveniles had a substantial surge in censored
animals that occurred within the first 60 days after radio
attachment that was not apparent for adults. This surge
also coincided with the period of high mortality of
juveniles. Some of this censoring was likely the result of
post-fledging dispersal into peripheral areas where our
searches were less intensive. Banded juveniles have
been occasionally recovered (usually dying or dead)
outside of the usual range of adults. Our radio telemetry
also revealed that juveniles will sometimes wander
throughout peripheral areas and later return to more


Adult








0 60 120 180 240 300 360 420 480 540 600 660 720 780
Day From Attachement


0 60 120 180 240 300 360 420 480 540 600 660 720 780
Day From Attachement

Figure 4-15. The percentage of radio-transmittered adult
and juvenile Snail Kites that were censored in each 60-
day time interval from the time of attachment.


typical habitats (i.e., temporary emigration). However,
some of the censoring probably was undetected
mortality. Birds that wander into atypical habitats are
likely to have encountered food shortages if apple snails
were not abundant. In addition, the risk of predation by
more terrestrial predators (e.g., Great-horned Owls) also
probably increased in these habitats compared to the
contiguous marshes usually inhabited. Search efficiency
for the radio signal would also decrease if either the
birds were in habitats not normally used by adults (i.e.,
we spent less time in areas where birds were usually not
present), or if mortality resulted in the radio being on the
ground (as opposed to up on a perch or flying) or
submerged in water.
Because we suspected that we may have been
failing to detect some mortality, we increased our search
effort during SY 1994 by hiring an additional field
biologist whose primary responsibility was aerial
searches for missing birds. Additional evidence that
some of the censoring of juveniles was actually
undetected mortality resulted from this increased effort.
We examined the proportions of censored and dead birds
during the first few months after attachment (i.e., < 1
September of each year) before radio batteries were









likely to have been exhausted; consequently, simple radio
failure from that cause was less likely. For adults, the
proportions of censored birds and birds confirmed dead
remained relatively constant for each year including 1994
when effort was increased (Fig. 4-16). In contrast, the
proportions for juveniles remained relatively constant for
SY 1992 and 1993, but showed a dramatic departure
from this pattern during SY 1994. During SY 1994 the
proportion of birds confirmed dead substantially
increased and the proportion of censored birds
substantially decreased. The proportion of censored
juveniles during 1994 also closely matched the
proportion of censored adults, which it had not during
1992 or 1993. This suggests that we were finding dead
birds during 1994; whereas a substantial number of dead
birds may have gone undetected during 1992 and 1993.


50

40
E
i,
so-

2 20-

1o-
n


1992 1993 1994
50 ]----------|
Juveniles Censo Birds
4 Dead Birds
so --
so

20-
I-
10 --

1992 1993 1994
Year

Figure 4-16. The percentage of adult and juvenile Snail
Kites from each sampling cohort (i.e., the year that they
fledged or were captured) which were censored or which
died between the time of capture and 1 September in each
year. Since all birds remaining in the sample will
ultimately be censored due to radio battery failure, we
used 1 September as a cutoff point which enabled
comparability among years, without the confounding of
battery failure.


Based on this evidence, we believe that we may
have overestimated survival of radio-transmittered
juveniles during SY 1992 and 1993. Although some


undetected mortality of both adults and juveniles was
certainly possible, even during 1994, the risk of biased
estimates probability was less during 1994. Our
estimates of juvenile survival from an independent data
source (i.e., banding data using either CJS or multi-state
CJS models) also were lower than the Kaplan-Meier
estimates for 1992 and 1993, but not during 1994. In
contrast estimates of adult survival from banding data did
not differ among years and the average estimates from
CIS and multi-state CIS models (0, = 0.92) were very
similar to the average of the annual estimates derived
from the Kaplan-Meier estimator (<0 = 0.90). Thus,
while we believe that survival of radio-transmittered
juveniles was overestimated for 1992 and 1993, we have
no evidence to support a similar conclusion for adults.

Small Sample Effect on Staggered Entry
Design- Small sample sizes can produce misleading, but
unbiased, estimates of survival. The Kaplan-Meier
product-limit estimator is a simple extension of a
binomial estimator (Kaplan and Meier 1958, White and
Garrott 1990). One of the characteristics of this
estimator is that the resulting product (i.e., the
cumulative estimate) is equivalent to a binomial estimate
over the interval of study (White and Garrott 1990). For
example, if we start with a total of 10 animals and during
the course of the study 2 animals die at different times,
the final estimate derived from the Kaplan-Meier
estimate will (9/10=0.9) x (8/9= 0.88) = 0.80 (Fig.
4-17); exactly the same as 8/10 from a binomial
estimator for the overall interval. However, when using
a staggered entry design the estimates for any given time
interval will be equivalent to a binomial estimate, but the
cumulative estimate (i.e., the product) will not
necessarily be the equivalent to a binomial estimate for
the overall interval (Fig. 4-17). For example, if only 2
animals are at risk over an interval, estimates of survival
can have only 3 possible outcomes: 1.0 (both lived), 0.5
(1 died and 1 lived), or 0.0 (both died). The expected
value (i.e., the mean of 6 from a very large number of
repeated experiments) of the estimate may still reflect the
"true" estimate, but the individual outcomes for intervals
with a low number of animals at risk may be an
inadequate reflection of the expected value. For an
estimate to be biased, the expected value of the estimate,
not the outcome of any particular trial, must differ from
the true parameter value.
In this study, we tried to circumvent this
problem by beginning our estimation at a time when a
sufficient sample size was established (i.e., April 15 of
each year). There is always a balance that must be
established between having sufficient sample size to
"trust" the results and not missing potentially "real"


Adults | Censored Birds
B DeadBirds






U-L *






































effects that may exist early in the sample interval. This
was less of a problem for adults after the first year
because our sample from the previous year is retained;
and newly captured animals merely supplement the
existing sample. However, juveniles from the previous
year could not be retained for our sample of juveniles
because they are no longer juveniles (they were added to
the adult sample after their first year). However, when
animals are partitioned into multiple groups for
comparison (e.g., region effects), the number of animals
at risk at any given time (even for adults) may have
been very low. Thus, we urge caution in interpreting
our Kaplan-Meier estimates without taking into account
the confidence intervals or number of animals at risk (r-)
at a given time. We have provided the details from each
Kaplan-Meier analysis (including r,) in appendices.



ASSUMPTIONS OF CORMACK-JOLLY-SEBER
MODELS

Capture and Release Occurs Over Brief Time
Interval- When this assumption is met all animals in the
study should have been exposed to the risks of mortality
for the same time (Burnham et al. 1987). This
assumption also enables a clear definition of the interval


over which survival is measured. The life history of
Snail Kites makes this assumption difficult to meet.
Snail Kites have a relatively long breeding season and
are not particularly synchronous in their breeding
attempts. Consequently, the time span over which
fledging, and therefore banding, occurs can be relatively
long. We have tried to minimize the violation of this
assumption by limiting our resighting period to the peak
four months of fledging (March-June). We do not
believe that violation of this assumption caused
substantial bias to our estimates. For adults, the highest
risk of mortality appears to be during the fall and winter.
Thus, animals within a given study year all experience
the same period of high risk. For juveniles, the highest
risk of mortality occurs over the first few months post
fledging and again all juveniles within a given cohort
were exposed to that period of high risk. Violation of
this assumption does, however, present some ambiguity
about the period of time over which survival is
estimated. As described in our methods section, we have
defined our estimates of survival from banding data
roughly as survival from one breeding season to the next.

There is No Band Loss or Misreading of
Bands- The loss of bands in studies of marked birds can
produce a serious negative bias in survival estimates
(Pollock 1981, Nichols and Hines 1993). We believe


9/10
1 8/9 =8/10 1

0.8 0.8-
1/2
S0.6- 0.6- x
*,,2/3 / 8/10
0 0.4- ( 0.4-

0.2 0.2
A B
t-- 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
a,--- 1 0 0 0 o 0 0 0 0 0 0 1 1 1 1 2 2 2 0 0 0 0
di- 0 1 0 1 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0
r--10 10 9 9 8 8 8 8 8 8 8 1 2 2 3 4 6 8 8 8 8 8

Figure 4-17. Hypothetical estimates of survival derived from Kaplan-Meier estimator where animals are entered into the
sample all at once at the beginning of the study (A) and with staggered entry over the first 30 days of the study (B). In each
case the total sample size was 10 animals and in each case 2 of the 10 animals died; however, the cumulative survival
estimates were not equal. To illustrate how the estimates were derived we have also shown the number of animals added to
the sample (a ) at time t, the number of animals that died (di), and the number of animals at risk (r ).









that the problem of band loss was negligible on our study
because almost all (859 of 913= 94.1%) birds were
marked with riveted aluminum bands that were virtually
assured of remaining on the bird. The other 5.9% were
made of PVC and anecdotal evidence suggests that band
loss from these bands also was extremely low. There
have been bands used on previous studies of Snail Kites
that were made from other materials (e.g., plastic and
Lynnply) more likely to have been lost (e.g., Bennetts et
al. 1988, Snyder et al. 1989a); however, we had no birds
with bands of these types in our sample. Thus, we
believe the potential for band loss effects on our results
was minimal.
We probably had greater potential for
misreading bands than for loss of bands. The misreading
of bands does not necessarily result in bias (unless there
is some systematic pattern to the misreading) but may
have an effect on the variance (G.C. White, Pers.
Comm.). We tried to minimize this potential effect two
ways. First, the numbering sequence of bands was
controlled by the manufacturer (ACRAFT Sign and
Nameplate Co. LTD) to minimize duplication errors.
For example, numbers that can be read as a different
number if the band were accidentally put on upside down
(e.g., 6S and 9S) were not used in a combination that
would allow accidental duplication of number
combinations. For example the number "19" could be
misread as "61" if the band were upside down. Thus,
only one of these two numbers would have been used.
Secondly, we did not record any band resighting until the
observer was virtually certain that the number had been
read correctly. Whenever possible, we also had a
second observer verify the number. This carefulness in
recording probably resulted in a lower overall resighting
probability because some bands that the observer wasn't
reasonably sure of went unrecorded; however, we were
reasonably assured that errors from misreading bands
were minimized.

Statistical Analyses are Based on Correct
Model- This assumption is the essence of all statistical
inference and its violation can seriously affect parameter
estimates (Burnham et al. 1987). In the strictest sense,
this assumption is always violated; however, our model
selection procedures and subsequent goodness-of-fit tests
help to ensure that violations are within acceptable limits;
however, when the reported model does not fit the data
then concern about bias of the estimator is warranted
(Lebreton et al. 1992).

Capture and Release Does Not Influence the
Subsequent Resighting of Animals- This is a common
and long-recognized problem for studies requiring


recapturing or resighting of animals (Pollock 1981,
White et al. 1982). We had some evidence of minor
violation of this assumption from our radio telemetry. A
few radio-transmittered birds (i.e., <5%) would depart
the immediate vicinity upon our approach after having
been captured. However, these birds usually just move
to a new perch a short distance away. We were usually
able to still read the bands through a slow careful
approach. In addition, many birds were nesting and
would consequently return to the nest vicinity within a
few minutes. There were a few (= 1%) non-nesting
individuals, however, that upon our approach, would
leave the vicinity and we would be unable to read their
bands. There were also occasionally unmarked
individuals that would depart the vicinity upon our
approach, suggesting that some birds are inherently more
wary, regardless of previous capture history. Because
radio-transmittered birds often had a more traumatic
capture experience (i.e., many were captured with a net
gun) than birds that were merely banded as nestlings, we
believe that our observations of radio-transmittered birds
overestimates the extent of avoidance as a result of
having been banded. Consequently, although our ability
to relocate birds was probably influenced by birds having
been captured, our observations of radio-transmittered
birds suggests that this influence is not substantial.

Causes of Mortality

Although our sample size of dead birds was not
large, it was sufficient to provide an indication of the
relative frequencies of different causes of death.
Predation appears to have been the most frequent cause
of death for both adults and juveniles (Fig. 4-18). We
caution, however, that without finding each carcass while
it is still fresh and conducting a necropsy, it is impossible
to distinguish predation form post-mortality scavenging.
However, we usually had ancillary information, in
addition to the carcass having evidence of having been
eaten (e.g., > 1 carcass at the site or the location
inconsistent with a site normally used by Snail Kites),
when predation was assigned as the probable cause of
death (Table 4-31).
Emaciation was the second most frequent cause
of death for juveniles, but was not an observed cause of
death for adults during this study. Here we caution that
our data were collected during non-drought years.
During widespread droughts, when food may be scarce,
emaciation may be a more frequent cause of death for
both age classes. Emaciation may also have been
underestimated in our sample. For example, we
suggested that some juvenile deaths may have been









included as "censored" birds as a result of birds
dispersing to habitats atypical of adults and subsequently,
less adequately searched. These areas may also be more
likely to have less predictable food resources. Two of
the emaciated birds were found in marine environments
where apple snails are completely lacking.
Other causes of death included vehicle collisions,
disease, and one probable gunshot. Vehicle collisions
were observed for both age classes and one additional
death from this cause was observed for an unbanded or
radio-tagged adult (i.e., it is not included in this sample).
Deaths from vehicle collisions may be more likely when
nesting or foraging concentrations occur adjacent to
roadways. One adult female died of an infection of the
coelomic cavity. One juvenile may also have had an
intestinal disease, but severe autolysis precluded
conformation. The skeletal remains of one juvenile had
a probable gunshot (shotgun) hole through its sternum,
but we were unable to confirm if this was the cause of
death.


Table 4-31. Known mortality of snail kites banded or radio-tagged during this study.

Year Age Location of Mortality Probable Cause of Death

1992 JUV WCA-3A Predation'
1992 JUV St Johns Marsh Predation'
1992 JUV Holmes Beach, FL Emaciation2
1992 JUV Florida Bay Emaciationz
1992 JUV Everglades Agricultural Area Vehicle Collision'
1992 JUV WCA-2A Unknown4
1992 JUV Agriculture Area (Collier Co.) Unknown4
1993 JUV West Palm Beach W.C.A. Predation'
1993 JUV Everglades National Park Predation'
1993 AD Lake Okeechobee Predation'
1993 AD Lake Okeechobee Predation'
1993 AD WCA-3A Infection of Coelomic Cavitys
1993 JUV Coquina W.C.D. Unknown6
1993 AD West Palm Beach W.C.A. Unknowns'7
1994 JUV WCA-3A Predation'
1994 JUV WCA-3A Predation'
1994 JUV WCA-3A Predation'
1994 JUV WCA-3A Predation'
1994 JUV Everglades National Park Predation'
1994 JUV Lake Kissimmee Predation'
cont.


ADULTS JUVENILES
(n=13) (n-30)
385% 36.7%



7.0.0%%
7.746.2% 3.3 46.7%

E,* 0 v-a c.El 0 m coao, n a P Man UO 0, \o


Figure 4-18. The percentage ofmortality of adult and
juvenile Snail Kites in each of 6 mortality classes.
Particularly for causes of death in which a necropsy was not
performed (e.g., predation), we can never determine the
cause of death with certainty. However, we assigned
mortality to each class only when ancillary evidence
supported our conclusion. In the absence of such evidence,
we assigned mortality to the unknown class.










Table 31. Continued


1994 JUV
1994 JUV
1994 AD
1994 AD
1994 JUV
1994 JUV
1994 JUV
1994 AD
1994 AD


Loxahatchee N.W.R.
Loxahatchee N.W.R.
Lake Kissimmee
Lake Okeechobee
Everglades Agricultural Area
WCA-3A
Big Cypress National Preserve
WCA-2A
Lake Okeechobee


1994 JUV Lake Kissimmee


1994 JUV
1994 JUV
1994 JUV
1994 JUV


Everglades National Park
East Lake Tohopekaliga
Lake Okeechobee
C- ll Basin


1994 AD East Lake Tohopekaliga
1995 AD Lake Okeechobee
1995 AD Oesceola Co. (private land)
1995 JUV Lake Marion
1995 JUV East Lake Tohopekaliga
1995 JUV Lake Marian
1995 AD Glades Co. (State Highway 78)
1995 AD St. Johns Marsh
1995 JUV Lake Marian


Predation'
Predation'
Predation'
Predation'
Unknown (not predation)8
Unknown (not predation)s
Unknown (not predation)8
Unknown (not predation)8
Unknown (not predation)s
Unknown4
Unknown4
Unknown4
Unknown4
Unknown4
Unknown4
Predation'
Predation'
Predation'
Predation'
Emaciation2'9
Vehical Collision2
Unknown (not predation)8'10
Unknown4


Carcass showing clear signs of having been eaten (e.g., feathers plucked, bones broken) with additional ancillary evidence
supporting having been taken by a predator (e.g., other carcasses found at the site, feathers of predator [Great-horned
Owl], and location or conditions under which the carcass was found [e.g. on limbs of trees]). However, we can never be
certain that some of these were not a result of scavenging after death.
2 Based on necropsies performed by the National Wildlife Health Research Center.
SBird found still alive along farm road with broken wing.
SCarcass too decomposed for evaluation.
SBased on necropsy performed by the University of Florida Laboratory of Widlife Disease Research.
6 Skeletal remains of bird banded by Jon Buntz (GFC)found at base offence post. Small hole in sternum was consistent with
shotgun pellet.
7 Bird was in excellent nutritional health and had 3 snails in esophagus. No external signs of trauma.
3 Carcass was completely intact (i.e., no sign ofpredation or scavenging), but was too decomposed for further evaluation.
9 Bird was severely emaciated; however, some evidence of intestinal disease which may have contributed to death.
"o Decomposition was too severe to determine cause of death, but bird was emaciated.


Previous Estimates of Survival

There have been several previous reports of
estimated survival (e.g.,Snyder et al 1989a, Beissinger
1995). To our knowledge these estimates emerged


primarily from 3 sources of information: (1) the annual
count, (2) a study by Snyder et al. (1989a) with banded
birds, and (3) a study by Snyder et al. (1989a, 1989b)
using radio transmittered birds.
Snyder et al. (1989a) reported the results of
resighting birds banded from 1968-1978. They did not












use standard capture-recapture methodology (i.e., CJS
models) presumably because of a lack of sufficient data
for most years. They point out that only a fraction of the
birds were checked for bands in any given year.
Consequently, they estimated a range for minimum
annual survival, taking into account all possible band
loss. Thus, actual survival estimates from these data
would range from their lower minimum estimate to 1.0.
All of the resightings they report were from 1979 when
their resighting effort was most intensive. Consequently,
their estimates of minimum annual survival also
confound adult and juvenile survival since all birds,
except those banded in 1978, were banded as nestlings
and resighted as adults. The estimate for birds banded in
1978 would not confound adult and juvenile survival,
since the estimates were only for minimum survival of
their first year (i.e., the estimate was for juveniles only).
For 4 of 10 years their lower estimate was 0 (i.e., actual
survival ranged from 0.0 to 1.0) and the average of their
lower estimates was 0.47 (i.e., survival ranged from
0.47 to 1.00). They suggested that adult survival
probably exceeds 0.9 under good conditions, but did not
present any estimation procedure to support this
suggestion. Beissinger (1995) later reported adult
survival during high-water years was 0.95 (0.03 SD).
He reported that these estimates were projected from the
studies by Snyder et al. (1989a, 1989b); however, he
reported no estimation procedure for either survival or
its standard deviation; nor could we find such procedures
in the sources cited. Beissinger later informed us (S.
Beissinger, pers. comm.) that these were estimates in the
sense of approximations and were not derived using a
statistical estimator (e.g., maximum likelihood), and that
the standard deviations were intended as a way of dealing
with the uncertainty of the survivorship information in
his stochastic model.
Estimates of survival during droughts,
particularly the 1981 drought, have been frequently
reported based on changes between years in the annual
count (e.g., Beissinger 1986, 1988, 1995, Takekawa and
Beissinger 1989). The count in 1980 (before the 1981
drought) was 652 birds (Sykes et al. 1995)(reported as
654 by Rodgers et al. [1988]). The count made in
December of 1981 (after the drought) was 109 birds (an
83% decline in the count)(Rodgers et al 1988, Sykes et
al 1995); although an alternative count for 1981 of = 250
(done independently by Beissinger [1982, 1984] in
March of 1982) is often substituted for the state-
conducted count, implying a population decline of 60%
(e.g., Beissinger 1984, 1986, 1988, 1995, Takekawa and
Beissinger 1989). We suggest that using the annual
count for estimating survival is not scientifically valid
because it is subject to multiple sources of error that are


inconsistent among years (discussed in detail in chapter
on Monitoring Snail Kite Populations in Florida).
Estimating survival from the annual survey during a
drought is particularly suspect because it is well known
that Snail Kites disperse in large numbers to peripheral
wetlands during droughts, where counts are not
conducted (Beissinger and Takekawa 1983, Takekawa
and Beissinger 1989) and this dispersal has not been
taken into account in these survival estimates. Snyder et
al. (1989a, 1989b) reported that at least 7 of 8 (one was
of unknown fate) radio transmittered birds survived from
May 1981 (just prior to the peak of the 1981 drought)
until their study period of 1982. Although their sample
size was small, this implies an estimate of survival
during droughts as 0.875 (rather than the 0.17 0.40
inferred from the annual count). If these data are
considered as following a binomial distribution (and
assuming that the unknown bird was dead), then a 95%
confidence interval would be 0.65 1.00 and does not
even include the estimates inferred from the annual
count. Beissinger (1995) later suggests that these data
are more applicable as lag-year (the year following the
drought) estimates because radios were attached toward
the latter part of the actual drought after these birds had
survived some of the dropping water levels and the risks
associated with dispersal to Lake Okeechobee. We
disagree for several reasons. First, our data show that
the risks associated with dispersal are encountered on a
regular basis (i.e., approximately 25% of the population
disperses every month) whether there is a drought or not.
We do agree that the risks of predation may be greater
in peripheral habitats because of increased proximity to
upland habitats and their associated predators
(particularly Great-horned Owls); however, dispersal to
these habitats is not likely to be extensive until alternative
wetlands have dried (i.e., at the peak of the drought).
We also suggest that much of the mortality associated
with droughts will occur during the peak of the drought
and during the first winter after a drought, when
depressed food levels are compounded by cold
temperatures, which decrease food availability even
further (Cary 1983, 1985). Both of these periods of high
risk were included in their sample of radio-transmittered
birds and therefore should better represent drought-year,
rather than lag-year, conditions. However, regardless of
how these data are interpreted, there still remains a need
for credible estimates of drought-related survival.







































ChapterS. REPRODUCTION


Beo-- the eof th" "pt o I" 0, debl



bchh--,; ... Igy) hllb c h I g-"le, oa
b-o b,,t 0 e 1 10 -W I'Mo t,, -1o, t( I
0 d by Beo O 1988,O ykeCtOb]199 5 I ).
-h B foO 1 15 10 0
Rp"od"l ooool SoooKoa\ t lot 0
'cl.dollah 0000iici 00iIlll I u nu

ed~t b el We h,,,, b hpl,,d a ,,pi-
ll 0 01, oh o h- -0-y Ie
ooouia o l,10adal 0,1,00 1'v '10p c a xpl
Ill ooooop'oolloinyrcl ho dlrlnmi '00,000
,ooohoooooooilr;lid 00 0nrd dolle 1,1I 3 0000000
........ leleuiede',pl..


ot female young) produced pe female iofeach age Cass
1 Ihe population Untorunael., lor many species,
including Snall Kbe. we c-nnol estlimae this paran-
eler directly Rlihcr. It is delved h1om the proportion
ol buds attempting to breed (a.), ihe propormn of
breeding ittemp, that are 1 ucceful (S,), and the num-
ber ol bleeding atlemllpt pc yea (fi,) For succesful
0estig atcmpt,, we all need o, knolw the lni0 be of
young produced (Y,) and the sex ratio of the young
piodu0ed (R,)(Brwn 1974,1 Caughicy 1977(Fig 5-
1) l Alni chaplel wc review the imotmalon that has
beeli pi v~otl y led on e"ach ol these pat -meel ,
a0 well a 1'pren 0 ll baed oo 0 lew00 dh0ata

SEMANTICS
Mlundl; tandlngs houtme ,asules ot reproduce
mn cl frequently be attrlbulble to a lick of car
deimtlans ol what is belng mnastulcd anid/or to wiat i


owr, to avoid Ihe Ilmet lype. we will begin or


























assessment of reproduction by providing operational
definitions for terms discussed below.

Breeding Attempt- There has been considerable
disagreement among researchers regarding what
constitutes a breeding attempt. For the purposes of this
report, we consider a breeding attempt to begin with the
laying of the first egg Steenhof (1987). Snyder et al.
(1989a) considered a breeding attempt to begin with nest
building, prior to the laying of the first egg. They
suggested that to ignore the period before egg laying in
analyses of reproductive success would be ill-advised
because of the high proportion (>0.33) of nests they
observed that failed prior to egg laying. We agree
entirely with Snyder et al. (1989a) that, for many
questions, the failure of nests prior to egg laying may
have important biological implications. These failures
may provide insight as to environmental conditions at the
time of egg laying and also may provide information
regarding behavioral aspects of mate choice. However,
we disagree that nests during the nest-building stage for
this species should be considered as a nesting attempt for
estimation of reproductive parameters. We have several
reasons for this conclusion.
First, inclusion of "pre-laying" failures may
include nests in which a pair bond has not even been
established between a male and female. Nest building is
initiated by the male as part of courtship (Beissinger
1988, Bennetts et al. 1988) and more than one male may
direct courtship toward a single female (Beissinger 1987,
pers. obs.). Thus, if two males initiated nest building as
part of the courtship toward a single female, this would
be considered as two nesting attempts using the definition
of Snyder et al. (1989a), even though only one of these
nests may produce young. Similarly, our observations
indicate that a single male may exhibit courtship


behavior, including nest building, towards several
females in succession. This behavior may last as little as
a few hours or may last several days and may then be
redirected to a new female if a pair bond is not
established. We observed a single radio-tagged male
direct courtship to as many as five different females
before a pair bond was established that resulted in egg
laying. Using the definition of Snyder et al. (1989a),
each of these courtship attempts would have been
interpreted as a failed breeding attempt. In contrast, we
view this as part of the mate selection (courtship) process
rather than as a demographic parameter.
Second, the passage of cold fronts and corresponding
temperature change often results in reduced food
availability (Cary 1985). Consequently, courtship is
often terminated with the passage of cold fronts and
resumed (often at a new location) when temperatures
return to pre-front conditions (Beissinger 1988, Bennetts
etal. 1988). Thus, if two cold fronts passed before eggs
were actually laid, the pair would have been considered
to have made three separate breeding attempts (with two
failures) even if the pair successfully raised a brood.
For demographic purposes, we view these
postponements as courtship interruptions, rather than
multiple breeding attempts with each interruption being
considered as a breeding failure. Third, because nest
building begins with the placement of the first stick and
many more courtship nests are probably initiated than are
ever detected, it creates a substantial bias in the estimate
of success if these early starts are not detected (Mayfield
1961, Miller and Johnson 1978, Johnson 1979, Hensler
and Nichols 1981).
Finally, it is well known that nesting raptors tend
to be considerably more sensitive to disturbance early in
the nesting cycle (Grier and Fyfe 1987, Steenhof 1987).
Although previous investigators have reported a high








proportion of nest abandonment by Snail Kites prior to
egg laying (e.g., Beissinger 1986, Snyder et al. 1989a),
we have seen no accounting for how much of this
abandonment might be attributable to disturbance by the
investigators themselves. In contrast, abandonment of
eggs or young by Snail Kites is extremely rare (Bennetts
et al. 1994, Sykes et al. 1995). Thus, measuring nesting
success after the first egg has been laid can reduce this
potential source of confounding and minimize
disturbance to this endangered species.
Based on these concerns, we define a breeding
attempt to begin with the laying of the first egg. Thus,
unless otherwise stated, references to nests in this report
implies the presence of eggs or young.

Successful Nest- For the purposes of this
report, a successful nest is one in which at least one
young reaches fledging age (Steenhof 1987). Because
birds after fledging may or may not be present at the
nest, we defined fledging age as 80% of the average age
of first flight (Steenhof and Kochert 1982). Snail Kites
are capable of first flight at approximately 30 days of age
(Chandler and Anderson 1974, Beissinger 1988, Bennetts
et al. 1988); thus, we considered a nest as having been
successful if it produced young that survived to at least


24 d (Bennetts et al. 1988). At this age, animals are
reasonably assured of still being at the nest and mortality
for most raptors between this time and fledging is
minimal (Milsap 1981, Steenhof 1987). In addition, we
banded birds at the time they were determined to be of
fledging age. Consequently, any mortality that occurred
after this age would have been included in our estimates
of juvenile survival from our mark-resighting program.

The Breeding Season

The initiation of nests (i.e., egg laying) has been
documented in all months of the year (Sykesl987c);
although, for any given year, Snyder et al. (1989a)
observed a maximum breeding season (interval over
which nests were initiated) of 31.7 weeks (7.9 months)
during an 18-year study in Florida. Although Snail Kites
in Florida can potentially lay eggs in all months of the
year, there is a very distinct seasonal distribution of nest
initiations (Table 5-1) (Fig. 5-2). Nest initiations begin
as early as November, but in most years widespread
initiations usually do not begin until January or February.
Peak initiations usually occur in March, but are often
several weeks later, peaking in April, in the northern
habitats (Toland 1994).


Table 5-1. The number of nest initiations reported in each month during studies from 1966 through 1995.
Sykes Snyder et al. Snyder et al. Bennetts et al. Toland This Proportion of
Month (1987c) (1989a)1 (1989a)2 (1988)' (1994)4 Study5 Total" Total
OCT 5 0 0 0 0 5 0.00
NOV 9 9 8 -- 0 0 17 0.01
DEC 5 35 35 0 7 47 0.04
JAN 26 90 81 14 3 74 184 0.16
FEB 35 98 78 102 22 66 201 0.17
MAR 36 147 119 125 49 106 310 0.26
APR 21 114 103 102 53 40 217 0.18
MAY 10 64 56 32 19 17 102 0.09
JUN 5 44 38 0 16 0 59 0.05
JUL 1 27 25 0 5 0 31 0.03
AUG 2 1 1 -- 0 0 3 0.00
SEP 1 0 0 0 0 1 0.00
Total 156 629 544 375 167 310 1177
Includes all years reported by Snyder et al. (1989a).
SIncludes only years reported by Snyder et al. (1989a) to have wide seasonal coverage (1970-1982).
Unpublised data from nests reported by Bennetts et al. (1988). Seasonal coverage was limited to
January through July.
Unpublished data courtesy ofB.R. Tolandfrom nests reported by Toland (1994).
Based on data from 1994 and 1995.
SBased only on data with wide seasonal coverage (Sykes 1987c, Snyder et aL 1989a [1970-1982 only],
Toland 1994, and this study.











S0.25-
020


S015 0

0.05

OCT NOV DEC JAN FEB MAR APR MAY JUN JUL AUG SEP
Month
Figure 5-2. The proportion of nest initiations for each month
of the year based on cumulative data reported by Sykes
(1987c), Snyder et at (1989a)(1970-1982 only), Toland
(1994), and this study.


The Breeding Population

AGE OF FIRST REPRODUCTION

Sykes (1979) reported that Snail Kites are
capable of breeding at 3 years of age. However, Sykes
(1979) suggested that some birds possibly breed at a
younger age. Beissinger (1986) later reported both male
and female birds breeding at one year of age and Snyder
et al. (1989a) reported one female breeding at nine
months. Our data are consistent with Beissinger (1986)
and Snyder et al. (1989a). During this study, we
commonly observed yearling Snail Kites attempting to
breed.

PROPORTION OF BIRDS ATTEMPTING TO BREED

Adults- Nichols et al. (1980) suggested that the
proportion of birds that attempted to breed during
favorable conditions was quite high. They suggested that
there was no reason to suspect that it was not 1.0 and,
consequently, assumed that value for their demographic
model. They reported, however, that this was a crude
estimate for lack of a better one. Beissinger (1995)
similarly reported that the proportion of adult Snail Kites
attempting to breed during high-water years was 1.0, but
also provided no empirical evidence. Although our data
for this parameter are very limited, they are consistent
with these earlier estimates. During 1995, we closely
monitored 25 radio-transmittered adult Snail Kites for
breeding activity in order to assess the proportion
attempting to breed and the number of breeding attempts
per year. Of these 23 adults, 14 were females and 9
were males. During the 1995 nesting season, we located
each bird on the ground approximately bi-weekly to


determine its breeding status (e.g., a nest, courtship, not
breeding). Birds in which no breeding activity was
detected were generally observed for 22 hrs and
subsequent visits, usually within 10 days, were required
to confirm a non-breeding status and to confirm any
nests for birds exhibiting courtship. During 1995 (a
relatively high water year throughout the kite's range),
all 23 (100%) adults attempted to breed at least once.
Our estimate is based on a relatively small sample
(N=23) and on only one year; however, it does provide
an empirical basis that most, if not all, adults may
attempt to breed in some years.
Sykes (1979) reported that he observed no
nesting attempts during 1971 (a widespread severe
drought; see Management and Conservation). Based on
this observation, Nichols et al. (1980) assumed that no
birds nested during 1971 for their demographic modeling
effort. Beissinger (1986) reported that during the 1981
drought 80-90% of the kites did not attempt to nest, and
Beissinger (1995) later reported that only 15% of adult
Snail Kites attempt to breed during drought years.
However, no empirical evidence was presented in
support these estimates. Based on anecdotal evidence,
we believe that the proportion of birds attempting to
breed during drought years may be highly variable
depending on the spatial extent of the drought (see
discussions on Water Management and Snail Kites in
Management and Conservation). We agree that during
a severe widespread drought, most birds probably do not
attempt to breed. However, in cases of more localized
droughts, where portions of the kite's range may not be
experiencing dry conditions, the proportion of birds
attempting to breed may remain very high. For
example, during 1991, the Everglades region was at the
end of a 2-3 year drought (whether it was a 2 or 3 year
drought depends on how a drought is defined). During
this year almost no nesting activity was observed in the
Everglades region (J.A. Rodgers Jr., pers. comm.).
This would appear consistent that a small proportion of
birds had attempted to breed. However, during this
year, record numbers of birds were nesting on Lake
Tohopekaliga (J.A. Rodgers and J. Buntz, pers. comm.)
and in the upper St. Johns Marshes (B. Toland, pers.
comm.), areas not influenced by the drought conditions
in the Everglades. Our data on movement strongly
suggest that the Florida population is one population that
moves frequently throughout its range, rather than a
meta-population of quasi-isolated subpopulations (see
Movements). Thus, in years where drought is not
widespread, birds may merely shift the location of
nesting activities. Consequently, we suggest that this
parameter may be quite variable and needs to take into
account the severity and spatial extent of a given









drought.
Sykes (1979) observed relatively few (n=6)
nests during 1972, the lag year following the 1971
drought. The average number of nests per year that
Sykes (1979) reported from 1968-1976, excluding 1971,
was 23. Based on this observation of reduced nesting
during this lag year, Nichols et al. (1980) assumed a
proportion of 0.5 adults attempted to nest during 1972.
Beissinger (1995) reported that a proportion of 0.8 adults
attempt to breed during lag years, although we could find
no empirical support for this estimate in any of the
sources cited. We suspect that, similarly to drought
years, this parameter may be highly variable depending
on the specific drought. Thus, we view this parameter
as also being unknown and subject to high variability.

Subadults-- Snail Kites have been reported to
breed as young as 9 months old (Snyder et al. 1989a);
thus, by a calendar-year definition Snail Kites are
capable of breeding as juveniles (i.e., < 1 year old).
However, these cases are ones in which the birds
attempted to breed during the nesting season following
the nesting season of their hatch year. Thus, this
parameter should be defined as the proportion of birds
attempting to breed during their second breeding season
(the first is the one in which they hatched).
Based on data from Snyder et al. (1989a),
Beissinger (1995) reported that 25% of subadults attempt
to breed during high water conditions. Snyder et al.
(1989a) observed 8 banded subadults breeding during
1979 out of a minimum of 74 that had survived from
their hatching year of 1978. Because Snyder et al.
(1989a) only checked 50.8% of the nests for bands, they
estimated that there were probably 16 subadult breeders
out of a minimum of 74 banded subadults (22%). Of
course, this estimate assumes that only the 74 subadults
observed alive in 1979 had survived and that there was
an equal probability of detecting a banded subadult that
was breeding in the sample of nests that were checked
and those that were not checked.
During 1992, we estimated a similar percentage
of 17% of the subadult birds attempting to breed
(Bennetts and Kitchens 1992). Our estimate was based
on only 2 breeding birds of 12 banded yearlings that we
observed during the 1992 breeding season.
Consequently, our estimate requires similar assumptions
that we suggested above for Snyder et al. (1989a).
During 1995 (a high water year throughout the
kite's range), we also closely monitored 9 radio-
transmittered juvenile Snail Kites for breeding activity
(as described above). Of these 9 birds 3 (33%) attempted
to breed. All of the estimates derived from the data of
Snyder et al. (1989a), as well as from our own data, are


very limited (i.e., small samples each from one year);
however, they do consistently suggest that a relatively
small proportion of subadults do attempt to breed during
some years.
Beissinger (1995) also reported that the
proportion of subadults attempting to breed during
drought years and lag years was 0.15. We could find no
empirical basis for this estimate in any of the sources
cited; but we agree with Beissinger that the average
percentage would probably be lower when conditions are
poor in part or all of their range.

Nest Success

Nest success has been among the most widely
estimated parameters of reproduction of Snail Kites.
However, it has probably also been among the most
confusing. There are several areas of disagreement
among researchers regarding estimation of nest success.
The disagreements center primarily on which nests
should be included in the sample and what estimator
should be used. Consequently, nest success has been
difficult to compare because different researchers have
used different estimators and have included or excluded
different categories of nests within their respective data
sets. We have attempted to summarize below the major
issues of contention. We have also summarized the
literature on nest success and explicitly pointed out which
estimator was used and what categories of nests were
included or excluded in the sample. Thus, readers can
make comparisons among studies and decide for
themselves which estimates are most appropriate for
their particular needs.


AREAS OF DISAGREEMENT
ESTIMATION OF NEST SUCCESS


REGARDING


Inclusion or Exclusion of Nests Found at
Different Stages- At what stage a given nest is found can
greatly influence its probability of success. Nests found
late in the nesting cycle have a higher probability of
success because they have less observation time during
which they are at risk. A Snail Kite nest requires at least
57 days to fledge young (27 days of incubation and 30
days for nestlings to reach fledging age). Thus, a nest
found during egg laying will have potentially >50 days
"at risk" (provided it does not fail earlier) to be
considered successful. In contrast, a nest found close to
the time of fledging may have only a few days "at risk"
to be considered successful. Consequently, estimates of
nest success that were derived using nests found late in
the nesting cycle tend to be biased high (Mayfield 1961,








1975, Miller and Johnson 1978, Hensler and Nichols
1981, Hensler 1985).
Nests at different stages also are vulnerable to
different risks. For example, rat snakes (Elaphe
obsolete) are believed to be one of the major predators of
Snail Kite nests (Bennetts and Caton 1988). Rat snakes
will readily take eggs or young that are less than one
week old; however, the larger size of older nestlings
largely precludes predation by rat snakes. Consequently,
nests found when young are > 1 week have an inherently
lower risk of predation by rat snakes.
Some researchers (e.g., Beissinger 1986, Snyder
et al. 1989a) also have included nests prior to eggs
having been laid (i.e., during nest building) in deriving
estimates of success. We disagree with this practice for
the reasons previously discussed (see definition of
Breeding Attempt in earlier section of this chapter on
Semantics).
Because of these biases, estimates of nest
success can be substantially influenced by what nests
(i.e., found at what stage) are included or excluded for
deriving a given estimate. This makes comparison of
previous estimates of nest success for Snail Kites difficult
because researchers have not used the same criteria for
inclusion or exclusion of nests found at different stages
when deriving their estimates. Steenhof and Kochert
(1982) suggested three ways to minimize this type of
sampling error for estimating nest success. First, they
suggest estimating success based on a pre-determined
sample of territorial pairs. However, because Snail
Kites do not maintain nesting territories from one year to
the next, this solution is not feasible for this species.
Secondly, they suggested using estimates derived only
from nests that were found during incubation (by
definition, they considered a breeding attempt to have
begun after they laying of the first egg). This suggestion
is feasible for kites; but of the previously reported
estimates, only Snyder et al. (1989a) reported estimates
using this criterion (but they also included manipulated
nests for which some strong assumptions were made; see
discussion of Manipulated Nests below). Their third
suggestion was to use the Mayfield Estimator (see
discussion of Mayfield vs Conventional Estimators
below), which is intended to account for the bias
imposed by not finding all nests during early stages. Of
the previously reported estimates, only Bennetts et al.
(1988) reported estimates using this estimator.
Given the differences in what nests were
included or excluded in previous studies, we urge caution
in making comparisons among previous studies. We also
agree with Steenhof and Kochert (1982) that estimates of
success should be derived either using only nests that
were found during incubation or using the Mayfield


estimator. Of these two approaches we prefer the latter
(discussed below in section on Mayfield vs Conventional
Estimators), although there remains disagreement among
researchers regarding this conclusion.

Manipulated Nests- Nests that occur in cattails
may have a tendency to collapse under conditions of high
winds or waves (Sykes and Chandler 1974). This led to
a previous practice of placing nests that were subject to
this type of failure in artificial nest baskets (Chandler
and Andeson 1974, Sykes and Chandler 1974). Because
this may influence the outcome of a given nest, whether
to include or exclude these nests has been the subject of
some debate (e.g., Beissinger 1986, Snyder et al.
1989a). Similarly, when these nests have been included
in samples from which estimates of nest success were
derived, there have been differences among researchers
(e.g., Sykes 1979, Snyder et al. 1989a) as to how these
nests were treated in the derivation of nest success.
Sykes (1979, 1987b) included 43 nests that were
placed in artificial nest baskets in his sample for
estimating success. These nests were not treated
differently than other nests. Snyder et al. (1989a) later
criticized this use of manipulated nests. They suggested
that the success of manipulated nests was higher than if
they had not been manipulated, and that this would have
biased Sykes's estimate of success upward. Snyder et al.
also presented estimates of nest success using 94
manipulated nests. They argued that because these nests
were in imminent danger of collapse, they considered
them all as failures. They suggested that to exclude
them, as was done by Beissinger (1986) and Beissinger
and Snyder (1987), would have also biased success
upward because these manipulated nests were not a
random sample with regard to their probability of
success (i.e., that they would have failed). In contrast to
their suggestion, we have observed collapsed nests
containing older (> 10 d old) nestlings that have been
successful. Although we agree with Snyder et al.
(1989a) that exclusion of these nests probably would
have biased success upward, we also believe that
including them all as failures probably would have biased
their estimate slightly downward. Our tendency is to
agree with the solution of Snyder et al. (1989a), but to
accept that there might be a slight bias toward
underestimation of success.
An additional concern that has not been
addressed by previous authors is that the susceptibility of
nests to collapse may be influenced by the investigators
themselves. The vulnerability of nests to collapse can be
greatly influenced by the paths of airboats while
conducting nests visits, particularly in cattails (Bennetts
1996). Airboat trails are often wide enough to allow








increased susceptibility to wind damage and/or to weaken
the structural support provided by the cattails adjacent to
the nest. This type of damage can be minimized, if not
eliminated, by maintaining a substantial distance from the
nest during an approach and either wading in to nests or
using a mirror pole from a distance to check them
(Bennetts 1996). Nest baskets have not been used in
recent years and we do not anticipate (or advocate) a
recurrence of their use. Although some nest collapse
still occurs in some areas, particularly on lakes (J.A.
Rodgers, pers. comm.), we do not believe that the
benefits of nest baskets warrant the effort or disturbance
for their use as a general management tool. They may,
however, be warranted for isolated special
circumstances. Previous use of nest baskets had been
initiated when numbers of Snail Kite probably were
much lower than are currently found. We do, however,
advocate that researchers must exercise extreme care to
avoid influencing the outcome of nests being monitored.

Mayfield vs Conventional Estimator- Mayfield
(1961, 1975) proposed an estimator for nest success that
was based on daily exposure (risk) such that a daily
probability of success was derived using only those days
in which a given nest was under observation. Overall
success is then derived by applying the daily success over
the length of the interval being estimated. This approach
provides an estimate of success that is unbiased with
respect to when a given nest was found, but requires an
assumption that the probability of success is constant
over the period (e.g., incubation) being estimated.
Hensler and Nichols (1981) later showed, using Monte
Carlo simulations, that this estimator was superior to the
conventional estimator under a wide variety of
conditions.
Bennetts et al. (1988) used the Mayfield
estimator for nest success of Snail Kites and found it to
perform favorably for this species. They found some
violation of the assumption of constancy (e.g., success
differed between incubation and nestling stages);
however, this assumption can be overcome by using
separate estimates for periods that differ (Hensler and
Nichols 1981). Snyder et al. (1989a) later argued that
the Mayfield estimator was inappropriate for Snail Kites
because the interval length for nest building was too
variable to apply this estimator. We agree with Snyder
et al. (1989a) that the Mayfield estimator would be
inappropriate for estimation of success during the nest
building stage. However, we also believe that the nest-
building period is inappropriate to include in estimates of
nesting success for this species (see discussion of
Breeding Attempt in earlier section of this chapter on
Semantics). Consequently, we disagree with Snyder et


al. (1989a) that the Mayfield estimator is inappropriate
for estimating nesting success of Snail Kites. Rather, we
agree with Hensler and Nichols (1981), Miller and
Johnson (1978), Steenhof and Kochert (1982), and
Steenhof (1987), that this estimator is preferable to
conventional estimates of nesting success because of its
ability to produce unbiased estimates of nesting success.

ESTIMATES OF NEST SUCCESS AND ITS
PROCESS VARIANCE

Given the wide disagreement among researchers
regarding nesting success, we suggest that future
researchers be specific about what is being included or
excluded, and that consideration be given to reporting
success both by conventional and Mayfield estimators so
readers have the ability to compare their results. We
have also provided a summary of the previously reported
estimates, showing what nests (i.e., found at what stage)
were included in each estimate, whether or not
manipulated nests were included, and which estimator
was used (Table 5-2).
We estimated the mean annual nest success (S,)
as 0.32 based on reported nest success from each year
using estimates that were based on nests (in which at
least one egg has been laid) that were found during the
egg stage (Table 5-3). However, some years had
extremely low sample sizes, which may have precluded
a reliable estimate for that year. If we had excluded
estimates for those years with < 10 nests, we would have
estimated mean annual nest success (S,) as 0.28.

Estimate of Process Variance- It is important
to recognize that there are several distinct variance
components associated with demographic parameters
(White et al. 1982, Burnham et al. 1987). A
demographic parameter (e.g., survival) may vary over
time (temporal variation) or among locations (spatial
variation). There is also likely to be heterogeneity
among individual (individual variation) in their
probability of survival due to genetic or phenotypic
variation (DeAngelis and Gross 1982). Each of these
sources of variation are a type of population variation
(Burnham et al. 1987). There is also variation
attributable to sampling populations. Unlike these
previous sources of variation, sampling variation is not
a measure of population variability, but rather is a
measure of sampling error. This latter source of
variation is important because it provides a measure of
the certainty for a given parameter estimate. However,
for demographic modeling, what is important is the
actual variability of parameter over time, space, and















































among individuals (collectively called process variance).
For modeling populations, sampling variation is a source
of noise and should be removed from the overall
variance estimate. Burnham et al. (1987) provided the
theoretical framework and formulae for estimating
process variance. We used this framework to estimate
process variance for nest success based on estimates
reported from 1968-1995 using only nests found after the
first egg was laid, but before hatching. Based on the data
from table 5-3, we estimated &o=0.08 and b=0.28.


INFLUENCES OF NEST SUCCESS

There are a multitude of factors that could
potentially influence the outcome of Snail Kite nests.
Factors that have been reported to significantly affect
nest success include location (i.e., area)(Snyder et al.
1989a), water levels (Sykes 1987b, Bennetts et al. 1988,
Snyder et al. 1989a, Toland 1994), date of initiation
(Bennetts et al. 1988), nest substrate (Snyder et al.


1989a, Toland 1994), nest height (Bennetts et al. 1988,
Toland 1994), distance to land (Sykes 1987c), and
interspecific coloniality (Snyder et al. 1989a). We used
logistic regression to test for the influence of each of
these effects, except interspecific coloniality, on a
sample of 854 nests using data from Bennetts et al.
(1988), Toland (1994, unpubl. data), and this study. Our
preliminary univariate analysis, which had a liberal
rejection criterion of a=0.25 (see methods) for each
effect indicated that all of these effects warranted
retention for further analysis (Table 5-4). However, our
results indicated that the specific substrate, rather than
herbaceous versus woody, was warranted for further
consideration. Similarly, our results indicated that a
categorical threshold distance to land of less than or
greater than 200m (Sykes 1987c) was warranted for
further consideration, rather than the actual distance.
Although our preliminary univariate analysis
supported the retention of these effects, a multivariate
analysis with each of the retained effects (but lacking
interaction terms) indicated that only year and date of


Table 5-2. Mean annaul nest success from major studies conducted since 1968. A successful nest is considered a nest in which at least one
young fledged. Also shown are the estimator used to estimate success, whether or not nests placed in nest baskets were included in the estimate,
and whether or not nests found in each of 3 stages were included in the estimate.
Stage Found

Annual Includes Nest
Success Range Years Location(s)' N Estimator Baskets Building Inc. Nestling Source
54% 17-85% 68-76 1,2,3,4,5,6 175' Conv" Yes' No Yes Yes Sykes (1979)
56% 17-85% 68-78 1,2,3,4,5,6 204 Convy Yes' No Yes Yes Sykes (1987b)
21% 0-41% 78-83 2,3,4,5,6,7 331 Convy No Yes Yes No Beissinger (1986)
30% 23-36% 86-87 4 367 Mayfield" No No Yes Yes Bennetts et al. (1988)
38% 30-46% 86-87 4 367 Conv' No No Yes Yes Bennetts et al. (1988)
31% 21-40% 86-87 4 358 ConWf No Yes Yes No Bennetts et al. (1988)
35% 26-44% 86-87 4 317 Convy No No Yes No Bennetts et al. (1988)'
9%' 0-30% 68-83 4,5,6,7 236 Convy Yese Yes No No Snyder et al. (1989)
29%' 0-100% 68-83 4,5,6,7 256 Convy Yese No Yes No Snyder etal. (1989)
23%' 0-100% 68-83 4,5,6,7 499 Conv' Yes' Yes Yes No Snyder et al. (1989)
29% 8-55% 90-93 8 167 Convy No No Yes No Toland (1994)
58% 53-63% 94-95 1,2,3,4,5,6,7 233 Cony' No No Yes No This study
SLocations are: WCA-1 (1), WCA-2A (2), WCA-2B (3), WCA-3A (4), Lake Okeechobee (5), Lake Kissimmee (6), Lake Tohopekaliga (7), and
Upper St. Johns Marsh (8).
SDiffers from 183 reported by source author because success was unknown for 8 of the 183 nests.
Conventional- Success estimated as (# Successful Nests/Total # Nests)
SNests placed in nest baskets were treated as all other nests.
SSuccess estimated using maximum likelihood estimator (MLE) described by (Mayfield 1961, 1975, Hensler and Mchols 1981).
SWas not reported by Bnennes e al. (1988), but data from their study were avalaible to derive estimate.
SNot reported by source authors, but estimated from annual success in Snyder et al. (1989a, Table 2). Annual success for 1978-1983 was
reported seperaely for lakes and WCA-3A by source authors, but combined here for estimate of mean annual success. All years were included
in estimate regardless ofsample size.
'AU nests placed in nest baskets were considered to have failed.


















































initiation were warranted at more restrictive rejection
criteria of a =0.05 (Table 5-5).
Our final model indicated an area, but not a year
effect, as was indicated by our preliminary analyses
(Table 5-6). However, area and year effects were highly
confounded in these data because the studies included in
this analyses that were conducted during different years
were also conducted at different areas. Thus, we do not
believe that we can reliably distinguish between these
effects. Differences in success among areas and years
are not surprising given the many causes of nest failures
(Sykes 1987c, Bennetts et al. 1988, Snyder et al. 1989a).
Our data indicated an effect from the date of
initiation in all phases of this analysis; although it was not
completely clear as to whether this effect was quadratic
or linear. Overall nest success (all years combined) was
highest during January with a generally decreasing trend


over time (Fig. 5-3). However, the overall trend is
somewhat misleading because it was heavily influenced
by one year (1987) of exceptionally high success in
January (Fig. 5-4). Most years had the peak of success
inFebruary (3 of 7) or March (2 of 7). In only one year
was peak success in January (1987) and one year in April
(1993). In only one year did we observe nesting during
December (1985), and success was lower than during
January, February or March of that year.
These temporal effects of success were
undoubtedly confounded with year effects because
studies conducted from 1991-1993 by Toland (1994,
unpubl. data), which were included in this analysis, were
conducted in the northern part of the kite's range where
the date of initiation was often several weeks later than
in the southern portion of their range. In contrast, most
of the data from other years were from the southern


Table 5-3. Annual nest success reported during studies from 1968 through 1995. Nest success was based
on nests found during the egg stage.
No. No. Nest Sampling
Year Nests Successful Success (S) Var (S)' Source
1968 1 1 1.00 0.0000 Snyder et al. (1989a)
1969 -- -
1970 1 1 1.00 0.0000 Snyder et al. (1989a)
1971 -- -- -
1972 3 1 0.33 0.0741 Snyder et al. (1989a)
1973 18 4 0.22 0.0096 Snyder et al. (1989a)
1974 13 0 0.00 0.0000 Snyder et al. (1989a)
1975 15 0 0.00 0.0000 Snyder et al. (1989a)
1976 18 0 0.00 0.0000 Snyder et al. (1989a)
1977 13 2 0.15 0.0100 Snyder et al. (1989a)
1978 59b 25 0.42 0.0041 Snyder et al. (1989a)
1979 78b 42 0.54 0.0032 Snyder et al. (1989a)
1980 2 0 0.00 0.0000 Snyder et al. (1989a)
1981 5b 0 0.00 0.0000 Snyder et al. (1989a)
1982 12b 1 0.08 0.0064 Snyder et al. (1989a)
1983 18b 5 0.28 0.0111 Snyder et al. (1989a)
1984 -- -
1985 -
1986 107 28 0.26 0.0018 Bennetts et al. (1988)
1987 210 92 0.44 0.0012 Bennetts et al. (1988)
1988 -
1989 -
1990 26 2 0.08 0.0027 Toland (1994)
1991 39 8 0.21 0.0042 Toland (1994)
1992 59 33 0.55 0.0042 Toland (1994)
1993 43 14 0.33 0.0051 Toland (1994)
1994 57 36 0.63 0.0041 This study
1995 176 94 0.53 0.0014 This study
"Sampling variance was not reported by the source authors, but was estimated based on a binomial distribution.
Annual success for 1978-1983 was reported seperately for lakes and WCA-3A by source authors, but combined herefor
estimate of annual success.










Table 5-4. Summary statistics from individual univariate logistic regression models for the factors effecting
nest success.

Source df X2' P> X21

Year 7 63.18 <0.001
Areab 3 45.90 <0.001
Nest Substrate (NSUB)' 6 29.50 <0.001
Herbaceous vs Woody Substrate 1 0.01 0.918
Date of Initiation (DOI)d 1 35.94 <0.001
DOI* DOI' 1 12.83 <0.001
Water Depth at DOI 1 19.98 <0.001
Nest Height (HOT) 1 5.18 0.023
Distance to Nearest Land (LAND) 1 0.31 0.578
Distance > 200 m (D200)' 1 2.42 0.120
" Chi Square was based on Wald Statistic (SAS Inc. 1988).
b Areas were WCA-3A, WCA2B, Upper St, Johns Marsh. Because other areas had insufficient sample sizes to be
effectively included, they were grouped into an 'other'category.
SSubstrates were Willow, Pond Apple, Meleleuca, Wax Myrtle, Cypress, Cattail, and 'other'
d Estimated Julian date offirst egg.
'A visual inspection indicated the possibility ofa quadratic, rather than linear, relationship ofDOI.
SSykes (1987b) suggested that nests within 200m of land were more prone to failure.




Table 5-5. Summary statistics from preliminary main-effects multivariate logistic regression model for the
factors effecting nest success.

Source df X2. P> I X21

Year 7 17.65 0.007
Area" 3 0.38 0.944
Nest Substrate (NSUB)' 6 8.66 0.194
Date of Initiation (DOI)d 1 7.66 0.006
DOI* DOI' 1 3.14 0.076
Water Depth at DOI 1 0.39 0.532
Nest Height (HGT) 1 1.31 0.253
Distance > 200 m (D200)' 1 2.49 0.114
a Chi Square was based on likelihood-ratio test of models between the fully saturated main-effects model with and
without each main effect (Hosmer and Lemeshow 1989).
b Areas were WCA-3A, WCA2B, Upper St, Johns Marsh. Because other areas had insufficient sample sizes to be
effectively included, they were grouped into an 'other'category.
SSubstrates were Willow, Pond Apple, Meleleuca, Wax Myrtle, Cypress, Cattail, and 'other'.
d Estimated Julian date offirst egg.
'A visual inspection indicated the possibility of a quadratic, rather than linear, relationship ofDOI.
f Sykes (1987b) suggested that nests within 200m of land were more prone to failure.


































100

o80


,0

0o0


0 N D J F M A M J J A S
Month
Figure 5-3. The percentage of nests that were successful
during each month. Data used in this analysis were from
Bennetts et a. (1988)(1986-1987), Toland (1994,
unpubLdata)(1990-1993), and this study (1994-1995).


portion of the kite's range. This probably also accounts
for the interaction effect of year with date of initiation.


Number of Young per Successful

Nest

In contrast to nest success, the number of young
per successful nest probably is one of the least variable
and has been the least controversial of the reproductive
parameters. The relative lack of variability for this
parameter is not surprising since it is not unusual for
raptors to produce normal numbers of young per
successful nest even when other aspects of reproduction
(e.g., proportion of population attempting to breed or


nest success) are depressed (Brown 1974, Steenhof
1987). For this reason, the number of young per
successful nest is not particularly informative in the
absence of these other reproductive parameters (Brown
1974).
Several studies have reported estimates for the
number of young per successful nest (Table 5-7) and the
average from 20 years of reported data is 1.9. Annual
estimates reported have ranged from a low of 1.4 (Sykes
1979, 1987b, Bennetts et al. 1988) to a high of 2.5
(Sykes 1979, 1987b)(Table 5-8).


Number of Nesting Attempts Per
Year
Snail Kites are capable of raising more than one
brood per year and attempts at multiple brooding may be
fairly widespread (Snyder et al. 1989a). However, the
extent of multiple nesting attempts has been poorly
documented. Snyder et al. (1989a) estimated the number
of attempts based on a verbal description of the following
calculation:
n
No. Nesting Attempts = No. Attempts per pair
n No. Pairs
2


where n, = the number of successful nesting attempts
found on Lake Okeechobee and WCA-3A in 1978, s =
the estimate of nest success (estimated as the number


Table 5-6. Summary statistics from the final most parsimonious (based on AIC and LRTs) logistic regression
modelfor the factors effecting nest success.

Source df f2. p> I x21

Year 7 9.80 0.200
Area' 3 10.78 0.013
Date of Initiation (DOI) 1 5.34 0.021
DOI DOl1 1 2.66 0.103
YR DOI 1 16.63 0.020
a Chi square was based on a LRT between models with and without the source term.
SAlthough the main effect ofyear was not sign ficant at a=0.05, it was retained in thefinal model because of its
significant interaction with date of initiation (DOI).
c Multivariate model indicated that this term did not need to be retained in subsequent models; however, the AIC was
lower with this term included. Year and area effects also were confounded because some years included a
restricted sample of areas (see Table 5-2).
dA LRT ofmodels with and without this term indicated that this term was not warranted; although the AIC was
slightly lower with the term included.












a0 0
I)40

0
0 N D J


W 80
dC
60



O 0


CD
a40
is




20
Z



wi
so

dl00-r
Uao



CD
so


8 4601
dl
z


ID


1986




F M A M J J A S


SN D J J A S


1991


I Is


O N D J F M A M J J A S


IILn~


0 i -- T--- -


O N D F A


1i00


1992


J A S


1993


0Ill7


SN A J J A S


1994


11[


O N D J F M A M J J A S


80 1995
o I
i6i-iil

v il


S D J F M A M
Month


Figure 5-4. The percentage of nests that were successful
during each month of each year. Data used in this analysis
were from Bennets et aL (1988)(1986-1987), Toland (1994,
unpubl data)(1991-1993), and this study (1994-1995).


J J A S


o
0 -


In-
Z
0o

(a
so



0 -


successful nests/ number nests observed) for nests found
at the nest building stage at Lake Okeechobee and WCA-
3A that were successful, and n, = the number of Snail
Kites counted during the 1977 annual count on Lake
Okeechobee and WCA-3A. Using the values reported
by Snyder et al. (1989a) produces the following estimate:

60
0.29 207
.9 2.7 attempts per pair
152 76
2


We have several concerns about this estimate
derived by Snyder et al. (1989a). First, it is important
to recognize that this estimate is only applicable to the
number of nest building attempts (i.e., courtship attempts
by our definition; see discussion of Semantics above). It
is not an estimate for the number of nesting attempts, if
a nesting attempt is defined as having produced at least
one egg. If we apply their procedure using nests in
which at least one egg was laid and using an estimate of
success derived from nests found during incubation (25
successful nests of 59 found during incubation
=0.42)(this is less biased than using nests found after
hatching; see above) then an analogous estimate for the
number of nesting attempts (in which at least one egg
was laid) would have been:
60
0.42 142
2-- 1= 1.9 attempts per pair
152 76
2


Regardless of how a nesting attempt is defined,
we believe that the approach used by Snyder et al.
(1989a) is inherently biased and produces an unreliable
estimate for this parameter. First, there are several
assumptions inherent in their calculations for which their
estimate is not robust to violation. They correctly
pointed out that they assumed (1) the count from 1977
was an accurate census, (2) there was a 1:1 sex ratio, (3)
all birds counted during 1977 were potential breeders in
1978, and (4) no birds died between the 1977 count and
the end of the 1978 breeding season.
As we discuss in considerable detail in a later
chapter (see discussion of the Annual Count in Chapter
on Monitoring the Florida Snail Kite Population), the
annual count is not a reliable census of the population, as
was assumed, and evidence suggests that it is not even a
reliable index to the population. Given that it is highly
improbable that all birds are counted during the annual
count (Rodgers et al. 1988), it is likely that the number



































of pairs would have been underestimated, which would
have resulted in the number of attempts per pair to have
been overestimated. We also have no means of
evaluating whether assuming a 1:1 sex ratio was
reasonable, but in the absence of such information we
agree with Snyder et al. (1989a) that this was a
reasonable assumption.
We disagree that all birds counted during the
1977 annual count can be reasonably assumed to be
breeders in 1978. Our data, as well as the observations
by Snyder et al. (1989a) indicate that juveniles have a
lower probability of attempting to breed than do adults.
Thus, all birds included in the 1977 count that were
juveniles were not potential breeders. This would tend
to inflate the denominator (i.e., to reduce the number of
pairs in their calculation) and, consequently produce a
negative bias; however, there is no reliable way to
estimate the extent of the bias, since the proportion of
juveniles in the 1977 count was unknown.
We agree with Snyder et al. (1989a) that they
had to assume that no birds died between the 1977 annual
count and the end of the 1978 breeding season; however,
the true assumption was substantially more extensive
than they reported. Because their calculation assumed
the same population at both time periods (i.e., during the
count and during the breeding season) they really
assumed that Lake Okeechobee and WCA-3A (the only
areas used in their calculation) were a closed population.
That is, that there were not only no deaths, but also that


there were no births, no immigration, and no emigration.
If the time period between the 1977 annual count and the
end of the 1978 breeding season were very short, this
would be a reasonable assumption. However, the count
was conducted in December of 1977 (see the Annual
Count in Chapter on Monitoring the Florida Snail Kite
Population) and initiation of the last reported nest of the
1978 breeding season was in August of 1978 (Snyder et
al. 1989a). Thus, the interval over which this
assumption applied was 8-9 months. We are not
especially concerned about the assumption of no births
because this assumption can be met by excluding young-
of-the-year juveniles from their analysis. Although it is
unlikely that no deaths occurred over this period of time,
we would also not expected a large bias from this effect
due to high adult survival. In contrast to births and
deaths, we find the assumption of no immigration or
emigration of substantial concern. Our data suggests that
the probability of an adult bird moving in any given
month is approximately 0.25 (see Movement). Thus,
over an 8-9 month period it is extremely unlikely that
there was no immigration or emigration to or from these
areas. Furthermore, our data suggest that at the time the
annual count is conducted, there is a greater probability
that birds will be in habitats not typically used for
nesting. This could result in a substantial overestimate
of the number of attempts per pair.
Snyder et al. (1989a) suggested that a bi-modal
seasonal distribution of nests that are spaced about 34


Table 5-7. The annual mean and overall (i.e., all years and locations combined) number young per successful
nest from major studies conducted since 1968. A successful nest was considered a nest in which at least one
young fledged.
Annual x Overall
No. Per No. Per Total No.
Successful Successful Annual Years Successful
Nest' Nest2 Range Included Location(s)' Nests Source
2.0 1.9 1.4 2.5 1968-1976 1, 2, 3, 4, 5, 6 84 Sykes (1979)
2.0 2.0 1.4-2.5 1968-1978 1, 2, 3, 4, 5, 6 103 Sykes (1987b)
2.0 -* 1978-1983 4,5,6,7 106 Beissinger (1986)
1.5 1.6 1.4 1.7 1986-1987 4 149 Bennetts et al. (1988)
2.0 --' 1968-1983 4,5,6,7 --4 Snyder et al. (1989)
1.9 2.0 1.5-2.1 1990-1993 8 57 Toland (1994)
1.8 1.8 1.6- 1.9 1994-1995 1,2,3.4,5,6,7,8 144 This study
The annual average number ofyoung per successful nest
SThe total (all years) number ofyoung per total number of successful nests
'Locations are: WCA-1 (1), WCA-2A (2), WCA-2B (3), WCA-3A (4), Lake Okeechobee (5), Lake Kissimmee (6), Lake
Tohopekaliga (7), Upper St. Johns Marsh (8),
SNot reported and/or insufficient information provided to estimate









Table 5-8. The number of successful nests, youngfledged, and number of young per successful nest
reported for each year from 1968 through 1995.

No. Successful No. No. Young per
Nests Young Fledged Successful Nest Source

1968 11 24 2.2 Sykes (1979, 1987b)
1969 8 13 1.6 Sykes (1979, 1987b)
1970 8 12 1.5 Sykes (1979, 1987b)
1971 0 0 --' Sykes (1979, 1987b)
1972 3 7 2.3 Sykes (1979, 1987b)
1973 12 29 2.4 Sykes (1979, 1987b)
1974 6 11 1.8 Sykes (1979, 1987b)
1975 14 35 2.5 Sykes (1979, 1987b)
1976 22 30 1.4 Sykes (1979, 1987b)
1977 8 20 2.5 Sykes (1987b)
1978 11 20 1.8 Sykes (1987b)
1979 54 1082 2.0 Beissinger (1986)
1980 _- -3 3
1981 0 0 -' Beissinger (1986)
1982 2 42 2.0 Beissinger (1986)
1983 10 202 2.0 Beissinger (1986)
1984 _3 -3 .3 -3
1985 -3 3
1986 45 65 1.4 Bennetts et al. (1988)
1987 104 172 1.7 Bennetts et al. (1988)
1988 _3 _-
1989 3 _3 -3
1990 2 3 1.5 Toland (1994)
1991 8 17 2.1 Toland (1994)
1992 33 68 2.1 Toland (1994)
1993 14 26 1.9 Toland (1994)
1994 50 80 1.6 This study
1995 94 181 1.9 This study
' No successful nests from which to estimate
2 Not reported, but inferred from number of successful nests and number young per successful nest
' No reported information


months apart during some years indicated widespread
multiple brooding. We agree that a bi-modal distribution
of nest initiations may be an indication of multiple
brooding during those years. However, a bi-modal
distribution may also be attributable in some years to the
starting and stopping of nest initiations due to early
season temperature changes; although these two
explanations are not mutually exclusive. We often


observed nest initiations during December or January
that coincided with periods of unseasonably warm
weather, particularly in the southern habitats. As colder
temperatures resumed nest initiations that had not
reached the egg laying stage were likely to terminate.
Initiations would resume when warm temperatures
returned, often a month or two later. This would create
a peak of initiations early during the season (usually in









January) followed by a second larger peak one or two
months later (usually in February or March) after warm
temperatures became prevalent. However, birds that
successfully initiated early were also likely to re-nest
because completion or failure of their first nest occurred
well within the primary breeding season.
Snyder et al. (1989a) correctly limited their
inference to 1978. Beissinger (1995) later extended this
inference to a general estimate for the number of
attempts per year; although he revised the estimate to
from 2.7 to 2.2 suggesting that this would be
conservative. In order to generalize this estimate to all
years, Beissinger (1995) had to assume that 1978 was a
"representative" year. Based on the data presented by
Snyder et al. (1989a), 1978 was far from representative.
Based on both number of nests found and on number of
nestlings banded, 1978 was an extremely high year for
reproduction (Fig. 5-5). Consequently, application of
this estimate to other years would very likely have
resulted in an inflated estimate.


388888E.&,~.



aso11111


8ee 7 88 68 70 71 72 73 74 75 76 77 78 79 80 81 82 83
Year











66 67 88 68 70 71 7 723 74 75 76 77 78 79 80 81 62 83
Year


Figure 5-5. he number ofnests (top) and number of
nestlings banded (bottom) during each year reported by
Snyder et al. (1989a). Snyder et a. (1989a) estimated the
number of breeding attempts per pair during 1978 and
correctly limited their inference of this estimate to 1978.
Based on these data, we suggest that estimates from 1978 can
not be reliably extended to other years.


Z
z



I



Z






E


z
-o


1 -




,0 -

a Ii


In contrast to the approach used by Snyder et al.
(1989a), we estimated the extent of multiple broods using
radio-transmittered birds. During 1995, we closely
monitored 23 radio-transmittered adult Snail Kites for
breeding activity in order to assess the number of
breeding attempts per year. Of these 23 adults, 14 were
females and 9 were males. During the 1995 nesting
season, we located each bird on the ground
approximately bi-weekly (x = 14.1 days 8.1 sd) and
determined its breeding status (e.g., a nest, courtship,
not breeding). Birds in which no breeding activity was
detected were generally observed for 22 hrs and
subsequent visits, usually within 10 days, were required
to confirm a non-breeding status and to confirm any
nests for birds exhibiting courtship. We found an
average of 1.4 (0.6 sd) nesting attempts per bird
(Table 5-9). The interval of our breeding status checks
could have resulted in a failure to detect an occasional
bird that initiated a nest that failed early during laying or
incubation. Consequently, this estimate may be slightly
low for 1995. However, most nest failure occurs during
the first week after hatching (Bennetts et al. 1988),
which would have required nesting activity for at least 4-
5 weeks. Consequently, the potential bias from having
missed nests over a 14-day period probably was
negligible. In addition, 1995 had favorable water
conditions throughout the Snail Kite's range in Florida.
Consequently, we might also expect our 1995 estimate to
be higher than an annual average.
In summary, we agree with Snyder et al.
(1989a) that multiple brooding by Snail Kites in Florida
is common during some years. However, our data
suggest that the estimate of 2.7 attempts per year by
Snyder et al (1989a) and even the "more conservative"
estimate of 2.2 attempts per year used by Beissinger
(1995) were substantial overestimates. A combination of
differences in our estimation procedures, difference in
our respective definitions of a breeding attempt, and
annual variability of this parameter probably account for
these discrepancies.

Conditional Probability of Attempting to Breed-
An alternative way to look at the number of attempts per
year is using conditional probability for the proportion of
birds attempting to breed (c,). That is, a" would be the
probability that a bird attempted to breed, given that it
had not attempted previously during that nesting season.
Of 23 birds we monitored for breeding activity during
1995, all 23 attempted to breed at least once. Thus, our
estimate of aC would be 1.0. Similarly, a2 would be the
probability that a bird attempted to breed, given that it
had previously made 1 attempt during that breeding
season. Based on our data from 1995, we would


RE P- -- -P-- IF.


W VH W










Table 5-9. Number of nesting attempts and number
ofattempts that were successful for each of 23 adult
Snail Kites during the 1995 breeding season.
y S Number of No. Attempts
Frequency Sex
Attempts Successful
152.698 F 1 1
152.584 F 1 0
153.496 F 2 0
153.860 F 2 1
152.739 F 3 1
153.931 F 1 1
152.039 F 2 1
153.969 F 2 2
152.494 F 1 1
152.169 F 2 1
153.979 F 2 1
152.777 F 1 1
152.499 F 1 0
152.369 F 1 1
152.869 M 1 1
153.900 M 1 1
152.128 M 1 1
153.390 M 1 1
153.290 M 1 1
152.848 M 1 0
152.858 M 1 1
152.539 M 1 0
152.379 M 2 1


estimate a2 to be 0.34 (8 of 23). The probability that a
bird attempted to breed, given that it had previously
made two attempts during that breeding season (a3) was
0.13 (1 of 8). The variance for these estimates could be
derived based on a binomial distribution, although the
formula traditionally used for this estimate:
&,(1 a,)
Var(&,) = --
ni



is intended for large samples (White and Garrott 1990).
Hollander and Wolfe (1973) provide alternative
procedures that could be used for smaller samples.


Number of Successful Broods per Year- Snyder
et al. (1989a) suggested that in some years it was
possible for Snail Kites to successfully raise four broods.
This was based on the length of the breeding season for
certain years (e.g., 1978 and 1979) and the assumption
that it would take 10 weeks (70 d) to raise a brood if
mate desertion occurred and 16 weeks (112 d) if no mate
desertion occurred. Although this is certainly
theoretically possible, we believe that the probability of
a Snail Kite successfully raising even three broods in a
given year is very close to zero. We base our conclusion
on several points. First, empirical data do not support
Snyder et al.'s (1989a) conclusion. There have been no
documented cases of Snail Kites successfully raising >
2 broods in a given year, and the occurrence of
successfully raising 2 broods appears quite rare. Out of
an 18-year study including 666 nesting attempts, Snyder
et al. (1989a) documented only 3 cases of Snail Kites
successfully raising 2 broods. Similarly, only 1 of 23
(4%) radio-transmittered birds that we closely monitored
for breeding activity during 1995 (a good year),
successfully raised two broods. We believe that Snyder
et al. (1989a) overlooked some critical aspects of the
breeding biology of Snail Kites when making this
suggestion. First, although the inclusive dates from the
first nest initiated in a given year to the last may span a
period of 6-7 months, the initiation of nests is not evenly
distributed throughout that period (see The Breeding
Season above). The majority of nests (82% of the nests
reported by Snyder et al. [1989a]) were initiated during
a five month period from January through May. This is
sufficient time for only two successful broods even when
mate desertion occurs. The longest nesting season (time
span over which nest initiations were observed) reported
by Snyder et al. (1989a) over an 18-year period was only
31.7 weeks. Given that they suggest that it takes 10-16
weeks per successful brood, the longest nesting season
they observed did not even have sufficient time to
successfully raise four broods (even with mate desertion
for all broods), and barely had sufficient time to
successfully raise two broods without mate desertion.
There also has been no consideration given to energetic
costs of raising successive broods. Thus, we believe that
a small percentage of birds (e.g., < 10%) may
successfully raise two broods during some years;
however, there is currently no empirical evidence to
support the conclusion that Snail Kites successfully raise
> 2 broods per year.





































Chapter 6. MOVEMENTS



Natal Dispersal of Juveniles

Natal dispersal s usually defined as the pr-
manent movement ol an animal (usually a juvenile)
away froman animals natal site to a new ste actual
or potential breeding (.g Howard 1960, Greenwood


dlspelsal ofa uvene from ts natal wetland wI h tie
undertandg that a given indivdual mght, and prob-
ably wdl, return to it natl wetland many times dur-
ing its 1lfetme.
The overall cumulative probabilityof uvenle
Snail Kites dispeing from their natal wetld dunng
their firt ye- was 0 81 based on an estmae derived
using a KaplanMeeresimator (Fig 6-) Ony 8 of
65 (12%) radlo-transmitterd birds ovel a the-ye
period that survived their enlire hist year and whoe
locations were known remained in their nata welland
for their entire irst year


64









TEMPORAL PATTERNS OF NATAL DISPERSAL

Within-year patterns- Of the birds that
dispersed during their first year (n=57), most (60%) did
so within the first 60 days after fledging and all did so
within the first 240 days (Fig. 6-2).





-o
i 40-








90 3 0 120 150 180 210 24 270 0 3
0






Days Since Fledging

Figure 6-2. The percentage ofjuvenile Snail Kites that
initially dispersed from their natal wetland in each 30-day
time period since fledging.


Differences among Years-Dispersal of juveniles
from their natal wetland was lowest during 1992 and
relatively higher in both 1993 and 1994 (Fig. 6-3).
Differences were significant (at a =0.05) between 1992
and each of the other years, but not between 1993 and
1994 (Table 6-1)(Appendix 6-1).


CD ------------


3.0"3
IE-

o 0.4 -J'



O APR MAY JUN JUL AUG SEP OCT NOV DEC JAN FEB MAR APR
Date

Figure 6-3. Kaplan Meier estimatesfor the cumulative
probability of dispersal in each of the three years.
Confidence intervalsfor estimates are not shown to minimize
cluttering, but are provided in detail in Appendix 6-1.


Table 6-1. Results of log-rank tests between
dispersal functions of juvenile Snail Kites
dispersing from their natal wetlands during
each study year (SY)(April 15 April 14).

Comparison X2 df Prob

1992 vs. 1993 5.246 1 0.022

1992 vs. 1994 4.049 1 0.044

1993 vs. 1994 0.129 1 0.720




DIFFERENCES IN NATAL DISPERSAL BETWEEN
NORTHERN AND SOUTHERN REGIONS

We did not have a sufficient sample to compare
differences in natal dispersal of juveniles among specific
wetlands and/or among specific regions. However,
because low waters levels occurred throughout the
southern portion of our study area just prior to our study,
we were interested in a spatial comparison of natal
dispersal. Consequently, we pooled our samples from
the southern regions (i.e., Everglades, Loxahatchee
Slough, and Lake Okeechobee), which were substantially
influenced by the previous drought, and the northern
regions (i.e., the Kissimmee Chain-of-Lakes, and Upper
St. Johns River Basin), which were relatively unaffected
by the previous drought. This analysis indicated that
natal dispersal from wetlands in the southern regions
(i.e., most affected by the previous drought) was
substantially lower than from wetlands in the northern
regions during 1992, but not during 1993 or 1994 (Table
6-2)(Fig. 6-4)(Appendix 6-2).



Table 6-2. Results of log-rank tests comparing
dispersal functions ofjuvenile Snail Kites
dispersing from the northern regions (i.e.,
Kissimme Chain of Lakes and Upper St. Johns
River Basin) and southern regions (i.e.,
Everglades, Lake Okeechobee, and
Loxahatchee Slough) during each study year
(SY)(April 15 April 14).

Comparison X2 df Prob

1992 7.530 1 0.006

1993 0.635 1 0.426

1994 <0.001 1 0.994
































DIFFERENCES IN NATAL DISPERSAL BETWEEN
LAKE AND MARSH HABITATS

Our sample was also insufficient to compare
the differences in natal dispersal among each individual
habitat; but was sufficiently large to compare between
lakes and marshes. We did not observe any differences
in dispersal out of lakes compared to marshes for any
juvenile cohort at a = 0.05; however differences were
significant during 1993 at a = 0.10 (Table 6-3).
Additionally, the number of animals in each sample
group (i.e., birds that were alive and whose locations
were known for each habitat type during each year) at
any given time was relatively low (! 15)(Appendix 6-3).
Thus, the power to detect even substantial differences in
dispersal also was relatively low (see Fig. 3-1 in Methods
section). There was, however, a consistent pattern for
the estimates of the cumulative probability of dispersal
out of marshes to be higher than from lakes for each
cohort (Fig. 6-5).

DISCUSSION OF NATAL DISPERSAL

High rates of natal dispersal are often found for
species that exhibit nomadic tendencies or that inhabit
fluctuating or unpredictable environments (Baker 1978,
Greenwood and Harvey 1982). Thus, the overall high
rates of dispersal we found are not surprising.
Two of the more commonly suggested reasons for high
rates of natal dispersal are to gain access to resources
(e.g., food or mates) in a saturated environment and
inbreeding avoidance. The latter will be addressed in a


Table 6-3. Results of log-rank tests comparing
dispersal functions ofjuvenile Snail Kites
dispersing from lake and marsh habitats
during each study year (April 15 April 14).
Comparison X2 df Prob
1992 0.599 1 0.439
1993 3.071 1 0.080'

1994 0.717 1 0.397
SUsing the alternative tests described by Cox and
Oakes (1984)(Appendic 3-3) that are slightly less
conservative (i e., have greater power, but higher
risk of Type I error) we estimated '=3.801,
P=0.051 and j=3.653, P=0.056for
alternative tests I and 2, respectively.



later section (see Natal Philopatry and Site Fidelity).
Most species studied in which gaining access to resources
has been suggested as a reason for dispersal have been
territorial species (reviewed by Greenwood and Harvey
1982). In contrast, Snail Kites defend only a few meters
around their nest site and only rarely defend feeding
areas (Sykes et al. 1995). Thus, access to resources is
unlikely to be limited by social behavior (except mate
choice), and is more likely due to the overall availability
of food or mates.
Given the lack of social exclusion from
resources, it might be predicted that if high dispersal
were attributable to gaining increased access to































resources, then dispersal should be high when local
resources are depressed. Our results were not consistent
with higher rates of dispersal during periods of low
resources availability. Unfortunately, we did not initiate
the collection of foraging data until 1993; however,
based on hundreds of hours of field observations of
birds, its was very apparent that food resources were
substantially lower in 1992 (the first year of our study
following the previous drought) compared to later years.
Our foraging observations (during 1993 and 1994) as
well as numerous previous reports (reviewed by Sykes et
al. 1995) indicated that the capture of snails often
requires < 3 minutes. In contrast, we often observed
birds during 1992 that would forage in excess of 30
minutes without capturing a snail. In a few instances
during our trapping period, we observed several foraging
birds for hours without observing a single snail having
been captured. Thus we would have predicted that
dispersal would have been highest during 1992.
Contrary to our prediction, dispersal was lowest during
1992. Furthermore, during 1992 dispersal was
substantially lower in the southern regions where food
was more depressed.

Movement Probabilities

In addition to the initial dispersal of juveniles
from their natal wetland we also examined general
movements of adult and juvenile Snail Kites. Here we
treat movement using a conditional logistic regression
model based on one month time intervals (see methods).
Thus, our model expresses the conditional probability


that, given a bird was alive and its location known at
time t, that it would be in the same location (or
conversely at a different location) at time t + 1. We
then explored several potential effects on this probability.
We based our analysis of general movements on 5,299
locations of radio transmittered birds (3,618 locations of
adults and 1,681 of juveniles).

THE EFFECT OF AGE AND SEX ON MOVEMENT
PROBABILITIES

A conditional logistic model of movement
probability (i.e., the probability of a bird moving
between time t and time t +1, given that it was alive and
its location known at t+ 1) indicated effects of both age
(X2=5.38, df=1, P=0.020) and sex (X2=6.16, df=2,
P=0.046) when analyzed separately. However, our
model of sex was confounded with age because sex was
categorized as male, female, or unknown; where only
juveniles were unknown. If these data were constrained
to only adults, we found no evidence that sex had an
influence on movement probability (X2=0.76, df=1,
P=0.384). Because the unconstrained model with age
(i.e., both age classes included) is a subset of the more
general model of sex (with 1 fewer parameter), an
alternative approach to test for the effects of sex is a
LRT. The LRT between the unconstrained univariate
models also indicated that sex does not add significantly
to the fit of these data over a model with only age (LRT
=0.76, 1 df, P=0.617). A comparison of AIC between
these two models adds additional support for this
conclusion (the model with age only had a lower
AICXTable 6-4). Thus, we concluded that age, but not


g 1. .....

o 0.8... -
--- --. -
= 0.6 J
S .. -.o
-o
a 0.4
S_.... .- i

n 0.2
jB-------------- mar&i
E 0
3 A' M J J A S O N D J F M A M J J A S O N D J F M A M JJ A S O NO D JJ F M A
1992 1993 1994 1995
Date

Figure 6-5. Kaplan Meier estimates for the cumulative probability of dispersal from lake and marsh habitats in each ofthe three
study years. Confidence intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 6-3.









Table 6-4. Summary statistics for conditional
logistic regression model for the factors affecting the
probability of movement between times t and t + 1
(at monthly time steps), given that an animal was
alive at time t and its location known. Shown are
the model description, number of estimable
parameters (np), relative deviance (-2bln[l), and
Akaike 's Information Criterion (AIC). The model
shown in bold would be the one selected from this
group based on AIC.

Model np -21n(S) AIC

Age 2 2502.95 2506.95

Sex 3 2502.19 2508.19




sex, had an influence on movement probabilities. Our
estimates indicated that adults had a higher overall
probability of movement (excluding any additional
factors) than juveniles (Fig. 6-6).


Age Class

Figure 6-6. Conditional probabilities that adult andjuvenile
Snail Kites that were alive and their location known at time t,
were in the same location (or conversely at a different
location) at time t + 1. Also shown are the standard errors
(rectangles) and 95% confidence intervals (vertical lines).



TEMPORAL EFFECTS ON MOVEMENT
PROBABILITIES

We began testing for an overall time
effect using a univariate model based on separate
parameter estimates for each month of each year of our
study (i.e., 12 months for each of 3 years =36
parameters). This test showed a strong effect of time
(X2=90.79, df=36, P<0.001)(Fig. 6-7). We then


16
E -
0

9.





04



16
0.4

a2-
0
91

50"
05
.0 .4
6)
.0


AMJJASONDJFMAMJJASONDJFMAMJJASONDJFMA
192 1993 199M 1995
Month/Year


A


AMJJASONDJFMAMJJASONDJFMAMJJASONDJFMA
1992 1903 1994 1995
Month/Year


Figure 6-7. Conditional probabilities that adult and juvenile
Snail Kites that were alive and their location known at time t,
were in the same location (or conversely at a different
location) at time t + Ifor each month during this study (solid
lines). Also shown are the 95% confidence intervals (dotted
nnes).


tested whether this effect could be accounted for with a
more parsimonious model using separate months, but not
for each year (i.e., 1 parameter for each month [12]
rather than 36 for the previous model). This model also
showed a significant monthly effect (x2=32.36, df= 11,
P<0.001), but a LRT indicated that the more general
model (with 36 parameters) was warranted
(LRT=73.81, 24 df, P< 0.001). We next explored a
series of models in which time was expressed as a
seasonal, rather than monthly effect. These models
reflected various combinations of 3 and 4 seasons in a
sliding window approach (i.e., each iteration shifted the
months included by one month) to determine how the
months should be divided into seasons. This analysis
indicated that one of the 3-season models (of 4
months/per season) was the most parsimonious based on
AIC (Table 6-5); however, 2 additional models (the
model with a separate parameter for each month and one
of the 4-season models) had similarly low AIC values
and could not be rejected based solely on AIC criteria.
Next, we compared a suite of models including


+


I I


i~


\q

































combinations of the above effects including their
interaction terms. This analysis indicated that a model
with season (the 3-season model selected from the above
analysis), year, and the interaction between season and
year was the most parsimonious based on AIC (Table 6-
6). As above, two alternative models (individual month
model with 36 parameters and the model with season and

Table 6-6. Summary statistics for conditional
logistic regression models for potential temporal
effects on the probability of movement between times
t and t + I (at monthly time steps), given that an
animal was alive at time t and its location known.
Shown are the model description, number of
estimable parameters (np), relative deviance (-
21ln[f), and Akaike's Information Criteria (AIC).
The model shown in bold would be the one selected
from these potential models based on AIC criteria.
Model np -21n(-) AIC
Time 36 2400.6 2472.6
Month 12 2474.4 2498.4
Season 3 2490.6 2496.6
Year 3 2477.4 2483.6
Month Year 14 2446.6 2474.6
Month Year Month*Year 36 2402.7 2474.7
Season Year 5 2460.9 2470.9
Season Year Season*Year 9 2451.3 2469.3


year without an interaction term) had similarly low AIC
values and could not be rejected based solely on AIC.
Likelihood ratio tests also indicated that the model with
the lowest AIC was preferred (X2 =50.72, df=27,
P=0.003 and X2 =9.55, df=4, P=0.049 for each of the
alternative models, respectively). Thus, our data support
that movement probabilities are influenced by season
(Fig. 6-8), year (Fig. 6-9), and an interaction between
season and year (Fig. 6-10).

SPATIAL EFFECTS ON THE PROBABILITY OF
MOVEMENT

Here we explored whether the location of a
given bird influenced whether or not the bird moved
between times t and t + 1. For this analysis, we were
not concerned with the destination of the bird (that will
be explored latter in the section on Spatial Patterns of
Movement); only its location at the time that a movement
did or did not occur. We began this analysis with the
general null hypothesis that the specific wetland where a
bird was located at time t did not influence the
probability of whether or not it moved to a different
location at time t + 1. We rejected this null hypothesis
based on a conditional logistic regression model (x2
=107.38, 16 df, P <0.001)(Fig. 6-11). We then tested
the same hypothesis using the region, rather than the
specific wetland, to determine if this might provide a
more parsimonious model. This test also rejected the
null hypothesis (x2 =53.82, 5 df, P <0.001); however,
a comparison of these two models indicated that our data


Table 6-5. Summary statistics for conditional logistic regression models for potential seasonal groupings affecting
the probability of movement between times t and t + 1 (at monthly time steps), given that an animal was alive at
time t and its location known. Shown are the model description, number of estimable parameters (np), relative
deviance (-21n[-]), and Akaike's Information Criteria (AIC). The model shown in bold would be the one selected
from these potential models based on AIC.

Season Model np -2ln(S) AIC

(JAN FEB MAR APR MAYJUN JUL AUG SEP OCTNOVDEC) 12 2474.44 2498.44
(JAN FEB MAR APR) (MAY JUN JUL AUG) (SEP OCT NOV DEC) 3 2490.66 2496.66
(FEB MAR APR MAY) (JUNJULAUGSEP) (OCT NOVDECJAN) 3 2498.75 2504.75
(MAR APR MAYJUN) (JULAUG SEP OCT) (NOVDEC JAN FEB) 3 2506.24 2512.24
(APR MAYJUNJUL) (AUG SEP OCTNOV) (DECJANFEB MAR) 3 2498.14 2504.14
(JAN FEB MAR) (APR MAY JUN) (JUL AUG SEP) (OCT NOVDEC) 4 2491.16 2499.16
(FEB MAR APR) (MAYJUN JUL) (AUG SEP OCT) (NOV DEC JAN) 4 2497.98 2505.98
(MAR APR MAY) (JUN JUL AUG) (SEP OCT NOV) (DEC JAN FEB) 4 2497.36 2505.36






















SPRING SUMMER FALLWINTER
(JAn-lp) (May-Au) (Sp-Dc)
Season

035

0.30 -

025




0.10

0.05
SPRING SUMMER FALLWINTER
(Jn-Ap) (My-Aug) (Sp-D-c)
Season

Figure 6-8. Conditionalprobabilities that adult andjuvenile
Snail Kites that were alive and their location known at time t,
were in the same location (or conversely at a different
location) at time t + I during each season. Also shown are
the standard errors (rectangles) and 95% confidence
intervals (vertical lines).



supported use of the more general model (i.e., specific
wetlands) based on both a LRT (X2 =52.04, 11 df, P
<0.001), and on AIC (Table 6-7).





Table 6-7. Summary statistics for conditional
logistic regression models of the probability of
movement between times t and t + 1. The model
with the lowest AIC (bold) would be selected if
based solely on this criterion.

Source -21n(g) np AIC

Specific Wetland 2394.12 17 2428.12


Region 2446.16 6 2458.16


Ad 1


0.35
Adib

0.30

2 0.25

0.20
g | 4-i-


2
0.15

0.10

Study Year

0.35

0 -


5 0.20
0 215 -


0.05 -
0.00
19S2 1993 1994
Study Year

Figure 6-9. Conditional probabilities that adult andjuvenile
Snail Kites that were alive and their location known at time t,
were in the same location (or conversely at a different
location) at time t + 1 during each study year. Also shown
are the standard errors (rectangles) and 95% confidence
intervals (vertical lines).



Pooling of Locations- We next explored
whether we could improve our model by some limited
selective pooling based on a combination of biological
and statistical criteria. Our goal for this exploration was
to determine if we could obtain a more parsimonious
model by pooling areas in which the overall relative use
and seasonal patterns of use were similar enough so as
not to warrant separate parameter estimates. We did not
attempt to pool some wetlands whose use patterns we felt
were biologically different (e.g., wetlands that were used
primarily during non-breeding with wetlands used
primarily for breeding) even though we could have done
so strictly based on statistical criteria. Thus, although a
more parsimonious model for the effects of location on
movement probability was possible, we preferred to
maintain separate parameter estimates for some areas to
better ensure the biological integrity of these models.
We began our exploration of potential pooling
with areas in the Southern Everglades. The first pooling
we considered was Everglades National Park (ENP) and
Northeast Shark River Slough (NESRS). Each of these






















SU FA SP SU FA SP 8U FA SP
S192 1993 1994
Season/Study Year


SU FA SP SU FA SP SU FA SP
1992 1983 1994
Season/Study Year


Figure 6-10. Conditionalprobabilities that adult and juvenile
Snail Kites that were alive and their location known at time 1,
were in the same location (or conversely at a different
location) at time t + 1 during each season of each study
year. Also shown are the standard errors (rectangles) and
95% confidence intervals (vertical lines).



10



I


a


-10
w04A1 9CAS 303B BICY 43 4 EE TO4 O 3 C PEW

Location

Figure 6-11. Adjusted residuals from a crosstabulation of
movement and location at time t. Residuals >0 indicate that
birds in this area moved more frequently than expected and
residuals <0 indicates that birds in that area moved less
frequently than expected.


I


2 030-
3 o.mo

0.20

0.10 -

0.00
o


a.o


these areas are administered by the National Park
Service, are part of the Shark River Slough (ENP has
areas not within the Shark River Slough, but these areas
were not used by radio-transmittered kites during our
study), receive low to moderate kite use, and are not
impounded at their outflow (each have levees at their
inflow). A statistical comparison indicated that separate
parameter estimates for each of these areas was not
warranted based on LRTs and AIC (Table 6-8). Next,
we considered including WCA-3B with ENP and
NESRS. WCA-3B is also within the Shark River
Slough, but is impounded at its outflow and is
administered by state agencies. However, its overall
relative use and its seasonal patterns of use (each of
these areas tended to be used most during early summer)
were quite similar to ENP and NESRS. A statistical
comparison indicated that separate parameter estimates
for each of these areas also was not warranted based on
LRTs and AIC (Table 6-8). We considered pooling Big
Cypress National Preserve (BICY) with ENP, NESRS,
and WCA3B; however, BICY was used more
extensively during fall and early winter than these other
areas and much of areas used in BICY consisted of
cypress prairie habitat which was not generally available
in these other areas. Consequently, we did not include
BICY with these other areas, even though we probably
could have justified doing so on a statistical basis.
Next, we considered pooling the A.R.M.
Loxahatchee National Wildlife Refuge (WCA-1), Water
Conservation Area 2A (WCA2A), and Holey Land
Wildlife Management Area (HOLEY). Each of these
areas represents northern Everglades habitats, although
their water management histories have differed. Our
statistical comparison indicated that separate parameter
estimates for each of these areas was not warranted
based on LRTs and AIC (Table 6-8).
We next considered areas within the Kissimmee
Chain-of-Lakes. First we considered pooling Lakes
Tohopekaliga (TOHO) and East Lake Tohopekaliga
(ETOHO). Our statistical comparison indicated that
separate parameter estimates for each of these areas was
not warranted based on LRTs and AIC (Table 6-8).
Next we considered including Lake Kissimmee (KISS)
with TOHO and ETOHO. Lake Kissimmee received
moderately heavy use compared to the other lakes, but
all are in close proximity, the seasonal patterns of use
were similar, and there was considerable interchange
among these lakes. Our statistical comparison indicated
that separate parameter estimates for each of these areas
again was not warranted based on LRTs and AIC (Table
6-8). In contrast, other lakes within the Kissimmee-
Chain-of-Lakes (e.g., Lakes Marion, Tiger, Walk-in-the
Water, and Marian) received substantially less use than











































KISS, TOHO, and ETOHO and the seasonal pattern of
use was quite different (i.e., they were used most
frequently during non-breeding periods). The seasonal
pattern of use of these smaller lakes more closely
resembled that of the peripheral habitats. Consequently,
we next considered pooling the smaller lakes of the
Kissimmee-Chain-of-Lakes with the peripheral habitats.
Our statistical comparison indicated that separate
parameter estimates for each of these areas were not
warranted based on LRTs and AIC (Table 6-8).
Finally, we compared various combinations of
pooling to the general unconstrained model and among
each other. This analysis indicted that the model
containing all of our proposed pooling was a substantial
improvement over the unconstrained model (i.e., with no
pooling). Additionally, the model with all of our
proposed pooling had the lowest AIC; although overall
differences among all of the models we compared were
relatively small. Consequently, we used the most
parsimonious model (Model 16) of this set of models in
further analyses of the influence of location on
movement probability. This model had 10 parameters


compared to the unconstrained model with 17
parameters; but was still a substantial improvement (x2
=47.61, 7 df, P <0.001) over the 6-parameter model
using regions, rather than location.

HYDROLOGIC EFFECTS ON THE PROBABILITY
OF MOVEMENT

We tested the influence of water levels on the
probability of movement for a subset of the data (1787 of
2994 months). We were unable to use the complete data
because applicable hydrologic data were not readily
available for all areas. Our subset, however, included
most of the major wetlands used by kites. Probably the
most notable exception was the Upper St. Johns marsh.
Data that were applicable to the areas kites used most
during this study (i.e., the Blue Cypress Water
Management Area and Blue Cypress Marsh Water
Conservation Area) were available only from 1991 to
present. Because our analysis used long-term averages
to assess relative water levels (see below), we did not
believe that these data were sufficient for this


Table 6-8. Summary statistics for conditional logistic regression models of the probability of movement between
times t and t + I to evaluate the pooling of some parameters. A failure to reject a LRT indicates that the
additional parameters of the more general (unconstrained) model may not be supported by these data. The model
with the lowest AIC (bold) would be selected if based solely on AIC.
Model No. Constraints (Pooling) LRT' (2) df P > X2 -21n(i) np AIC
1 Unconstrained -- 2394.12 17 2428.12
2 NESRS = ENP 2.78 1 0.10 2396.90 16 2428.90
3 NESRS = ENP=WCA-3B 2.80 2 0.25 2396.93 15 2426.93
4 HOLEY = WCA1 0.07 1 0.79 2394.20 16 2426.20
5 HOLEY = WCA1=WCA2A 0.14 2 0.93 2394.26 15 2424.26
6 TOHO= ETOHO 1.46 1 0.23 2395.59 16 2427.59
7 TOHO= ETOHO=KISS 1.48 2 0.48 2395.61 15 2425.61
8 PERIPHERAL = KISSCH 0.00 1 0.99 2394.12 16 2426.12
9 Reduced Model (2,4,6,8) 4.31 4 0.36 2398.44 13 2424.44
10 Reduced Model (2,4,7,8) 5.39 6 0.49 2399.52 11 2421.52
11 Reduced Model (2,5,6,8) 4.38 5 0.50 2398.51 12 2422.51
12 Reduced Model (2,5,7,8) 4.40 6 0.62 2398.52 11 2420.52
13 Reduced Model (3,4,6,8) 4.34 5 0.50 2398.46 12 2422.46
14 Reduced Model (3,4,7,8) 4.36 6 0.63 2398.48 11 2420.48
15 Reduced Model (3,5,6,8) 4.41 6 0.62 2398.53 11 2420.53
16 Reduced Model (3,5,7,8)2 4.43 7 0.73 2398.55 10 2418.55
' Based on comparison with unconstrained model
2 Includes allproposed constraints (pooling)








assessment. Long-term data were available for the
adjacent Blue Cypress Lake (which were used for
drought assessments; see Conservation and
Management), but these data did not sufficiently
represent the areas most used by kites during the years
of our study 0==0.48 for monthly averages from gauge
S-251E [the most applicable gauge for recent kite use]
with Blue Cypress Lake), undoubtedly due to changes in
water management policies during these most recent
years. However, based on our results, exclusion of this
area would have been highly unlikely to have influenced
our conclusions.
Assessing the effects of water levels on a
biological response of Snail Kites (or any other species
that can move large distances in short periods of time)
can be extremely difficult because of the spatial and
temporal variation in both the water levels and the
animals. We were unable to use stage (i.e., elevation of
the water surface) directly as our measure of water level
because the ground elevation differences among areas
result in stages among areas being incomparable. At
first glance, it seems that local water depth is a suitable
alternative to stage; however, we were very concerned
about two problems that result from using depth. First,
reliable ground elevation data (which are required to
determine depth from stage) are sorely lacking for most
areas. Although many of the gauges have known ground
elevation, there was often considerable distance between
the gauges and the areas used by birds such that these
known elevations were not reliable indicators of
elevations in the areas of interest. Secondly, the lake
habitats in particular, often had steep elevation gradients
such that a substantial range of water depths may have
been used even by a single bird in any given day. The
range of depths used by a given bird in a single foraging
bout on some lakes (e.g., Lake Kissimmee) often
exceeded the range of depths over hundreds of square
kilometers in marsh habitats. Consequently, our
assignment of elevations and subsequent depths would
have been quite arbitrary for many areas. To overcome
these problems we wanted a measure that was robust to
spatial variation (i.e., "standardized" such that it was
comparable among areas) but did not rely on our having
to assign an arbitrary elevation to a local site as would
have been required by using depth. We also wanted a
measure that would adequately reflect the temporal
variation of hydrologic conditions. For example, most
areas in Florida exhibit seasonal variation in water
levels. Consequently, we wanted a measure that enabled
us to assess the influence of water levels independently
of seasonal variation. Based on these considerations we
used the departure from monthly average stages for each
area as a measure of relative water levels. Specific


gauges used for this analysis are provided in Appendix 6-
4. A monthly average over the 26-year period from
1969-1994 (i.e., an overall average of the annual
monthly averages) was calculated for each area. The
period of record was used for areas that did not have
reliable records for all 26 years. The difference (i.e.,
the residual) between a specific month of a given year
and the long term average for that month was used as
our measure of relative water levels. The 26-year period
of time was sufficiently long to encompass long-term
variability and also coincided with the period of time that
Snail Kites have been monitored on an annual basis.
Because averages were calculated by month, rather than
annually, seasonal variation was taken into account.
Because this measure is using stage, rather than depth, it
also did not require assignment of a local ground
elevation. Consequently, we could apply the same
measure to birds within, as well as among, wetlands.
However, one general caveat that must be included is
that this is a measure of relative, rather than absolute,
water levels. If birds respond to absolute water
conditions (e.g., depth) then this measure may be
misleading. There has been considerable effort in recent
years to collect reliable data on elevation that can be used
to estimate depth. A re-analysis of these data may be
warranted in the future as these improvements become
more available; however some problems (e.g., steep
elevation gradients) will persist even with improved
information on depth.
We began our analysis with a univariate
approach to logistic regression (Hosmer and Limeshow
1989) using the departure from average stage (described
above) as a continuous independent variable. This
analysis initially indicated an effect of relative water
level (X2=8.27, 1 df, P=0.004). However, because
location (i.e., wetland) and water level are confounded
and our previous analyses indicated an effect of location,
we next tested a model that included the effects of both
location and water level. This test was consistent with
our earlier analysis indicating a location effect
(X2=55.88, 9 df, P< 0.001); however when included in
a model with location, the effect of water level on the
probability of movement was no longer apparent
(X2=0.01, 1 df, P=0.934).
Although these results indicated that relative
water level was not a major influence on the probability
of movement, we must emphasize that the hydrologic
conditions under which our study was conducted were
generally high water conditions throughout the study
area. Consequently, low water conditions that might
have triggered movements generally did not occur during
this study and inferences regarding the effects of low-
water conditions on movement probabilities could not be









made. However, previous studies (e.g., Beissinger and
Takekawa 1983, Takekawa and Beissinger 1989) have
indicated substantial dispersal of Snail Kites during low-
water conditions. Given that apple snails may die or
become unavailable to kites during dry conditions, these
previous reports are certainly reasonable, and may
indicate that the lack of an effect of water levels we
observed applies only to generally high water conditions.

THE EFFECT OF FOOD RESOURCES ON
MOVEMENT

One of the most commonly cited reasons for
animals to move is the availability of food (e.g., Krebs
et al. 1974, Greenwood and Swingland 1984, Pyke
1984). Animals move if food resources are low or if
there is potential for better resources elsewhere (Pyke
1984). Nomadic species are particularly believed to
move in response to sporadic food conditions (Andersson
1980). The most commonly suggested reasons suggested
for Snail Kites to move are water levels and food (e.g.,
Beissinger 1988, Bennetts et al. 1994, Sykes et al. 1995)
and low water levels are generally implied to represent
low food availability.
During 1993 and 1994 we conducted 343 hours
of foraging observations (including 814 prey captures) to
assess the influence of food resources on movement. To
minimize confounding variation, all observations were
conducted on adult birds between 2 hours after sunrise
and 2 hours before sunset. In addition, we restricted our
observations to days that were not unseasonably cold
(i.e., during the passage of cold fronts), were not
raining, and winds did not exceed 10 mph. For
comparisons of food acquisition, we also used only
complete observations of foraging bouts. That is,
observations in which we observed an individual for the
entire length of time it took to capture a snail.
There was also potential confounding attributable
to foraging behavior. Course hunting (hunting by low
flight over the marsh) is the most commonly used
method of prey capture in Florida (Beissinger 1983a,
Sykes 1987a) and accounted for 671 of 814 (82%) of the
captures we observed. Still (perch) hunting accounted
for the remaining 143 (18%) captures and the proportion
of use among these 2 behaviors was highly dependent on
habitat type (X2=249.78, 3 df, P<0.001). Perch
hunting was primarily observed in cypress prairie
habitats (Fig. 6-12). In contrast, we had a more reliable
sample of course hunting observations in each habitat
type, season, and year. Thus, our comparisons among
seasonal and annual food acquisition by kites was limited
to course hunting to reduce confounding between these
different behaviors.


100- S


a i


g20
0 -ARM
Qr Mi0 tNoh Ljh- Ok-ho0b CypO Prfi
Habitat Type
P.ch Hunng 400 0Hun0in 4

Figure 6-12. The percentage of captures by perch hunting
and aerial (course) hunting in each of 4 habitat types.

We compared food acquisition among seasons
and years using the foraging time per capture as the
dependent variable in an ANOVA model. Our results
showed a difference among years, seasons, and an
interaction between year and season (Table 6-9).
Capture times were lowest during summer, relatively
higher during spring, and still higher during autumn of
each year (Fig. 6-13). Capture times also tended to be
lower during 1994 compared to 1993 for each season.
These results suggest that higher movement probabilities
corresponded with seasons and years of higher food
availability (see Temporal Effects on Movement
Probabilities above). This is the opposite result of what
would be expected if movements were attributable to low
food availability.

Table 6-9. Analysis of variance table from model of
foraging time per capture as the dependent variable.
Mean square (MS) and F values are based on type III
partial sums ofsquares (i.e., they are adjustedfor all
other terms in the model and are not dependent on
the order of entry)(SAS Inc. 1988).

Source df MS F P
Year 1 386.79 29.65 <0.001
Season 2 374.80 28.73 <0.001
Year x Season 2 166.65 12.78 <0.001
Error 175 13.04


In addition to comparisons among seasons and
years, we also observed 8 movements of 7 radio-
transmittered birds (one bird moved twice) for which we
had foraging observations prior to a subsequent
movement. This enabled obtaining foraging observations
immediately after the movement for paired comparisons











1993 1994
E 25 1-








SPR SUH FALJ sPR SUM FALL
eason 20ear)
5 10








Figure 6-13. The mean ( SE)foraging time to capture
snails. Sample sizes (number of complete bouts observed)
are also shown.


of food acquisition before and after moving. To reduce
confounding, we restricted this analysis to those
observations where the movement was within 30 days of
obtaining the first foraging observation. This reduced
the potential for seasonal differences in food acquisition
to be confounded with differences between locations. To
further reduce confounding, we matched the time of
observations before and after moving. Thus, if the
foraging observations before moving were conducted
between 1100 and 1300 h, then observations after
moving were also conducted between 1100 and 1300 h.
We then tested the null hypothesis (Ho) that the
difference in mean foraging time per capture before and
after moving was zero. These data indicated no
difference in food acquisition before and after moving
(t=0.60, P=0.57). Furthermore, in 4 of the 8
movements we observed, birds increased the time
required to capture snails (Fig. 6-14). In the remaining
4 cases, birds decreased the time required to capture


1 2


12 12 12 1 2
IS5OI 155171 11A140 70555
Location
(Fl7p--hs


1 2 1 2
(53.83 153.6*


Figure 6-14. The mean ( SE) foraging time to capture
snails by radio-transminered birds before (location 1) and
after (location 2) moving. Sample sizes (number of complete
bouts observed) are also shown.


snails. Although these data are limited, they do not
support the hypothesis that birds are moving to enhance
their foraging opportunities. However, the only case
(bird 153.564) where we observed a second movement
immediately following a previous movement, was when
the time required to capture snails was higher than at all
other locations. This suggests that there might be a
threshold of food acquisition, below which birds will
move if there are more favorable sites elsewhere.

MODEL SELECTION AND DISCUSSION OF
EFFECTS ON MOVEMENT PROBABILITIES

We began our model selection using the
procedure described by Hosmer and Limeshow
(1989)(see Methods). First we conducted a univariate
analysis of each potential term (i.e., age, season, study
year, and location). Based on the preliminary analyses
above, we used location, rather than region. We also
used the 10-parameter model for location rather the full
17-parameter model. Because all terms were significant
at the a level (0.25) suggested by Hosmer and Limeshow
(1989), we next used each term in a multivariate model
of only main terms (i.e., no interactions were included
at this point). In both the univariate and multivariate
analyses all terms were significant (Table 6-10) and
consequently retained for further model selection
procedures. Next we began exploring which, if any,
interaction terms should be included in the model. A
LRT between a fully saturated model (i.e., with all
interaction terms) and the multivariate model without any
interaction terms indicated that at least some interactions
were warranted (Table 6-11, test 23). A comparison of
the fully saturated model with a model excluding the 4-
way interaction term indicated that the 4-way interaction
was not warranted (Table 6-11, test 24). However,
comparisons among models with various combinations of
2 and 3-way interactions indicated that some 2-way
interaction terms (Table 6-11, tests 2,3,7,8,10,12, and
13) and some 3-way interaction terms (Table 6-11, tests
17,18,19,20,22) were warranted. Our tests did not
support retaining an age location, age season, or age
* year interaction terms (Table 6-11, tests 1, 4, and 5);
but did support retaining 2-way interaction terms for
season x location and year location at a= 0.05 (Table
6-11, tests 2 and 3) and a season year interaction at
a= 0.10. All further tests using season x location and
year location interaction terms indicated that these
terms were warranted (Table 6-11, tests 7,8,10,12 and
13); however, retention of the season year interaction
was not supported when included with these other
interactions (Table 6-11, tests 9, 11, and 14). Thus, we
retained season*location and year*location interaction


| + +


6


4
3
2-

1










Table 6-10. Maximum likelihood analysis of variance table for univariate models (i.e., each source term
represents a separate model) and multivariate models (i.e., each source term is contained within 1
model) of potential sources of variation of the probability of movement between times t and t + 1.

Univariate Multivariate
Source 2 df P < 2 1 df P < X2
Age 5.38 1 0.020 9.87 1 0.002
Season 17.85 2 <0.001 22.35 2 <0.001
Year 28.28 2 <0.001 25.48 2 <0.001
Location 103.36 9 <0.001 110.98 9 <0.001
SBased on Wald / statistic (SAS inc. 1988).



Table 6-11. Likelihood ratio tests (LRTs) comparing conditional logistic regression models of the probability of
movement between times t and t + 1. The null hypothesis (H) from a LRT is that the reduced model (i.e., the model
with fewer parameters) fits the data equally well as the more general model (i.e., with more parameters). Thus, a
rejection of H favors the more general model and failure to reject Hefavors the more reduced model.

LRT
No. General Model' Reduced Model' LRT df P >

1 A S Y L A*L A S Y L 13.61 9 0.14
2 ASYL S*L ASYL 45.22 18 <0.01
3 ASYL Y*L ASYL 30.82 18 0.03
4 ASYL A*S ASYL 1.01 2 0.60
5 ASYL A*Y ASYL 2.99 2 0.22
6 ASYL S*Y ASYL 8.12 4 0.09
7 A S Y L S*L Y*L ASYL S*L 34.44 18 0.01
8 A S Y L S*L Y*L ASYL Y*L 48.84 18 <0.01
9 A S Y L S*L S*Y A S Y L S*L 4.64 4 0.33
10 A S Y L S*L S*Y ASYL S*Y 41.74 4 <0.01
11 ASYL S*L Y*L S*Y A S Y L S*L Y*L 2.51 4 0.64
12 ASYL S*L Y*L S*Y A S Y L Y*L S*Y 45.12 18 <0.01
13 ASYL S*L Y*L S*Y A S Y L S*L S*Y 32.31 18 0.02
14 ASY L + all2-way Interactions A S Y L S*L Y*L 22.23 17 0.18
15 ASYL S*L Y*L A*S*Y ASY L S*L Y*L 5.16 4 0.27
16 ASYL S*L Y*L A*S*L A S Y L S*L Y*L 21.06 18 0.27
17 ASYL S*L Y*L A*Y*L A S Y L S*L Y*L 32.76 18 0.02
18 ASYL S*L Y*L S*Y*L A S Y L S*L Y*L 59.63 36 <0.01
19 AS YL S*LY*L S*Y*L A*Y*L A S Y L S*L Y*L S*Y*L 33.59 18 0.01
20 ASYL S*L Y*L S*Y*L A*Y*L A S Y L S*L Y*L A*Y*L 60.77 36 <0.01
21 A S Y L +all 2 & 3-way Interactions ASY L S*L Y*L S*Y*L A*Y*L 39.57 39 0.44
22 ASY L +all 2 & 3-way Interactions AS Y L + all 2-way Interactions 110.87 76 <0.01
23 Fully Saturated (all interactions) ASYL 247.78 16 <0.01
24 Fully Saturated (all interactions) A S Y L +all 2 & 3-way Interactions 35.02 36 0.52
'Model terms areA =Age, S=Season, Y=Study Year, L=Location (with poolingfrom Table 6-8)








terms, but not a season* year term. Using a base model
with these 2-way interaction terms we tested whether the
data supported the addition of any 3-way interaction
terms. Our results indicted that two of the 3-way
interaction terms (Age*Year*Location and
Season*Year*Location) were supported by the data
(Table 6-11, tests 17,18, 19, and 20) and the remaining
two were not (Table 6-11, tests 15 and 16). Thus, based
on LRTs, our data support a model with all main effects,
two 2-way interaction terms (Season*Location and
Year*Location) and two 3-way interaction terms
(Age*Year*Location and Season*Year*Location).
The results from using AIC were not in
complete agreement with LRTs. The model with the
lowest AIC (Table 6-12, Model 7) was a model with all


main effects and only one 2-way interaction term
(Season*Location); however, the AIC of the model with
both of the 2-way interaction terms supported by LRTs
(Table 6-12, Model 12) was very close. AIC did not
support inclusion of the 3-way interaction terms.
Differences between models indicated by LRTs and AIC
are not uncommon and represent conceptual differences
between an approach of hypothesis testing (LRTs) and
optimization (AIC). Both approaches were consistent in
that they indicated that the effects of age, season, year,
and location were all supported by the data and that there
is an interaction between time and location. The
methods only differ in suggesting to what degree of
complexity the interactions are supported by the data.
Our results indicated that considerable


Table 6-12. Summary statistics used for model selection of conditional logistic regression models of the
probability of movement between times t and t + 1. The model with the lowest AIC (bold) would be selected if
based solely on AIC criterion.
Model Model Description -21n(g) np AIC
No.
1 Age (A) 2502.95 2 2506.95
2 Season (S) 2490.66 3 2496.66
3 Year (Y) 2477.48 3 2483.48
4 Location (L) 2398.55 10 2418.55
5 A S Y L 2339.23 15 2369.23
6 A S Y L A*L 2325.62 24 2373.62
7 A S Y L S*L 2294.01 33 2360.01
8 A S Y L Y*L 2308.41 33 2374.41
9 A S Y L A*S 2338.22 17 2372.22
10 A S Y L A*Y 2336.24 17 2370.24
11 A S Y L S*Y 2331.11 19 2369.11
12 A S Y L S*L Y*L 2259.57 51 2361.57
13 A S Y L S*L S*Y 2289.37 37 2363.37
14 A S Y L S*Y Y*L 2302.20 37 2376.20
15 ASY L S*L Y*L S*Y 2257.06 55 2367.06
16 A S Y L with all 2-way Interactions 2237.34 68 2373.34
17 A S Y L S*L Y*L A*S*Y 2254.41 55 2364.41
18 A S Y L S*L Y*L A*S*L 2238.06 69 2376.06
19 A S Y L S*L Y*L A*Y*L 2226.81 69 2364.81
20 A S Y L S*L Y*L S*Y*L 2199.63 87 2373.63
21 A S Y L S*L Y*L S*Y*L 2166.04 105 2376.04
22 A S Y L with all 2 & 3-way 2126.47 144 2414.47
23 Fully Saturated 2091.45 180 2451.45








movement occurs that is not directly related to water
levels. We cautioned, however, that this analysis was
conducted entirely during a period of relatively high
water. Based on anecdotal evidence we believe that Snail
Kites do move in response to low water levels. Thus, we
believe that there is a threshold response to water levels.
If water levels become low enough to negatively affect
food resources, then kites will likely move from that
area. However, during most years there is considerable
movement that appears to be independent of current
water levels. We suggest a hypothesis for this movement
pattern below.
Dispersal is also generally thought to be favored
when local resources (e.g., food) are low or better
conditions exist elsewhere (Horn 1984). In contrast, our
results from both within-year and between year
comparisons suggest that higher probabilities of
movement occur when food resources are high. We also
found that natal dispersal of juveniles was lower in areas
where food resources were likely to have been
depressed. At first it may seem counter-intuitive to leave
an area if food resources are high; however, we suggest
a hypothesis that this may be a reasonable strategy given
the dynamic and unpredictable nature of a kite's
environment. A virtual certainty about any specific
wetland inhabited by Snail Kites is that it will go dry.
What is not certain is which wetlands will go dry in
which years. Thus, there is an advantage for kites to
have experience regarding the availability of wetlands
throughout their range so that when a local drying event
does occur past experience reduces the need for "blind"
searching for suitable alternative habitats. Thus, in years
that food is not limiting, which for kites may be most
years, high food resources may enable kites to "explore"
their potential habitats with little risk of starvation. The
resulting experience from many locations may then help
kites to locate food resources faster during times when
food is limiting.

Spatial Patterns of Movement

Here we ask the question that given a bird at
location r at time t has moved, what is the probability
that it will be at location s at time t + 1. Thus, our
conditional setting is that a bird is alive, its location at
times t and t + 1 are known, and it has moved from its
location at time t.
We approached this problem by first establishing
whether the locations at time t and t + 1 are independent
(i.e., our null hypothesis [Ho] is that the probability of
moving to a given location does not depend on the bird's
previous location). For our initial exploration we used a


fully unconstrained data set with no pooling of locations).
This hypothesis (Ho) is strongly rejected (2= 1014.41,
df=256, P < 0.001). Thus, our data suggests that the
probability of movement to a given location does depend
on the bird's previous location. The residuals from this
analysis provide some insight into how these movement
patterns departed from what would have been expected
if Ho were true (Fig. 6-15).

EFFECT OF DISTANCE

What becomes apparent from the above residuals
is a tendency to move to areas in relative proximity to a
bird's current location. For example, birds at WCA-3A
had the greatest positive departure from expected when
they moved to Everglades National Park and Big
Cypress National Preserve, both of which are
immediately adjacent to WCA-3A. The greatest negative
departures from expected were movements to the lakes
within the Kissimmee chain-of-lakes (at the opposite end
of the kites' range in Florida. This pattern of higher
than expected values for locations in proximity and lower
than expected values for distal areas can be seen
throughout these data.
We further explored the relationship between
distance and the location of movements by ranking each
location (wetland) with respect to its relative distance
from each other location. Thus, the closest wetland to
any given wetland was given a rank of 1 and the furthest
was given a rank 15. We excluded the peripheral
habitats and the Kissimmee Chain-of-Lakes from this
analysis because they were not contiguous and it would
have been virtually impossible to perform an analysis for
every wetland. When more than one wetland was
immediately adjacent to another, we used the areas that
were most frequently used by kites within each wetland
to determine their relative ranking. We then tested for
the effect of distance using the null hypothesis (HJ that
the distance rank is equal for all movements (i.e., the
expected values for a goodness-of-fit test among distance
classes were equal). We strongly rejected H.
(X2=320.83,13 df, P <0.001) and concluded that
distance does have an influence on the locations to which
a bird moves. Not surprisingly, the residuals from this
analysis indicated that birds exhibited a greater than
expected frequency of moving to a new location in
proximity to their previous location, and a lower than
expected frequency of moving to areas that were most
distant from their previous location (Fig. 6-16). This
does not imply that birds will not move long distances
(our data show that they do); but rather, that long
distance movements have a tendency to be made in short
increments.










-a to EVWdONMWWIPk(ENP)
I P 1




E10 E Kim HOLEY. S P...

t0 -Cyp -N- "P.W...BCY)


cc I F
(DI
10






BEf A W LOX "CAB JM TOHO K-CHA
5I NoWth EaM Shwk Slo ug P NES.C
|-o|
10
WI
-Dc










:5 --_ .











0
Of KHE- WOOSB CA OE 0 XE i IS ETOHO PET

u A M LLorhzw N4W.R. LOX)


Hooy ULa W.MA (HOLE)
z
'0 M IM
0

-B
0


o B
aw 6p d aW W0 a8 AW00400f
^ .--O-"-- jgg










0E NEwRS WCAB WCAO HOLEY OEE 0 O ErH0O PER
V WOAOA WOS LOX IW It S TON K'

to



W NI OCB W"A2A HOLEY4 a* 000 ETOHO pow
KY WCA WC- LOX WLo -t TOMO K1CH












CI
cc
< oV E 91 000WIW0A0LEY OIWE KSa ETwHO PERt
S0----
3 LOxM t Skeo gho (WPBE)


0 SLJotnrMwhtSJM)

* I
cc
:3
WI
I 5
a


W a-


HL Er -10
OVCY W W LmOX W HL S S K-000 40 EEOOa WO SW.AA C LOXX 5 I TOHO" K-00
Location at Time t+1 Location at Time t+1


Figure 6-15. Adjusted residuals from the crosstabulation of the frequency of movements to each location (at time t + I)from each
location at time t. A residual value of 1.96 is an approximate indicator ofthat residual being significant at an a = 0.05 level of
significance.


_81


I -1














I---.


WEN Y 10 WCAOO WOO HOXLY WE US 01 T
tOY WOCM WOOB tLOX 01B LU TO0 000


-a 10 ijiTohcdioi

05
LO
o
S Tgg^gl
1<.5


BCt WCA3 WOS LOX W6P 00 T0O K-0HA

E L W' ElLT&opwlg>(ETOHO)



o
t-5

E0 0 0


'a 10------ OChneMl esW (KCHN




1 .5
a g a


I PW-ItW H0bb (PER1F)

II I
C-0
^ a~ o~ g g g-

____________< ll ---- ------ r


0e1 I0 WC0 S WOS1. H00 OEE MU E100 000

Location at Time t+1
Figure 6-15. Cont.


5 -






_10-


2 3 4 5 6 7 8 9 1011 1 13 1 15
Distance Rank

Figure 6-16. Standardized residuals from a contingency
table of the frequency of movements to wetlands in each
relative distance class. Expected values were derived under a
null hypothesis (H.) of equal probability for each distance
class.





THE EFFECT OF AGE, SEX, AND TIME ON
MOVEMENTS BETWEEN SPECIFIC LOCATIONS

We tested for the effects of age, sex, and time
(i.e., season and year) on the location-specific movement


probabilities using a LRT between a log-linear model of
specific location effects (using the reduced model from
Table 6-10) and models which include the effects of each
of these independent variables. We found no effect of
age (X2= 17.7, df=23, P=0.773), sex (X2 =27.9,
df=47, P=0.988), season (X2=39.04, df=44,
P=0.684), or year (X2=11.72, df=48, P>0.999).
This does not imply that there were no effects from these
independent variables, but rather that if there were such
effects, our data were insufficient to detect them. Given
the large number of parameters (> 100) in these models
this latter result would not be surprising.

SEASONAL SHIFTS IN LATITUDE

It has been previously suggested (e.g., Sykes
1983a, Sykes et al. 1995) that Snail Kites tend to move
south during colder winters. Although no data have
previously been presented in support of these
suggestions, our data are consistent with this conclusion.
There was a general tendency for Snail Kite occurrence
to shift north during the summer (May-August) and south
during Fall/Winter (September-December)(x2=27.68,
df=4, P<0.001)(Fig. 6-17).
We used a LRT to test whether this relationship
was influenced by age, sex, or year by comparing a fully


15 10 MW
10 IWO 1issi0t 0098)
0


'.10
^ 1 i-i- -i-i- r


anIEU


000 00*00s- w WOOs Hoim 000 1000 0.00


am- I


E P S WOYOL WCA o EY Oi 0 OM ETOHO tW
Y 0C00 WOcS LOX WPB UM TOLHO K-C0
Location at Time t+l


W I


1


1= li
* IN do M


I


Easc~


w





















J--AWp May-Aug S. -D.C
Season
Figure 6-17. Adjusted residuals from a contingency table of
the frequency of movements to wetlands in northern, central,
and southern regions during each season.


saturated model to one without each of these effects.
These tests indicated that these seasonal shifts in latitude
were influenced by age (LRT=15.68, df=4, P
=0.004), sex (LRT=22.47, df=8, P=0.004), and year
(LRT=51.60, df=8, P<0.001).
Although we found evidence to suggest that there
were effects of age, sex, and year on these seasonal
shifts, the AIC was substantially lower for a reduced
model without these effects (9,837 compared to 12,651
for the closest model). Although this result appears
contradictory, it is not uncommon, and reflects the
difference between hypothesis testing where a specific
effect is being tested and model selection, which views
model selection as more of an optimization problem
(Spendelow et al. 1995). In other words, there may be
real influences of age, sex, and year, but our data are
insufficient to effectively incorporate separate parameter
estimates for each of these influences.

SHIFTS IN REGIONAL USE

Factors such as resource abundance and
disturbance that influence patterns of distribution at one
scale may be expressed quite differently at a different
scale (Wiens 1989). Consequently, we looked at broad-
scale movement patterns over the duration of the study
in addition to patterns occurring within monthly time
steps. We accomplished this by examining how the
proportion of birds in different regions shifted throughout
the study.
Because our capture protocol used the proportion of
birds in different areas as a biasis for our sampling, we
believe that our initial conditions were a reasonable
representation of the distribution of birds at the start of


this study. We then looked at the proportion of locations
each month in each region to examine how these
proportions changed over time. It was quite apparent
from these data that shifts in regional use had occurred
over the duration of the study. We tested for a linear
trend over the 37 month period from April 1992 through
April 1995 using linear regression on the proportion
(after an arc-sin transformation) of locations in each
region. We tested the null hypothesis (Ho) that the
observed slope over this period was zero using a t test
(SAS Inc. 1988).
The proportion of birds using the Everglades
Region decreased over this interval (t=6.27,
P<0.001)(Fig. 6-18). The proption of birds using the
Okeechobee and Upper St. Johns Regions decreased over
the interval (t= -5.77, P<0.001 for the Okeechobee
Region and t=-3.94, P <0.001 for the Upper St. Johns
Region). The proportion of birds did not exhibit a linear
trend for the Kissimmee Region (t=-0.028, P=0.916),
the Loxahatchee Slough Region (t=-1.415, P=0.166),
or the Peripheral Region (t=-1.088, P=0.284); although
the seasonal shifts in some regions (e.g., the Peripheral
Region) may have been better represented by a higher
order polynomial regression.

Effect of Hydrology- Our analysis of regional
shifts is strictly correlative and can not reliably be used
to infer cause and effect. However, it is quite possible
that, at least some of the shifts in regional use, were
attributable to hydrologic changes. For example, the
Everglades Region experienced a relatively severe
drought just prior to our study. It is quite likely that the
low proportion of birds in this region at the beginning of
our study was attributable to this drought. Our anecdotal
observations of foraging birds in this region during 1992
clearly indicated that food resources were diminished.
We often observed birds foraging for long periods
(sometimes > 1-2 hours) without capturing a snail;
whereas, under good conditions, foraging birds capture
snails in just a few minutes (see Effect of Food Resources
on Movement).
The relatively high proportion of birds in the
Okeechobee and Upper St. Johns Regions at the
beginning of this study may well have been birds
displaced from the Everglades Region. The Okeechobee
Region also had experienced the drought preceding our
study; however, virtually all of the birds we observed in
this region at the beginning of our study were confined
to the outer marsh of Lake Okeechobee which had not
dried during the drought. Over the duration of our study
the proportion of birds decreased at Lake Okeechobee
and those birds present in this region shifted from the
outer marsh to more interior marshes, that had been dry











Everglades Region


S-





Okeechobee Region
80
so
-40









Kxahatee ChainSlough Region
so0





60
I40



20
40 -

20 -

Upper St Johns River Region



40
S20
S-
0-
Lonuhalchee Slough Region


015.

so



Peripheral Region
880so

60

o40

20

0
AMJJASONDJ FMAMJJASONDJ FMAMJJASO N O J FMA
1992 1993 1994 1995
Month (Year)

Figure 6-18. The percentage of Snail Kite locations in each
region during each month from April 1992 through April
1995.


during the preceding drought. The Upper St. Johns
Region had not experienced a drought just prior to our
study. The Northern Lakes Region also had not
experienced the recent drought and a higher proportion
of birds in this region might also have been expected.
However, the Northern Lakes Region had extremely
high numbers of birds in 1991 (J. Rodgers Jr, J. Buntz,
GFC, Pers. Comm.). Thus a high proportion of birds
probably had been using this region, but had already
dispersed by the start of our study.

SEASONAL SHIFTS IN HABITAT USE

During this study it became apparent that there
were substantial seasonal shifts in the use of different
habitats by Snail Kites. Consequently, we conducted a
"post hoc" evaluation of these habitat shifts. Because
this study was not designed to evaluate habitat
relationships and because there is high variability in
micro-habitat, our analysis was limited to five broad
habitat categories (described in Study Area). We assigned
each location to one of these five habitat classes. We then
examined how the proportion of use of each habitat for
each month shifted over the duration of the study.
Our results indicated that Snail Kites exhibit
strong seasonal patterns in their relative use of some
habitats (Fig. 6-19). Relative use of Lake
Okeechobee fluctuated greatly, but was not as
predictable seasonally as other habitats. Use of Lake
Okeechobee was relatively high for the first year of our
study then dropped to very low use during the winter of
1993-1994, and increased again to moderate use
(discussed above in Shifts in Regional Use).
The use of other habitats exhibited more
seasonal fluctuation. Peak use of graminoid marshes and
northern lakes coincided with the periods of major
breeding activity; whereas, cypress prairies and
peripheral habitats were used more extensively outside of
the primary breeding season. One potential explanation
for the use of graminoid marshes and lake habitats
during breeding is the relative predictability of
hydrologic conditions in these habitats compared to
cypress prairies or peripheral habitats. Cypress prairies
tend to have shorter hydroperiods than are typically used
by breeding birds. Given that a breeding attempt
requires 10-16 weeks per clutch (Snyder et al. 1989a),
the probability of an area drying out in cypress prairie
during a breeding attempt is greater in this habitat
compared to those typically used by nesting kites. Thus,
the probability of a nesting attempt failing probably is
also greater. Similarly, peripheral habitats were usually
more ephimeral wetlands or were used for agricultural
purposes and were subject to extreme fluctuations for










Okeechobee




a
so .


Nortarn LBa


Gramlnold Marsh


1992 1993 1994 1995
Month (Year)
Figure 6-19. The percentage of Snail Kite locations in each
habitat during each month from April 1992 through April
1995.


agricultural use. Thus, the probability of this habitat
remaining in suitable hydrologic condition for the extent
of a breeding attempt also was likely diminished.
Although hydrologic predictability may provide
an explanation for why graminoid marshes and lake
habitats were more likely used during breeding seasons,
it does not explain why cypress prairies and peripheral
habitats were used during non-breeding periods. We can
offer some potential hypotheses, all of which remain to
be tested. First, birds in cypress habitats tend to hunt
from perches, rather than by flight (see The Effect of
Food Resources on Movement). This could offer an
energetic advantage since perching requires less
energetic expenditure than flight. Secondly, using
alternative habitats to the primary breeding habitats could
allow snail populations to be repelinished prior to the
next breeding season. The summer months when birds
shift to cypress and peripheral habitats also coincides
with peak breeding activity of apple snails. Lastly,
exploring alternative habitats when food resources are
not limiting may better enable kites to locate food during
periods of localized drought. This latter hypothesis was
previously discussed in more detail in the section on
Model Selection and Discussion of Effects on Movement
Probabilities.
The detectability of Snail Kites is probably very
low in both cypress prairies and peripheral habitats
compared to graminoid marshes, the northern lakes, or
the Lake Okeechobee habitat types. Many of the
peripheral habitats were either on private land (e.g.,
agricultural areas) or on public land in which access was
limited (e.g., water control districts). Much of the
cypress habitat also had very limited access (e.g., some
management units of Big Cypress National Preserve do
not allow airboat access) and even large numbers of
birds in this habitat type were very easy to overlook
because of the dense vegetation. On several occasions
we had reports of a few birds in cypress habitats only to
discover with more intensive searching that the "few"
birds turned out to be a large number of birds (50-400).
This seasonal use of habitats with low
detectability can have important implications regarding
population monitoring of Snail Kites in Florida. For
example, the annual count conducted each year (see
Monitoring of Snail Kite Populations in Florida) in
Florida coincides with the period of relatively high use
(July-January) of the peripheral and cypress prairie
habitats. Consequently, the number of birds detected
during these surveys may be greatly influenced the
number of birds in these habitats. This can cause a
substantial undercounting of birds and may influence the
variability among counts depending on what proportion
of the population is in these habitats in any given year.









These problems are discussed in detail in the chapter on
Monitoring ofSnail Kite Populations in Florida.
Beissinger et al. (1983) reported that Snail Kites
in Florida have long been known to "disappear" from
their usual haunts in summer and subsequently
"reappear" during mid-October. This observation led to
the speculation that Snail Kites in Florida may move to
Cuba (Beissinger et al. 1983). No birds that had been
banded in Florida were found during an expedition to
search for such birds in Cuba (Beissinger et al. 1983).
Although movement to Cuba is certainly possible, our
data suggest that a more simple explanation for the
disappearance of kites during summer may be the shifts
in habitat use. The time of disappearance of birds
reported by Beissinger et al. (1983) coincides with the
time that we observed increased use of peripheral and
cypress habitats (where they are less likely to be
observed). The reappearance in mid-October also
coincides with the time that birds begin shifting back to
the graminoid and lake habitats.



Natal Philopatry and Site Fidelity

Philopatry is usually defined as the tendency for
an animal to remain at (or very near) its natal site (e.g.,
Mayr 1963, Shields 1984). However, Snail Kites by
their nature do not typically remain anywhere, but may
have an affinity to return to their natal site.
Consequently, for our analyses, we have relaxed the
definition of philopatry to the tendency for a bird to
occur in its natal wetland even though birds may, and
probably do, regularly also occur elsewhere. We also
emphasize from the outset that our analyses of natal
philopatry and site fidelity were intended only as a "post
hoc" cursory exploration. Our study was not designed to
answer questions regarding these topics, and we urge
caution in interpretation of our results beyond
preliminary exploration. However, because almost no
information exists on natal philopatry or fidelity for this
species (Sykes et al. 1995), we believed that this
preliminary exploration was warranted.

NATAL PHILOPATRY

Our primary data for assessing natal philopatry
were the resightings of 414 banded Snail Kites, whose
natal wetlands were known. Of these sightings 157
(40%) were of birds that were sighted in their natal
wetland. The proportion of birds observed in their natal
wetland was not dependent on the breeding status (i.e.,
actively nesting or not nesting) (X2=0.45, 1 df, P=0.50)


based on a subsample of 255 birds whose breeding status
was known (Fig. 6-20). The proportion of birds
observed in their natal wetland was, however, influenced
by the bird's sex (X2=6.02, 1 df, P=0.014)(Fig. 6-21),
natal wetland (x2=39.75, 7 df, P<0.001)(Fig. 6-22),
and the year (x2= 10.12, 3 df, P=0.018)(Fig. 6-23).


enag Not BRre
01=106) f-149)
Breeding Status
Figure 6-20. The percentage of banded birds that were
observed to be actively breeding (including courtship) or not
breeding that were or were not resighted in their natal
wetland. Only birds whose breeding status was known were
included in our sample.







4-

a ,,

to
0 -T




a
(-125) Sex 0=170)


Figure 6-21. The percentage ofmale and female banded
birds that were or were not resighted in their natal wetland.


Differences in natal philopatry between males
and females are not uncommon among bird species and
the avoidance of inbreeding is a frequently suggested
reason for these sexual differences (e.g., Bengtsson
1978, Greenwood et al. 1978); even though there are
few examples of the harmful effects of inbreeding in
birds (Greenwood and Harvey 1982). Differences
among sexes in natal philopatry usually are reported for
philopatric species and not generally known to occur in
nomadic species (Greenwood and Harvey 1982). Snail
Kites in Florida, however, probably have a more










120
|inNdWfand
S100- 0 Not In Nd Wellald


06

S40



WCA3A WCA2B OKEE 08 SJM TOHO ETOHO OTHER
0_a V_ 62 6.114 0-60 t-M0 0_ -1M I.-i)
Natal Wetland



2 4.








WCA3A WCA2B OKEE KISS SJM TOHO ETOHO OTHER
Natal Wetland

Figure 6-22. The percentage of banded birds from each
(natal) wetland that were or were not resighted in their natal
wetland. Only birds whose whose natal origins were known
(n=414) were included. Also shown are the adjusted
residuals from a crosstabulation of natal wetland and
whether or not they were resighted in their natal wetland.
For visual clarity, only the residuals from birds that were
resighted in their natal wetland are shown; however the
residuals from birds that were not resighted in their natal
wetland are an approximate mirror image of those shown.


restricted range than most nomadic species and
consequently, similarly to philopatric species, may
have a greater probability of inbreeding. Thus, our
results are not necessarily inconsistent with the
inbreeding avoidance hypothesis; nor do we have any
data to support this potential explanation for the
differences in philopatry between male and female Snail
Kites.
The differences among years and specific
wetlands are even less clear. Our data were insufficient
to explore the details of interactions among these sources
of variation; however, interaction effects would be
likely. For example, if one part of the Snail Kites' range
had an ongoing drought then it would be unlikely for
birds that had fledged from that part of their range to be
in their natal wetland until conditions improved.
Consequently, our results should be viewed in light of
current conditions and probably should not be


extrapolated beyond the environmental conditions we
encountered. The extent of fidelity to a given birds natal
wetland is undoubtedly influenced by the current
condition of that natal wetland and other wetlands
throughout the state.

SITE FIDELITY

In the previous section we addressed the
tendency for Snail Kites to return to their natal wetland.
Using radio telemetry, we also examined whether
individual birds have a tendency to return to the same
wetlands from year to year, irrespective of their natal
area and breeding status. To test fidelity, White and
Garrott (1990) suggest using a chi-square test of
independence for situations in which locations of animals
can be assigned to discrete areas (e.g., our designations
of wetlands). For our initial exploration, we used this








procedure, without any pooling, for all individuals for
which we had sufficient data. Because we have already
shown that there are strong seasonal effects in locations
used by Snail Kites, we conducted analyses
independently for each season (using the season
designation previously supported by our data). We did
not include any individuals in this analysis for which we
had fewer than five radio locations in each season of
each year of comparison. Although five locations was an
arbitrary criterion, we believed that fewer locations
would likely have produced spurious results because we
would have likely had representation from only a small
portion of the season. Based on our sampling interval
(= 14 d), five locations were likely to have represented
locations from over half of the season of interest (the
expected number of locations during any one season, if
the radio were operational for the entire season, was =
8). Additionally, chi-square tests have been known to
perform poorly when the expected values of cells are less
than five (Cochran 1954). Our criterion was intended to
reduce these problems.
A departure from fidelity was common for Snail
Kites in each season. During the late-winter/spring
(January through April) season 5 of 7 (71%) of the
individuals, for which we had sufficient data to test
fidelity, showed a departure from fidelity at a=0.05
level of significance (Table 6-13). Similarly, 16 of 24
(67%) individuals showed a departure from fidelity
during the summer season (May through August)(Table
6-14), and 4 of 8 (50%) showed a departure from fidelity
during the autumn/early-winter season (September
through December)(Table 6-15).


Because each test was an individual analysis
from an individual bird, we did not attempt any detailed
model selection for additional effects (e.g., sex or year).
We also did not believe that the data warranted any
extensive meta-analysis for these effects. On a purely
descriptive note, however, we found no obvious patterns
with respect to sex or year. There was a slight tendency
for a greater departure from fidelity during the non-
breeding seasons (summer-early winter) early during the
study, but given the small sample sizes, we are not
convinced that this is anything more than random
variation.
We emphasized at the beginning of this section,
that this was a "post hoc" exploratory analysis. In
particular, there are two problems which need to be
identified. First, except for the summer season, in
which we had 24 individuals meeting our criteria, these
tests were conducted using a very small number of
individuals. These small sample sizes were likely a
result of the battery life of our radios. The expected
life of our radios was 12 months and this may have been
an optimistic expectation from some of the
manufacturers. Thus, our analyses were only possible
for individuals whose radios exceeded their expected life
(or were replaced) and whose locations were generally
known. Because most radios were attached during
spring, the summer season was the first full season
following attachment and the most likely to have had
individuals for which we had > 1 year of data. This is
probably why our sample was so much higher for
summer. The specifications of the mortality switches of
our radios precluded some radio designs that would have


Table 6-13. Test results from j tests of independence for the number of locations of individual radio-
transmittered Snail Kites during the late-winter/spring season (January April) at each location between years.
Freq Sex Year 1 Year 2 N 1' N 22 X2 df P
152.104 M 93 94 5 5 6.67 1 0.010
152.630 M 93 94 7 6 4.55 2 0.103
152.362 F 93 95 17 7 24.00 4 <0.001
152.030 F 94 95 5 7 2.86 1 0.091
152.480 F 94 95 8 7 4.77 1 0.029
152.499 F 94 95 6 8 14.00 3 0.003
152.584 F 94 95 7 6 13.00 2 0.002
STotal number of locations for individual during this season for year 1.
Total number of locations for individual during this seasonforyear 2.
SBecause df = rows-1 columns-1 and all individuals had locations in only years (rows) for this season, the total number of
locations we observed for the individual equals df+1.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs