Front Cover
 Title Page
 Table of Contents
 Procedures and data
 Conceptual framework
 Pesticide residues in Florida strawberries...
 Modeling and empirical analysi...
 Empirical results
 Summary, conclusions, and...
 Appendix A. Derivation of a weighting...
 Back Cover

Group Title: Bulletin
Title: Empirical relationships between pesticide residues, producer attributes, and production practices for Florida grown tomatoes and strawberries
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00008441/00001
 Material Information
Title: Empirical relationships between pesticide residues, producer attributes, and production practices for Florida grown tomatoes and strawberries
Series Title: Bulletin
Physical Description: v, 50 p. : ill. ; 28 cm.
Language: English
Creator: Stevens, Thomas J., 1952-
Kilmer, Richard L
University of Florida -- Agricultural Experiment Station
Publisher: University of Florida, Agricultural Experiment Station, Institute of Food and Agricultural Sciences
Place of Publication: Gainesville Fla
Publication Date: 1999
Subject: Pesticide residues in food -- Florida   ( lcsh )
Pests -- Integrated control -- Florida   ( lcsh )
Field experiments -- Florida   ( lcsh )
Strawberries -- Florida   ( lcsh )
Tomatoes -- Florida   ( lcsh )
Genre: government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
non-fiction   ( marcgt )
Bibliography: Includes bibliographical references (p. 47).
Statement of Responsibility: Thomas J. Stevens III and Richard L. Kilmer.
General Note: "February 1999"--Cover.
 Record Information
Bibliographic ID: UF00008441
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: ltqf - AAA6706
ltuf - AME7040
oclc - 41232676
alephbibnum - 002441827
lccn - 99523995
issn - 0096-607X ;

Table of Contents
    Front Cover
        Front Cover 1
        Front Cover 2
    Title Page
        Page i
        Page ii
        Page iii
        Page iv
    Table of Contents
        Page v
        Page vi
        Page 1
    Procedures and data
        Page 2
        Page 3
    Conceptual framework
        Page 4
        Page 5
        Page 6
        Page 7
    Pesticide residues in Florida strawberries and tomatoes
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
    Modeling and empirical analysis
        Page 15
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
    Empirical results
        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
    Summary, conclusions, and recommendations
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
    Appendix A. Derivation of a weighting function for pesticide applications
        Page 49
        Page 50
    Back Cover
        Page 51
        Page 52
Full Text

February 1999 Bulletin 329

Empirical Relationships Between

Pesticide Residues, Producer

Attributes, and Production Practices

for Florida Grown Tomatoes and

Thomas J. Stevens III and Richard L. Kilmer

Agricultural Experiment Station
Institute of Food and Agricultural Sciences


Bulletin 329

February 1999

Thomas J. Stevens Ill is a post-doctoral associate and Richard L. Kilmer is a professor in the Food and Resource Economics Department
at the University of Florida in Gainesville, Florida.

Empirical Relationships Between Pesticide Residues, Producer Attributes, and
Production Practices for Florida Grown Tomatoes and Strawberries

Thomas J. Stevens III and Richard L. Kilmer


Funding for this research was provided by the United States Department of Agriculture
(USDA) Economic Research Service (ERS) and Agricultural Marketing Service (AMS).
Numerous individuals in industry and government provided information and advice that
were essential to the completion of this study. Some of those who were frequently called
upon for this purpose included: Biing-Hwan Lin with the USDA-ERS; John Schaub of the
USDA-AMS; George Fong, Gail Parker, and Sherv' Spencer of the Florida Department of
Agriculture and Consumer Services; Mack Willhite, Aubrey Bordelon, and Robert Freie of
the Florida Agricultural Statistics Service. Individuals from various departments within the
Institute of Food and Agricultural Sciences at the University of Florida were: Michel Roy,
Scott Smith, and Ron Muraro with Food and Resource Economics; Jeff Brecht, Phyllis
Gilreath, George Hochmuth, Steve Sargent, Bill Stall and Robert Stall of Horticultural
Sciences; Dick Matthews and Norman Nesheim from Food Science and Human Nutrition;
and Karl Butts and Ben Castro of the Cooperative Extension Service. Other agencies or
associations which provided input include: the South-East Agricultural Weather Service
Center, the Florida Tomato Committee, the Florida Strawberry Growers Association, the
Florida Fruit and Vegetable Association, and the National Association of State Departments
of Agriculture. Finally, special thanks are due to the hundreds of Florida growers, packers,
and distributors of strawberries and tomatoes who made this study possible by voluntarily
providing detailed information on their individual cultural or handling practices.


This research delves primarily into the technical relationships between the cultural practices
used to produce fresh tomatoes and strawberries in Florida and the occurrence or lack of
pesticide residues in these commodities. Data on the levels of 19 different pesticide residues
found in Florida tomatoes and strawberries are matched to firm and decision-maker
attributes as well as the cultural and handling practices used in their production. Three
different regression models are developed and operationalized for each commodity at the
producer stage. The first model relates the levels of different aggregate pesticide residues to
socio-demographic attributes of pest control decision-makers and their firms. The second
explores the relationships between a wide array of general production practices and various
aggregate residues. The third model examines the relationship between levels of specific
pesticide residues and the application of these particular pesticides during production.
Model performance and individual factor results vary considerably between commodities
and types of residues. Altogether, over fifty empirical relationships are tested. Among
those found to be influential are decision-maker education, firm affiliation, the ratio of
rented verses owned acres used in production, soil type, weather, fertigation. drip irrigation,
soil testing for pests, length of growing period, plant variety, and use of economic thresholds
in pest control decisions. Topical as well as methodological recommendations are provided
for future research and policy directives.

Key Words: pesticide residues, strawberries, tomatoes, Florida, grower attributes, cultural
practices, pest control, integrated pest management.

Table of Contents

A cknow ledgm ents ................... ..... ...... ........ .................... iii
A b stract .................................................... ...................................... iv
Introd auction ............................................ ....... .............. ...... ................ . 1
Problem Setting and Objectives ............................ ...................... 1
Scope of Research ................................................. 1
Organization of Report ........................................................ 2
P procedures and D ata ........................................... ............. ........... 2
Conceptual Fram work ......................................................... .... ............ 4
Pesticide Residues in Florida Strawberry and Tomatoes .............................. 8
M odeling and Em pirical Analysis .................................... ............ 15
B asic M odel Specification ..................... ... ........ .... ................. 15
Aggregation of Left and Right Hand Side Variables ............................... ... 17
Data-Related Regression Problems ..................................................... 17
Em pirical R results ........................... ............. ... .................. 24
A attributes M odels ..... .............................................. .................... 24
General Practices Models ................. ................................. 28
Fertilizer and Fertilizer Management ................................. 28
Integrated Pest Management ..................................... .......... 31
Irrigation, Crop Rotation, Days from Plant to Harvest, and
Plant Density ........................................... .......... 32
Plant Varieties .............. .. ......... .. .............................. 33
W weather .................................. ............ 33
Specific Practices M odels ........................ ............. ... ................. 33
Straw berries ......................................... .......................................... 34
Tom atoes ........................ . ..................... ........... 39
Summary, Conclusions, and Recommendations ..................................... 40
Sum m ary ........ .... ...... ...... ................. .. .. .. .............. 40
Conclusions ............. ......... .... ....... ..................... 40
Recommendations ........... ..... ...... ......................... 44
Topical Recom m endations .............................. ................................ 44
M ethodological Recom m endations .......................................... ...... 46
B ib liog rap h y ........................................ ............................. ....... .... 47
A ppend ices ...................................... ............................... 49
A. Derivation of a Weighting Function for Pesticide Applications.

Empirical Relationships Between Pesticide Residues, Producer Attributes,
and Production Practices for Florida-Grown Tomatoes and Strawberries


Problem Setting and Objectives

During the last two decades there have been growing societal concerns over issues
related to public health, environmental quality, and food safety. One of the major
controversies inciting these concerns involves the production and consumption of fresh fruits
and vegetables. Research has shown that diets with greater proportions of fruits and
vegetables can prevent or delay many debilitating and life threatening diseases. At the same
time, public acceptance and adoption of these findings is being discouraged by ongoing re-
evaluations of the possible health risks associated with minute amounts of pesticide residues
sometimes found in these foods. These concerns have motivated a number of government
agencies to tighten their regulatory guidelines and controls on the use of pesticides in food
production and processing. Recently, the Clinton Administration proposed that a one year
project be undertaken to establish commodity-specific pesticide use reduction goals that
would be achieved by the year 2000 (EPA 1993). The Administration believes that one of the
best means of accomplishing such goals would be through the widespread adoption of
integrated pest management (IPM) techniques.

While a significant amount of scientific research is being directed toward finding safer
ways to combat agricultural pests, in many instances very little is known about how cultural
and handling practices commonly used in our current food production and processing
systems influence the occurrence of pest infestations, the effectiveness of various chemical
pesticide treatments used to control them, and the fate of these pesticides after they have been
applied. The primary objective of this study is to explore these relationships by empirically
modeling the levels of pesticide residues found in Florida fresh fruits and vegetables as a
function of the cultural and handling practices used to produce, process, and distribute them.
The implications and conclusions from this investigation can help researchers, government
agencies, and private decision-makers identify existing cultural and handling practices that
have the potential to reduce pesticide applications and residues in these foods without
imposing significant disruptions to their production or supply. A second objective of the
study is to evaluate the relationships between residue levels and various socio-economic
characteristics of growers and handlers. A better understanding of these associations could
help policy makers more effectively target extension and education programs aimed at
reducing pesticide use and/or residues.

Scope of Research

The scope of this investigation was limited to fresh tomatoes and strawberries grown in
the state of Florida from October 1990, to June 1993. Data on cultural and handling practices
as well as firm and decision-maker attributes were collected and evaluated for the grower,
packer, and distribution stages of each market channel. Together, nineteen different pesticide

residues were detected in these two commodities. These included insecticides, miticides,
fungicides, and herbicides. The production activities examined in this analysis included
general cultural practices in the areas of fertilization, water, pest, and basic crop
management, and specific practices related to the application of particular kinds of pesticides.
Handling practices at the processing and distribution stages of the market channel include
cleaning, waxing, sorting, grading, storage, ripening, and packing or packaging activities.
This investigation did not encompass the collection or analysis of costs and returns of
production, processing, or distribution.

Organization of Report

This report is organized into seven sections including this one. The second and
following section is entitled Procedures and Data. It reviews the procedures used to
achieve the previously stated research objectives and provides some background on the data
sources and collection process. The Conceptual Framework is developed in the third
section and is later used to specify the theoretical and empirical models. A descriptive
examination of the data is provided in the forth section entitled, Pesticide Residues in
Florida Strawberries and Tomatoes and Characteristics of the Industries. The fifth
section, entitled Modeling and Empirical Analysis, documents the development and
implementation of the empirical models. The outcome of the empirical analysis is discussed
in the sixth section, Empirical Results. From the empirical findings, implications,
conclusions, and recommendations are derived and presented in the seventh and last section
entitled, Summary, Conclusions and Recommendations. An assessment of the overall
success of the project is included in this section with suggestions for possible future
methodological improvements.

Procedures and Data

The objectives of this project were attained in essentially five stages with considerable
overlap between stages. First, background research was conducted into the relative economic
and dietary importance of various fruits and vegetables grown in the state of Florida. The
nature and extent of pesticide residues found in these fruits and vegetables was explored.
Also the current technology of pesticide residue detection in food products was reviewed.
Based on this preliminary research, tomatoes and strawberries were chosen as the two
commodities to analyze for this investigation. Once this decision was made, general
information on the cultural and handling practices commonly used in tomato and strawberry
production, processing and distribution were gathered and compiled.

Developing a conceptual framework and postulating sets of testable hypotheses
comprised the second stage of this study. Based on information obtained from background
research, numerous potential relationships between pest infestations, pest control, pesticide
residues, and various aspects of agronomic production, processing, and distribution were
formulated or hypothesized. A conceptual framework was then devised to structure these
relationships and provide a theoretical foundation from which empirical models could be
specified. This framework of hypothesized relationships was then articulated into a set of

survey questionnaires. The questionnaires were employed to collect data on the cultural and
handling practices used in producing, processing and distributing tomatoes and strawberries,
as well as the socio-economic attributes of growers, packers and distributors.

In the third stage of the project, data necessary for the empirical analysis were acquired
from primary and secondary sources. Data on pesticide residues found in strawberries and
tomatoes were obtained from the Florida Department of Agricultural and Consumer Services
(FDACS), Division of Chemistry, Chemical Residue Laboratory (CRL). This laboratory is
responsible for the chemical analysis of poisonous or deleterious chemical residues
remaining in or on human food and animal feed that is produced or marketed in Florida.
Field agents for CRL collect samples of fresh and processed agricultural products from
around the state at various stages of the marketing channel. These are rushed to CRL
laboratories where they are checked for chlorinated hydrocarbon, organophosphate and
carbamate pesticide residues using multi-residue screening procedures. If any of these types
of pesticide residues are detected by the screenings, then appropriate quantitative chemical
assays are performed. Chemical analysis of most samples is completed within 48 hours after
collection. In order to maximize the effectiveness of its monitoring effort, CRL sample
collection is weighted more heavily toward those commodities and growing seasons which
have a greater potential for accumulated residues. Consequently, the sampling procedures
used by the CRL are not entirely random.

The residue analysis results of 307 tomato and 113 strawberry samples were acquired
from CRL and converted to a micro-computer database format. These samples provided a
list of potential interviewees from whom corresponding cultural and handling practices data
could be collected. Survey questionnaires for each market stage were developed to elicit the
characteristics of, and production and handling practices used by the sampled growers,
packers, and distributors. A review of previous National Agricultural Statistics Service
(NASS) Chemical Use Survey questionnaires was useful for this purpose. The enumeration
of this survey was carried out by the Florida Agricultural Statistics Service (FASS) under a
contractual arrangement. FASS personnel also contributed substantially to the development
of the questionnaires. A total of 338, or 80 percent, of a possible 420 interviews were
successfully completed. Weather data consisting of monthly observations for rainfall and
average temperature at 34 locations within the state of Florida were obtained from the
National Weather Service. These locations were chosen for their proximity to the FDACS
sample collection sites.

The fourth stage of the project consisted of reviewing and editing the residue data
obtained from CRL, and the practices and attributes data collected through FASS. A number
of problems with the accuracy and completeness of the data were encountered. These
included: incomplete recording of residue sampling data that was considered important for
accurate field or lot identification at the grower and packer stages; inconsistent coding of
respondent's refusals to answer specific survey questions; inconsistent responses among
related survey questions; and considerable variability in compliance to different questions in
the survey.
The final stage of the project involved measuring and evaluating the statistical
relationships among the data in order to substantiate or refute the stated hypotheses. Results

from the survey indicated that pesticides were only being applied to tomatoes and
strawberries at the grower level. Since there was generally no coordination or trace-back of
residue testing by CRL between market stages, it was impossible to determine how much
residues were in or on tested samples before they arrived at the packing or distribution stages.
Consequently, empirical models of pesticide residues found at the packing and distribution
stages were not developed. For the grower stage, three different empirical models were
specified for each commodity. These are identified as (1) firm and decision-maker attributes,
(2) general cultural practices, and (3) specific cultural practices. Regression analysis was
carried out on each model using ordinary least squares and principal components procedures.
Separate regressions were performed for various levels of aggregation of pesticide residues
including: all residues, fungicide residues, insecticide residues, and residues of specific active
ingredients. A total of 27 regressions were implemented in all.

Once the statistical analysis for the grower data was completed, implications were
articulated, and conclusions developed. Recommendations for risk assessment, pesticide
regulation policy, and changes in production and handling practices were then formulated.
Finally, the overall success of this research effort was evaluated and recommendations for
future methodological improvements were suggested. A number of related problems which
could be opportunities for further research were also mentioned .

Conceptual Framework

From a technical standpoint, the occurrence of pesticide residues in fresh produce is
just one element of a set of cultural outcomes which are simultaneously determined by the
cultural practices (including inputs), environmental factors, and pest pressures associated
with or necessary to their production and processing. These outcomes include the quantity of
output, various qualitative aspects of output (including levels pesticide residues), and any
externalities impacting the environment and/or other economic agents. In a typical
agronomic or economic analysis, the technical relationships between factors of production
and these outcomes would be modeled with some type of mathematical production function.
Such functions typically designate continuous quantities of inputs and relevant environmental
factors as determinants of output and other cultural outcomes. Unless assumed otherwise,
these functions would have parameters and a functional form to accommodate economies of
scale or size. Also, since some agricultural production inputs, like pesticides, may influence
the variability as well as the level of output, it would be appropriate for such a production
function to incorporate the statistical properties of output as well (Just and Pope, 1979).

Due to a variety of inadequacies, it was determined that the traditional production
function modeling approach was impractical for this study. Foremost, the quantity and
quality of data needed to empirically operationalize a rigorously specified production
function would be much greater than that which producers and handlers could be reasonably
expected to furnish voluntarily. In addition, the traditional production function does not
typically address the socio-economic or industrial organization dimensions of the pesticide
residue problem. As a result of these problems, a more descriptive approach was developed.

The development of a conceptual framework for this analysis is based on the
classification of various potentially influential inputs, conditions, and practices according to
their inherent nature, mode of causality, and specificity of influence. These potentially
influential variables can be classified into three different categories by their intrinsic nature.
These are Practices, Environmental Factors, and Attributes. Practices are defined here as
deliberate actions or strategies undertaken by growers or handlers in order to produce,
process, or distribute the commodity. Environmental Factors are various weather conditions
that may impact pest pressures and pesticides in a variety of ways. Factors such as
temperature and rainfall differ from Practices in that they are beyond the control of the
decision-maker. They are also distinctive in that they may interact with a variety of practices
to initiate, preclude, or alter their influence on residues. Socio-economic characteristics of
the decision-maker and organizational characteristics of the firm are defined here as
Attributes. Attributes may be associated with or influence the selection of particular
practices which may then impact pesticide residues. Thus Attributes are distinguished by
having a consistently indirect affect on residues. Despite this indirect relationship, these
variables may be valuable for identifying the characteristics of firms and decision-makers
that are associated with residues. This knowledge can be used to identify particular segments
of the industry for further study and/or for targeting future policy directives.

Another important criterion by which agronomic practices can be classified is the
specificity of their influence on residues. In other words, does a particular practice influence
residue levels of an individual pesticide, pesticides of a certain type, or all pesticides?
Clearly, practices which only involve the application of a specific chemical pesticide should
be related solely to the occurrence of that particular chemical residue. For the purpose of this
analysis, practices with this type of association with residues are referred to as Specific. In
complementary fashion, practices which directly or indirectly effect residues of more than
one specific pesticide are designated as General Practices. Examples of General Practices
would include most activities associated with soil fertility or water management.
Environmental Factors and Attributes would also be classified as general because of their
potential impact on residue levels of many different pesticides.

The selection of Attributes, General Practices, Specific Practices, and Environmental
Factors that might explain pesticide residues in tomatoes and strawberries was based on
research into the associated production technologies and their agronomic relationships to
pests and pesticide residues for these commodities. The selected variables were classified
into one of these four categories as follows:

Attributes: Decision-maker role, age, experience, and education;
pesticide applicator certification; firm size, land rent-
ownership mix, objectives, organization, affiliation; soil type
and location.

General Practices: Type of irrigation and freeze protection used; use of plastic
mulch and crop rotation; length of growing season; plant
density and variety; types of fertilizer used; use of soil-testing
and plant tissue analysis; the application of fertilizers relative

to soil and tissue analysis recommendations; scouting for
pests; use of economic thresholds, beneficial, bio-control,
pheromones, irrigation practices, planting dates, mechanical
cultivation, alternating pesticides, soil testing for pests. and
sprayer calibration in pest management.

Specific Practices: Product applied, target pest, level of infestation; timing or
frequency of applications; number of applications;
application rate and method; pesticide product form; use of
surfactants, stickers and non-adjuvants; and decision factors.

Environmental Factors: Temperature and rainfall during growing and harvesting
periods, soil type, and location.

Specific and General Practices along with Environmental Factors may directly
influence the occurrence of residues in tomatoes or strawberries in one or more of three
ways or modes of action. First, any given practice or environmental factor may act to
encourage or discourage the development of pest infestations, possibly impacting
different pests or types of pests in different ways. For instance, over-fertilization of tomatoes
or strawberries may make them more attractive to insect pests or susceptible to fungal
diseases, thus requiring more applications of pesticides which could lead to higher residues.
In subsequent paragraphs this will be referred to as mode one (1). Second, some cultural
practices and environmental factors may substitute for or influence the effectiveness of
pesticides used in crop protection. If such practices or environmental factors resulted in
lower pesticide use, then a negative relationship would exist between their use and residue
levels. For example, the use of pheromones to monitor insect infestations may reduce
unnecessary spray applications while pheromones used as a control agent could substitute for
insecticides by interfering with insect reproductive cycles. Of course other types of practices
and environmental variables may diminish the effectiveness of pesticides, leading to more
frequent or higher rate pesticide applications. In subsequent paragraphs this type of
complementary or substitute action will be referred to as mode two (2). Third. various
practices and environmental factors may influence the fate of pesticides after they have
been applied to the crop. For example certain additives to pesticide spray mixes may retard
or accelerate the breakdown of the pesticide's active ingredient. At another level, post-
application overhead irrigation may wash off external residues. In subsequent paragraphs
this will be referred to as mode three (3). With respect to Attributes, none of these three
previous modes of action apply directly. As previously discussed, attributes are presumed to
influence the decision-maker's selection of general and specific practices which in rum
operate through one or more of the above modes to affect residues.

A wide assortment of inter-relationships may exist within and between general
practices, specific practices, environmental factors, and attributes as defined above. Thus
the paths and modes of causality between these variables and pesticide residues are not all
direct or independent. Under such circumstances, it would be inappropriate to specify
residues as a direct function of all these variables within a single equation. Consequently, a

hierarchical framework was developed to model the relationships described in the previous

While it is clear that residues are affected directly by specific practices and
environmental factors as defined above, i.e.,

directly Specific Practices and
(1) Residues = f f
S Environmental FactorsJ

at the same time, the intensity of these pesticide applications will be determined in part by
the general practices chosen by producers as well as environmental factors, that is,

) Specific f General Practices and
Practices Environmental Factors

In turn, choices between different general practices may be influenced by various attributes
related to the individual making the decisions or the firm to which they belong, in addition to
environmental factors, or

General fAttributes and
(3) Practices = Environmental Factors

It is also feasible that some general cultural practices can directly influence residue levels.
Notably, the type of irrigation used, the plant density levels chosen, the frequency of sprayer
calibration, and alternating between different pesticides, among other practices, while not
exclusively related to pest management, may have both direct and indirect effects on
pesticide residues in conjunction with environmental factors, i.e.,

{ General Practices and
(4) Residues = Environmental Factors)

Thus a hierarchical framework of causality has been derived and can be used to guide the
specification of the empirical models. This framework can be represented symbolically as
equation (5) or graphically as shown in Figure 1.

RSpecific n rac Practices = f(eefAttributes & Environmeal Factors)
Residues = f General Practices = f Attributes & Environmental Factors)
and Environmental Factors

Figure 1. Direction and Source of Causality Between Firm or Decision-maker Attributes,
General Practices, Specific Pesticide Practices, and Pesticide Residues (Equation 5).

Equations (1) through (5) imply that the model for pesticide residues should be
specified as a simultaneous system. Unfortunately the difficulties of operationalizing such a
system were beyond the scope and data resources of this study. Consequently, specification
of the empirical models was simplified so that only the ultimate impact of each class or type
of variables on residues was estimated. Since some types of survey data were found to be
less reliable than others, this approach offers some consolation in that the best data can be
analyzed without being tainted by the worst.

Pesticide Residues in Florida Strawberries and Tomatoes

While there are literally hundreds of different pesticide products and product
formulations labeled for use on strawberries and tomatoes, many are of little or no concern
with respect to food safety. Many pesticides do not leave residues because they are not
applied directly to the plant or fruit, or they are so volatile that they breakdown or dissipate
long before harvest and consumption. The residues detected in the samples used for this
analysis were confined to essentially 19 different pesticides for both commodities, or 13 each
for strawberries and tomatoes individually. A listing of these pesticides, including their EPA
tolerance levels, is provided in Table 1.

Table 1. Pesticide Residues Found in Florida Department of Agriculture and Consumer
Services, Chemical Residue Laboratory Tomato and Strawberry Samples, 1990 1993.

EPA Strawberry Tomato
Title 40 Pest Found tolerance tolerance
EPA name article No.' Trade name type 2 in 3 ppm. ppm.
Acephate 180-108 Orthene I T 0.00 0.00
Captan 180-103 Captan F S 25.00 25.00
Carbaryl 180-169 Sevin I S 10.00 10.00
Chlorothalonil 180-275 Bravo / Daconil F S&T 0.00 5.00
Chlorpyrifos 180-342 Lorsban I T 0.50 0.50
DCPA 180-185 Dacthal H S&T 2.00 1.00
Diazinon 180-153 Diazinon I S&T 0.50 0.75
Dicofol 180-163 Kelthane M S 5.00 5.00
Endosulfan 180-182 Thiodan I/M S&T 2.00 2.00
Ethion 180-173 Ethion I/M T 2.00 2.00
Iprodione 180-399 Rovral F S 15.00 0.00
Malathion 180-111 Cythion I S 8.00 8.00
Maneb 180-110 Manzate/Dithane F T 0.00 5.00
Methamidophos 180-315 Monitor I/M T 0.00 1.00
Methomyl 180-253 Lannate I S&T 2.00 1.00
Mevinphos 180-157 Phosdrin I/M S 1.00 0.20
Parathion 180-121 Parathion (generic) I/M T 1.00 1.00
Permethrin (B) 180-378 Ambush/Pounce I S&T 0.00 2.00
Vinclozolin 180-380 Ronilan F S&T 10.00 3.00

Code of Federal Regulations, Title 40, "Protection of Environment." Chapter 1, Subchapter E, Part 180.
Office of the Federal Register, Washington D.C.
2M = Miticide, F = Fungicide, H = Herbicide, I = Insecticide.
S = Strawberries, T = Tomatoes.

Referring to Table 1, the vast majority of pesticide residues detected in Florida
tomatoes and strawberries are insecticides, miticides, and fungicides. Overall, there were 13
different insecticide-miticide compounds detected and five different fungicides. One type of
herbicide residue was found in just two of the 338 samples for which interviews were
conducted. No nematocides or bactericides were detected. EPA tolerances for pesticides
vary between strawberries and tomatoes. In several instances, a particular pesticide has no
tolerance level for one commodity but is approved for use on the other. This variation
between commodities is often due to allocations made by EPA based on estimates of total
dietary intake or quantitative risk assessments of a given pesticide residue from the
consumption of many different foods over a lifetime. Such allocations are sometimes based
on the relative importance or average level of consumption of each commodity. In some
cases the importance or benefit of a particular pesticide in protecting a given commodity
from pest infestations may be considered. The variation in tolerance levels between
different pesticides is in large part due to differences in their toxicologic, oncogenic and
mutagenic properties.

Descriptive statistics on aggregated types of residues found in strawberry and tomato
samples for which interviews were successfully completed are shown in Tables 2 and 3.
Generally, there is no coordination between produce sampled at different market stages.
Occasionally, if a problem is first discovered at the packer or distributor stage, it may be
traced back up the market channel to the producer. When samples are found to have residues
at levels above EPA tolerances, harvesting is suspended or the lot is held in storage until
confirmatory sampling can be done within three to five days. Samples taken by FDACS for
this purpose were not used to calculate descriptive statistics for residues. Statistics provided
for each type of pesticide include: the proportion of samples found to have no detectable
residues, the proportion found to have a residue level exceeding its EPA tolerance, the
minimum and maximum parts per million found for each residue, the mean, standard error,
median, standard deviation and coefficient of variation for each type of residue. It should be
noted that relatively small sample sizes for strawberries at the packer and distribution market
stages resulted in large standard errors for the means as shown in the last column of Table 2.

By examining Tables 2 and 3, and Figure 2 below, some inferences about the nature
of pesticide residues and pesticide use in strawberries and tomatoes can be deduced.
Looking at Table 2 and Figure 2, it can be seen that fungicides are the dominant type of
pesticide residue found in Florida strawberries. At the grower stage less than five percent of
the samples contained no detectable residues of fungicide, with a mean level of 4.92 parts per
million (ppm). Roughly similar statistics for fungicide residues are found for strawberries at
the other two stages, although the sample sizes for these are much smaller. In comparison to
fungicides, insecticide residues were found in approximately 38 percent of strawberry
samples from the grower stage at an average level of 0.2388 ppm. Descriptive statistics for
the pesticide residues found in tomatoes are shown in Table 3. In contrast to strawberries.
the predominant type of residues found in tomatoes are insecticides and miticides. Overall.
tomatoes carry a considerably lower load of pesticide residues than do strawberries.

Table 2. Descriptive Statistics of Pesticide Residues Found in Samples of Florida Strawberries, Collected between October, 1990 and
June, 1993 at the Grower, Packer and Distributor Market Stages.

Proportion of Proportion
samples with of samples
Market Type of Sample no residue with residue Min. Max. Mean Standard
stage pesticide size detected > tolerance ppm. ppm. ppm.2 error
Fungicides 47 4.26% 2.13% 0 21.00 4.9203 0.8080
Insecticides 47 61.70% 2.13% 0 3.10 0.2388 0.0915
Pesticides 47 2.13% 4.26% 0 21.00 5.5483 0.8653
Fungicides 17 10.53% 0% 0 13.72 3.4318 0.9277
Insecticides 17 63.16% 0% 0 5.15 0.4758 0.2704
Pesticides 17 10.53% 0% 0 14.67 3.9076 1.0345
Fungicides 9 0% 0% 0.005 17.81 9.0400 3.0154
Insecticides 9 71.43% 0% 0 0.54 0.1171 0.0808
Pesticides 9 0% 0% 0.005 17.81 10.6371 2.6743

SFor type of pesticide: Pesticides represent the aggregate of fungicides and insecticides.
2 1 n .n
SThe mean of aggregate residues of pesticide type i are computed as r, =- rk with k = 1 ... n observations or
samples andj = 1 ... m individual pesticides of type i. T'he mean of all pesticide residues is similarly calculated.

Table 3. Descriptive Statistics of Pesticide Residues Found in Samples of Florida Tomatoes, Collected between October, 1990 and
June, 1993 at the Grower, Packer and Distributor Market Stages.

Proportion of Proportion
samples with of samples
Market 1Type of Sample no residue with residue Min. Max. Mean Standard
stage pesticide size detected > tolerance ppm. ppm. ppm. error
Fungicides 87 66.67% 0% 0 1.10 0.0502 0.0188
Insecticides 87 31.03% 27.59% 0 1.01 0.1877 0.0221
Pesticides 87 14.94% 27.59% 0 1.24 0.2379 0.0271
Fungicides 122 81.15% 0% 0 0.67 0.0207 0.0075
Insecticides 122 52.46% 5.74% 0 1.00 0.0813 0.0160
Pesticides 122 35.25% 5.74% 0 1.00 0.1021 0.0170
Fungicides 39 92.31% 0% 0 0.12 0.0033 0.0031
Insecticides 39 64.10% 2.56% 0 0.64 0.0521 0.0208
Pesticides 39 58.97% 2.56% 0 0.64 0.0554 0.0209

SFor type of pesticide: Pesticides represent the aggregate of fungicides and insecticides.
2 n ues "icide type e compwith k
'The mean of aggregate residues of pesticide type i are computed as r, - r,,, with k
A 1 i l

1 ... n observations or

samples and j 1 ... m individual pesticides of type i. The mean of all pesticide residues is similarly calculated.

Figure 2. Percent of Strawberry and Tomato Samples with no Detectable Residues of Different Types of Pesticides at the Grower,
Packer and Distributor Market Stages, Collected between October, 1990 and June, 1993

100% 1

90% 1---_,, 100%
80%- .

70% --- T- -


0%- 20%

D I,-- 0%


The statistics for aggregate residues found in tomatoes at three market stages shows
that both fungicide and insecticide/miticide residues diminish as the product moves through
the market channel. The average level of fungicides found in tomatoes at the distributor
stage are less than a tenth of those from the grower stage. Insecticide residues at the
distributor stage are reduced to almost one fourth the levels found at the grower stage. A
complimentary trend, though not as strong, is apparent for the frequency of occurrence of
tomatoes found with no detectable residues. T-tests in Table 4 confirm these inferences,
with highly significant differences found between average residues at the grower and packer
stages, and grower and distributor stages. The reduction in residues at the packing stage is
most likely due to the washing and rinsing processes performed at this stage. The fact that
tomatoes are optimally stored at relatively higher temperatures than strawberries (near 60
degrees Fahrenheit compared to slightly above freezing) could account for the continued
decline in residue levels through the distribution stage.

Although the number of observations for strawberries at the packer and distributor
market stages is small, pesticide residues, particularly fungicides, were found to occur at
significantly higher levels at these downstream stages. Since most strawberries are packed
into retail containers at harvest and subsequently refrigerated at temperatures slightly above
freezing, there is apparently little opportunity for residues to be removed or break down in
this fruit between harvest and retail merchandising. Survey results found no packers or
distributors applying pesticides to strawberries or tomatoes. While it is not impossible that
residues could concentrate in strawberries during packing and distribution, it is strongly
suspected that if tests were conducted using larger more randomly selected samples, then
this relationship would disappear.

Table 4. T-tests and P-values for Statistical Differences in Average Pesticide Residue
Levels between Market Stages for Strawberries and Tomatoes (2 tailed test). '

Strawberries Tomatoes
Fungicide Insecticide Pesticide Fungicide Insecticide Pesticide
residues residues residues residues residues residues
Grower- Test -1.0605 1.0622 -1.0836 -1.6282 -4.0049 -4.4537
Packer P-value 0.2940 0.2938 0.2836 0.1072 0.0001 <0.0000

Packer- Test 2.3896 -0.7904 2.8884 -1.2981 -0.9541 -1.4464
Distributor P-value 0.0268 0.4402 0.0088 0.1967 0.3419 0.1507

Grower Test 1.7315 -0.5054 2.0675 -1.6659 -3.7776 -4.2520
Distributor P-value 0.0895 0.6158 0.0438 0.0994 0.0003 0.0001

A negative test value indicates that the average downstream market stage pesticide residue level was less
than the upstream market stage (t = downstream upstream).

Modeling and Empirical Analysis

The specification of the empirical models was inspired and guided by the conceptual
framework and theoretical relationships described in equations (1) through (5) and Figure 1.
A number of simplifying modifications were necessary in order to accommodate a variety of
data, statistical, and modeling problems. Details of this specification process are discussed

Basic Model Specification

Because of operational- and data-related problems, none of the intermediate models
or relationships as described in equations (2) or (3) were implemented for the empirical
analysis. The theoretical models described in the previous sections were simplified by
specifying residues as three independent functions of the right hand sides of equations (1),
(2), and (3). These three models were specified with linear functional forms and estimated
using OLS and principal component regression procedures. Final specifications varied
somewhat between commodities and residue types due to data and implementation problems
which are discussed later. The first model relates residue levels directly to attributes and
environmental factors and is referred to as the Attributes model. It can be represented in
general terms as;

(6a) Residues = f{Attributes and Environmental Factors}.'

or with specific variables as,

Decision-maker Role, Education, Age, Experience;
(6b) Residues = f Pesticide Certification; Firm Size, Acres Rent/Own, Objectives,
Organization, Affiliation; Soil Type, Temperature, and Rainfall.

The second model captures the relationship between general cultural practices and residues.
It is referred to as the General Practices model and in general terms looks like,

(7a) Residues = f{General Cultural Practices and Environmental Factors }

or in terms of specific variables as

Fertilizer Type, Soil Testing, Plant Tissue Analysis,
Follow Test Recommendations, Scouting, Thresholds,
Beneficials, Bio control, Pheromones, Soil Test for Pests,
(7b) Residues = f Calibration, Irrigation Practices, Cultivation,
Alternate Pesticides, Irrigation Types, Plastic,
Freeze Protection, Rotation, Season length, Density,
Variety, Temperature, and Rainfall

The third model captures the relationship between specific cultural practices and associated
residues. In general terms the Specific Practices model is:

(8a) Residues = f Specific Pesticide Practices and Environmental Factors ,

or substituting specific variables,

Target, Infestation Level, Rate, Pre harvest Interval.
Times, Spray Interval, Form, Method, Non-adjuvant.
(8b) Residues = f<
(8b) Residues Surfactant, Sticker, Post Overhead Irrigation,

Decision Type, Temperature, and Rainfall

All models were specified with a linear functional form and intercept term, but equation (8b)
required some transformations to account for the non-linear interaction of rate and time on
pesticide residues. Equation (9) was used to transform the application rate. number of times
sprayed, pre-harvest interval and regular spray intervals to a single weighted component of
active pesticide ingredient applied (see Appendix A for derivation).

n-' 1
(9) Qt QaZ p+ki where
k=0 2 h

Qt = quantity of pesticide theoretically remaining on the commodity at time t (its
sampling date);
Qa = the uniform application rate for a reported spray regimen. (Data on pesticide
applications were collected on a spray regimen basis, so that rates and
formulations for a given series of applications were identical;)
n = the total number of pesticide applications;
k = indexes pesticide applications in reverse order of how they were applied;
That is k = 0 represents the last application and k = n 1 represents the first:
p = the pre-harvest interval (i.e., the number of days between the last pesticide
application and the sampling date of the commodity);
i = the number of days between regular pesticide applications;
h = estimated half life of chemical pesticide in days (Hubbell and Carlson, 1993).

Thus equation (8b) can be respecified as,

Target, Infestation Level, Weighted Rate,
Form, Method, Non adjuvant, Surfactant,
(8c) Residues = f
Sticker, Post Overhead Irrigation,

Decision type, Temperature, and Rainfall

The final specification of each model was based on research into the production
technology and their agronomic relationships with pests and pesticides. As discussed above,
it was often concluded that more than one mode of action could be at work between residues
and a particular attribute, practice, or environmental factor. Also any given practice may
have different impacts on different types of pests and the pesticides used to control them.
Thus in many cases, it was not possible to hypothesize a definitive sign or direction to a
relationship. A brief accounting of the modes of action and expected signs of all the
hypothesized relationships for the three models is provided in Table 5. Where appropriate,
the modes of action are coded numerically as was described in the Conceptual Framework
section (i.e., either 1, 2, or 3). The expected signs are either negative (-), positive (+), or
ambiguous (). Indirect modes of action, as described with respect to attributes, are
indicated with an arrow (-) followed by the relevant code number. A brief description of
the empirical representation of each variable is given in Table 6.

Aggregation of Left and Right Hand Side Variables

Attribute and General Practice models were implemented with explanatory variable
data aggregated and disaggregated across commodities. No individual growers were
sampled for both tomatoes and strawberries. Preliminary models using aggregated
strawberry and tomato data did not perform as well as those for each commodity alone. The
use of intercept and slope shifting dummy variables as a way around this was considered,
but found to be impractical given the loss in degrees of freedom and the added difficulty of

Each of the three models were regressed on residues aggregated at two of three
different levels. Both Attribute and General Practice models were regressed against the
unweighted sum of all types of residues and then on residues aggregated as either
insecticides or fungicides (there were only three grower samples containing herbicide
residues). The Specific Practices models where specified for each particular generic
chemical residue detected and then for residues aggregated as either insecticides or

Data-Related Regression Problems

There were a small proportion of survey questions which experienced significant
rates of refusal by interviewees. These included: firm objectives, planting and harvest dates,
crop yields and grades, target pests, levels of infestation, pesticide application dates, and
pesticide application rates. Generally, the more detailed or personal the questions, the
higher the refusal rate. A respondent's refusal to answer a specific question in an interview
creates missing values for that observation in the data set. To avoid losing a whole
observation, these missing values were amended using the zero-order method. This involves
replacing each missing value with the average of all non-missing observations or responses
for that particular variable or question. Provided there is no self-selection bias, this is a
statistically neutral and valid means of mitigating this problem. In some cases, a high
refusal rate contributed to a decision to drop certain variables from the analysis altogether.

Table 5. Hypothesized Relationships and Coefficient Signs for Attributes, General
Practices, and Specific Practices Model Variables.

Explanatory Variable Mode Sign Comment
Attributes Model
Decision-maker role -*1,2,3 + principal agent effects, owner verses manager
Education ->1,2,3 education => use of IPM =:> 4" residues
Experience -*1,2,3 managerial experience => T use of IPM => pesticides
Certification -+2,3 pesticide training = 4 mis-use of pesticides = residues
Firm size -->1,2,3 size = in quality of pest management => V pesticides
Rent vs. Own -41,2,3 + equity risk exposure, stability, planning horizon?
Objectives -41,2,3 + high yield objective -> pesticides residues
Affiliation ->1,2,3 4-risk and adverse selection => pesticides => residues
Soil type 1 + exploratory, no hypothesis
Location 1 + further south => T pests?
General Practices Model
Fertilizer types 1 + exploratory, no hypothesis
Soil-testing frequency 1 fertility manage. => 1 pest pressure = 4I pesticides
Plant tissue analysis 1 1 fertility manag. => pest pressure => 4 pesticides
Fert. recommendations 1 1 fertility manage. = 4 pest pressure = 4- pesticides
Scouting 2 monitoring => pest control efficiency = + pesticides
Economic thresholds 2 more effective/efficient use of pesticides r pesticides
Beneficials 1,2 substitutes for pesticides => pesticides > residues
Bio-control 1,2 substitutes for pesticides => pesticides = residues
Pheromones 1,2 substitutes for or T pesticide efficiency : residues
Irrigation practices 1,2 pest pressure & T pesticide efficiency = pesticides -
Adjusting planting dates 1,2 pest pressure & T pesticide efficiency => pesticides
Mechanical cultivation 1.2 exploratory, no hypothesis
Alternate pesticides 2 buildup of pest resistance = 4. pesticides => T residues
Soil test for pests 2 + 1 intensity of pest control => T pesticides i> ? residues
Sprayer calibration freq. 2 more accurate & lower applications rates => pesticides
Type of irrigation 1,2,3 + those that wet plants => pest pressures => pesticides
Plastic mulch 1,2 pest pressure r = pesticides => residues
Crop rotation 1 pest pressure = 4 pesticides > residues
Freeze protection 1,3 if overhead, washes off residues => residues
Season length 1,3 + length -> pests=> I pesticides =, ? residues
Plant distances / density 1.2,3 distance between plants = 41 pests 4, pesticides
Plant variety 1 + exploratory, no hypothesis

Ssee the Conceptual Framework section for discussion of coding.
" meaning of symbols and numbers under this heading is as follows:
(-) indicates an indirect action through one of the following modes.
(1) a variable which may encourage or discourage the development of pest infestations, possibly
impacting different pests or types of pests in different ways.
(2) a variable that may replace, enhance, or reduce the effectiveness of residue producing pesticides used
in crop protection.
(3) a variable that may interact with practices and the environment to influence the fate of pesticides once
they have been applied.
3 the preceding effect implies that residues should change in the same direction.

Table 5 (continued). Hypothesized Relationships and Coefficient Signs for Attributes,
General Practices, and Specific Practices Model Variables.

Explanatory Variable Mode Sign Comment
Specific Practices Model
Target pest 2 exploratory
Level of infestation 2 + 1 infestation => T pesticide rates => residues
Timing frequency 2,3 + I frequency > 4 spray intervals => residues
Number of applications 2,3 + applications = I pesticides => T residues
Application rate 2,3 + 1 rate => pesticides => T residues
Application method 2,3 + banding = 4- pesticide use, but not necessarily, on plants
Surfactants 2,3 efficiency of pesticides => 4 pesticides = 4- residues
Stickers 2,3 + makes pesticides adhere to crop => residues
Non-adjuvants 2,3 + exploratory
Decision factor 2 econ. threshold = 1T efficiency of control => 1 pesticides
Environmental Variables
Avg. temperature (current) 1,2,3 temperatures = faster breakdown of residue
Rainfall (current) 1,2,3 T rainfall can wash off pesticide => residues
Avg. temperature(lagged) 1,2,3 temperatures s 1 pests = 1 pesticides => residues
Rainfall (lagged) 1,2,3 + rainfall = 1 pests (fungus) > T pesticides = residues

see the Conceptual Framework section for discussion of coding.
2 meaning of symbols and numbers under this heading is as follows:
(--) indicates an indirect action through one of the following modes:
(1) a variable which may encourage or discourage the development of pest infestations, possibly
impacting different pests or types of pests in different ways.
(2) a variable that may replace, enhance, or reduce the effectiveness of residue producing pesticides used
in crop protection.
(3) a variable that may interact with practices and the environment to influence the fate of pesticides once
they have been applied.
3 the preceding effect implies that residues should change in the same direction.

Table 6. Description of Numerical Representation of Variables Used in Attribute, General
Practices, and Specific Practices Empirical Models.

Explanatory Variables Numerical Type Description
Attributes Models
Decision-maker role Discrete choice I=owner. partner. 3-manager
Education Discrete choice 1 6. for increasing levels of education
Experience Continuous years
Certification (sum) Sum of binaries 1 3. for certification of different employees
Firm size Continuous acres
Rent vs. Own Continuous acres rented + total acres
Affiliation Binary yes = 1
Soil type M.E. set of binary sand, loam, hydroponic, sandy loam.
General Practices Models
Fertilizer types Set of binary dry, liquid, foliar. fertigation. other
Soil-testing frequency Discrete choice less than, equal to, or more than once per year
Plant tissue analysis Binary yes = 1
Fert. recommendations Discrete choice 1=below, 2=at, or 3=above recommendations
Scouting (in-house) Binary scouting was done by self or employee = 1
Economic thresholds Binary yes = 1
Beneficials Binary yes = 1
Bio-control Binary yes = 1
Pheromones Binary yes = 1
Irrigation practices Binary yes = 1
Adjusting planting dates Binary yes = 1
Mechanical cultivation Binary yes = 1
Alternate pesticides Binary yes = 1
Soil test for pests Binary yes = 1
Sprayer calibration freq. Discrete choice l=weekly, 2-monthly, 3=season, 4=annual.
Type of irrigation Set of binary drip, micro-jet, seep. overhead, hydroponic
Plastic mulch Binary yes = 1
Crop rotation Binary yes = 1
Freeze protection Set of binary overhead, micro-jet, seep. covers, heaters
Season length Continuous days, test date minus plant date
Plant distances/density Continuous inches
Plant variety M.E. set of binary' 9 tomato and 6 strawberry varieties

"M.E. set of binary" stands for a Mutually Exclusive set of binary variables.

Table 6 (continued). Description of Numerical Representation of Variables Used in
Attribute, General Practices, and Specific Practices Empirical Models.

Explanatory Variables Numerical Type Description
Specific Practices Models
Target pest Discrete choice 20 different discrete choices
Level of infestation Discrete choice I=light, 2=moderate, 3=heavy
Timing frequency Continuous period of application no. of applications
Number of applications Continuous 1 through 31
Application rate Continuous pounds of active ingredient applied per acre
Application method Discrete choice banded=l, broadcast
Surfactants Binary yes=l
Stickers Binary yes=l
Non-adjuvants Binary yes=l
Decision factor Binary yes=l
Environmental variables
Avg. temp. (current) Continuous average degrees, harvest month
Rainfall (current) Continuous cumulative inches, harvest month
Avg. temp. (lagged) ((lagged) Continuous average degrees, month before harvest
Rainfall (lagged) Continuous cumulative inches, month before harvest

S"M.E. set of binary" stands for a Mutually Exclusive set of binary variables.

This was the case for firm objectives in the Attributes model, and target pest and level of
infestation in the Specific Practices models. Statistics on the number and proportion of
refusals for each survey question or variable in the analysis are summarized in Table 7.

There were a number of other data/numerical problems which made it necessary to
drop variables from the three models. Linear dependencies in the General Practices model
were the most troubling. This was probably due to the relatively large number of binary or
discrete choice explanatory variables in this model. Deleting at least one of the suspect
variables was the only means available to remedy this kind of difficulty. Some
dependencies involved more than six variables and could never be completely identified.

Multicollinearity was also a problem in most regressions. Condition numbers were
calculated for each regression and ranged from 43.9 in the Specific Practices model for
Chlorothalonil, to 730.29 in the General Practices model for strawberries. Serious
multicollinearity occurred between a number of binary response variables (with low
variance) and the intercept term. For example, more than 94 percent of all growers indicated
that they generally used economic thresholds for making insect and fungus pest control
decisions. Since more detailed information on this issue was available in the Specific
Practices section of the interviews, it was decided to drop these variables from the General
Practices models. In the Attributes models, variables which were highly collinear included:
(1) acres owned and rented, and gross revenue; (2) firm form and affiliation; (3) different
levels of pesticide certification; (4) various components of the weather variables, and; (5)
location variables with a wide array of other variables such as firm size and soil type. In the
General Practices models, soil-testing frequency and following test recommendations were
highly collinear, and so were comparable variables for plant tissue analysis.

One means of mitigating the consequences of multicollinearity is through the
application of principal components techniques. Principal components are a set of mutually
orthogonal vectors composed from linear combinations of the original regressors. They are
derived sequentially in order of decreasing variability. Consequently, the last few principal
components will represent only a small fraction of the variability of the original data, or
conversely, a significant proportion of the multicollinearity in the data. Thus it is often
possible to drop these last few principal components from the model and substantially
improve the power of the estimated coefficients. Detailed expositions on principal
components techniques can be found in Chatterjee and Price, and Maddala.

The deletion of a subset of principal components as described above, in effect
imposes a restriction on the model consistent with the multicollinearity of the data. Of
course, in general, any restriction inappropriately imposed on an empirical model will
statistically bias the coefficient estimates. To minimize this undesirable effect, conservative
guidelines were used to implement these procedures. A sufficient number of components
were retained so that at least 95 percent of the variation of the explanatory data was
preserved, or F-tests for the restrictions imposed by the dropped components had a type-one
error probability in excess of 80 percent. Because of these conservative criteria, principal
components procedures were not applied to every regression. Consequently, there were
several cases where multicollinearity could not be mitigated.

Table 7. Percentage of missing observations for 95 tomato and 53 strawberry
survey questionnaires.

Explanatory Variables Missing obsv. Explanatory Variables Missing obsv.
Attributes Stby.% Tom.% Specific Practices Stby.% Tom.%
Decision-maker role 0 0 Timing frequency 24.09 26.01
Education 0 0 Number of applications 6.45 20.09
Experience 0 0 Application rate 6.88 23.35
Certification 0 0 Application method 1.51 1.49
Firm size 1.89 1.05 Surfactants 2.80 4.16
Rent vs. Own 1.89 1.05 Stickers 2.80 4.16
Objectives 13.21 9.47 Non-adjuvants 29.68 9.38
Affiliation 0 0 Decision factor 29.68 1.17
Organization 0 0
Soil type 0 1.05
General Practices Stby.% Tom.% Environmental Variables Stby.% Tom.%
Fertilizer types 0 0 Avg. temperature (current) 0 0
Soil-testing frequency 0 0 Rainfall (current) 0 0
Plant tissue analysis 0 0 Avg. temperature(lagged) 0 0
Fert. recommendations 0 0 Rainfall (lagged) 0 0
Scouting 0 0
Economic thresholds 0 0
Beneficials 0 0
Bio-control 0 0
Pheromones 0 0
Irrigation practices 0 0
Adjusting planting dates 0 0
Mechanical cultivation 0 0
Alternate pesticides 0 0
Soil test for pests 0 0
Sprayer calibration freq. 0 0
Type of irrigation 0 0
Plastic mulch 0 0
Crop rotation 0 0
Freeze protection 0 0
Season length 0 12.63
Plant distances / density 0 2.11
Plant variety 1.89 9.47

Empirical Results

Model statistics, estimated coefficients and t-tests for the OLS-Principal Component
regressions are presented in Tables 8 through 15. Results for each commodity in each
model are given in individual tables. Within each table, results for different aggregations of
the dependent variable are shown in separate columns and labeled accordingly.
Performance statistics for each regression are located in the top section of each table. These
include: the number of observations and regressors; the degrees of freedom; the number of
principal components retained in the model (if this number equals the number of regressors,
then the principal components procedure was not applied); F tests for the significance of the
overall regression and the principal components restriction when applicable, and; measures
of the goodness of fit in the form of R2 and adjusted R2 statistics. Coefficients and
significance levels (p values) for each variable can be read horizontally from left to right
across the lower half of each table. All variables were centered and normalized as a
prerequisite for the application of principal components (Chatterjee and Price). This has the
added benefit that coefficients are comparable in terms of proportions of their standard
deviation. Shaded cells are used to indicate those regressions and coefficients which were
significant at the 0.10 level or better in a 1-way test.

Attributes Models

The outcome of six regressions using the Attributes model (equation (6b)) are given
in Tables 8 and 9. Attribute models for tomatoes and strawberries were each regressed on
insecticide and fungicide residues separately, and then combined. All regressions were
significant at the 0.02 level, except for insecticides in strawberries. The poor performance
of this regression is probably due to the fact that only 38 percent of the strawberry samples
were found to contain any insecticide residues. In comparison, fungicide residues were
detected in over 94 percent of strawberry samples. Adjusted R2 statistics for the significant
attribute regressions ranged from 0.330 to 0.438. The regressions on insecticides and
fungicides combined did not out-perform those on each type of pesticide individually,
except of course for insecticides in strawberries. Principal components were not used in the
regressions for combined fungicides and insecticides, and fungicides alone in strawberries.

The first four variables in each attributes regression measured the relationships
between residues and the characteristics of the pest management decision-maker. The
position or function of this decision-maker within the firm was coded so that higher values
were assigned to individuals further removed from ownership (i.e., an owner was designated
as a 1; partner, 2; and a manager, 3). For tomatoes, decision-makers further removed from
ownership had significantly lower fungicide residues but greater insecticide residues. In the
strawberry regressions the function coefficients had the same signs as in the tomato model
but none were significant. Coefficients for education and certification were significantly
positive for fungicides and aggregate residues in strawberries, and education was
significantly positive for fungicides in tomatoes. These results contradict the hypothesized
relationship that education and training should reduce the incidence of pesticide residues in
produce. Years of experience was insignificant in all regressions.

Table 8. Strawberry grower attributes model regression results, using ordinary least squares
and principal components

Dependent Variable
Regression Statistics All Residues Fungicidesu Insecticides
Observations 53 53 53
Regressors 13 13 13
Principal-Comp. retained 13 13 10
Degrees of freedom 39 39 42
F test for regression 3.0091 b 2.9726 0.8416
Signif. level for regression l(i.u u) 0.0042 0.5923
F test for restriction NA NA 0.2296
Signif. level for restriction NA NA 0.9201
R squared 0.5008 0.4977 0.1669
Adjusted R squared 0.3343 0.3303 -0.0314
Dependent Variable
Explanatory All Residues Fungicides Insecticides
Variable' Coef. P, 1 tail Coef. P, 1 tail Coef. P, 1 tail
Function -0.028 0.423 -0.089 0.269 0.120 0.218
Education 0 457 0.004 U.548 0.001 -0.201 0.131
Experience 0.165 0.116 0.072 0.300 0.149 0.190
Certification 0.337 0.014 0.352 0.011 -0.030 0.425
Acres rented/total -0.172 0.143 -0.271 0.048 0.052 0.361
Acres, total 0.038 0.384 0.071 0.294 -0.001 0.499
Affiliated 0.013 0.464 0.077 0.303 -0.293 0.043
Loam -0 443 0 001 -0.430 (0.00 -0 S5 0.295
Sandy Loam -0.331 0.044 -0.375 0028 0.029 0.421
Avg. temp. harvest 0.024 0.458 -0.046 0.420 -0.036 0.339
Rainfall. harvest 0.617 0.017 0.557 0.027 -0.071 0.238
Avg. temp. growing 0.615 0.016 0.557 0.026 0.025 0.370
Rainfall, growing 0.113 0.341 -0.036 0.449 0.085 0.191

a If the number of principal components retained equals the number of regressors then
principal component restrictions were not applied.
b Shaded entries indicate statistical significance exceeding the 0.10 level.
c Includes all fungicides, insecticides and three instances of DCPA herbicide residues
d Fungicides in strawberries include Captan, Chlorothalonil, Iprodione, Iprodione2, and
e Insecticides in strawberries include Malathion, Methomyl, Mevinphos, Permethrin,
Carbaryl, Diazinon, Dicofol, and Endosulfan.
f The numerical representation of each variable is described in Table 6.
g Coefficients for soils are relative to base type of Sand.

Table 9. Tomato grower attributes model regression results, using ordinary least squares
and principal components

Dependent Variable
Regression Statistics All Residues Fungicidese Insecticides
Observations 95 95 95
Regressors 14 14 14
Principal-Comp. retained :' 13 13 11
DcLurce of freedom 81 81 83
F test for regression 5.720 b 5.462 7.635
Signif. Level for regression 0.000 0.000 0.000
F test for restriction 0.3314 0.1764 0.1057
Signif. Level for restriction 0.7189 0.8386 0.9802
R squared 0.4786 0.4671 0.5029
Adjusted R squared 0.3950 0.3816 0.4371
Dependent Variable
Explanatory All Residues Fungicides Insecticides
Variable Coef. P, 1 tail Coef. P, 1 tail Coef. P, 1 tail
Function -0.250 0.029 -0.522 0.000 0.162 0.048
Education 0.124 0.103 0.145 0.072 0.030 0.362
Experience 0.001 0.497 -0.021 0.435 0.047 0.308
Certification 0.027 0.420 0.071 0.298 0.010 0.458
Acres rented/total -0.181 0.013 -0.247 0.002 -0.017 0.414
Acres, total 0.379 0.000 0.100 0.175 0.361 0.000
Affiliated -0.289 0.005 0.076 0.248 -0.403 0.000
Loam 0.014 0.469 -0.060 0.373 0.119 0.049
Hydroponic' -0.054 0.297 -0.052 0.305 -0.007 0.471
Sandy Loam f 0.536 n.000. 0.373 0.000 0.330 0.000
Avg. temp, harvest ti.543 0.005 0.862 0.000 -0.081 0.013
Rainf.ill. harvest 0.143 0.087 0.154 0.074 0.035 0.360
Avg. temp, ro,\% in- -0.400 0 (.107 -0.644 0.000 0.063 0.255
Rainfall, growing -0.015 0.440 -0.085 0.200 0.044 0.319

a If the number of principal components retained equals the number of regressors then
principal component restrictions were not applied.
b Shaded entries indicate statistical significance exceeding the 0.10 level.
c Fungicide residues in tomatoes include Chlorothalonil and Maneb.
d Insecticide residues in tomatoes include Chlorpyrifos, Endosulfan, Methamidaphos,
Methomyl, Parathion, Permethrin, Acephate.
e The numerical representation of each variable is described in Table 6.
f Coefficients for soils are relative to base type of Sand.

Coefficients for the proportion of rental acres, total acres, and affiliation measure the
association between organizational attributes of the firm and pesticide residues. Acres
rented as a proportion of total acres had significant negative coefficients in both tomato and
strawberry fungicide regressions. Total acres, a proxy for firm size, was positive and
significant for insecticide and aggregate residues in tomatoes. This suggests that tomato
growers who run larger operations tend to rely more heavily on insecticides than smaller
operations. Both tomato and strawberry operations that were affiliated with downstream
market stages were linked to lower insecticide residues. The significant negative signs for
affiliation confirm the hypothesis that vertical coordination can reduce residues through
better coordination and/or risk reduction. Affiliation was the only significant coefficient in
the strawberry attribute regression for insecticides.

Soil type was represented by a mutually exclusive set of binary variables in the
attribute regressions. As a result, one soil type had to be left out of the regressions in order
to avoid a linear dependency. Sand was chosen for this purpose. Consequently, the
significant negative coefficients for loam and sandy loam in the strawberry regressions
indicate that these soils produced berries with lower residues relative to those grown on
sand. For tomatoes, this relationship was partially reversed. Tomatoes grown on sandy
loam soils were associated with higher residues relative to those grown on sand type soils.
The binary variable for hydroponic soil type in tomatoes operated as a dummy for
greenhouse operations. The positive sign of this coefficient was unexpected since it was
presumed that greenhouses would have fewer insect pest problems than open field crop
operations. Consultations with experts in this area revealed that many greenhouses in Florida
have been experiencing significant problems with the sweet potato white fly. This may
partially account for this result.

Since one type of soil is often predominant to a particular geographic area, it is
possible that these variables captured some type of regional or geographically related
phenomenon that influences pest infestations and/or residues. For this reason, the
relationship between soil types and the county from which each sample was taken was
examined. It was found that 79 percent of the tomatoes grown on sandy-loam soils
originated from Dade county, which is located in the southeastern corner of the state. Soil
types for strawberry samples were found to be much more evenly distributed. In fact, all
three types are found in the largest strawberry producing county in the state, Hillsborough.
Although it was presumed that temperature and rainfall data would accurately represent
many of the geographically related aspects of pest infestations and pesticide residues, the
influence of other (unknown) regional type of phenomenon in this analysis cannot be
completely precluded. Thus, soil type variables may be partially reflecting such
phenomenon, particularly for tomatoes.

Temperature and rainfall variables were included in the attributes models primarily
to avoid the possibility of erroneously associating decision-maker and firm characteristics
with residues that actually resulted from random or unusual weather phenomenon. For
strawberries, harvest month rainfall and the prior month average temperatures were
significantly positive for both fungicides and aggregate residues. For tomatoes, harvest
month temperature and rainfall were positive and significant for fungicides and aggregate

residues, but harvest month temperatures were negative for insecticide residues. Growing
season average temperatures had the opposite effect of harvest month temperatures on
fungicides and aggregate residues in tomatoes. These results for temperature on fungicides
were diametrically opposite to the expectations. It was rationalized that warm harvest
temperatures would break down pesticides while warm growing season temperatures would
encourage infestations.

General Practices Models

Tables 10 and 11 present the results of the General Practices regression models for
strawberries and tomatoes, respectively. As with the Attributes models, all the General
Practices regressions were significant at the 0.02 level with the exception of insecticides in
strawberries. Based on adjusted R2, the significant General Practices regressions
outperformed their Attribute Model counterparts with statistics ranging from 0.395 for
aggregate pesticides in tomatoes to 0.533 for fungicides in strawberries. This improvement
in explanatory power could be owed to the more direct relationships between general
practices and residues than is the case with the Attributes models. Combining or
aggregating insecticides and fungicides did not improve results compared to models for each
type of pesticide alone. Significant coefficients were more common for fungicides in
strawberries and insecticides in tomatoes.

Fertilizer and Fertilizer Management

Seven variables were included in the models to explore what impact fertilizer types
and fertility management activities might have on residues. Five of these were binary
variables indicating the types or forms for fertilizer applied to the crops. Most respondents
used more than one type of fertilizer. Thus fertilizer types did not comprise a mutually
exclusive set and there was no need to exclude one type as a base. It should also be noted
that no accounting was made for interactions between types of fertilizer. Looking at Tables
10 and 11, the use of any form of chemical fertilizer other than liquid appears to diminish
the occurrence of fungicide residues in strawberries. Liquid fertilizer was also negative, but
not significantly so. Fertigation was found to be negatively related to fungicide residues.
Liquid and foliar fertilizers had a positive effect on insecticide levels in tomatoes.

Soil-testing frequency had a positive impact on fungicides, insecticides, and
aggregated pesticides in tomatoes, but was insignificant in all strawberry regressions. The
implicit sign of this coefficient is the reverse of its empirical result due to the coding scheme
used for data collection. Therefore these results can be interpreted to imply that more
intensive fertility management is associated with higher residues. This is opposite to prior
expectations. It was presumed that better fertility management would avoid conditions such
as over-fertilization that could encourage certain types of pest infestations. This implication
is supported by the significant positive coefficient of plant tissue analysis for aggregate
pesticides in tomatoes.

Table 10. Strawberry grower general practices model regression results, using ordinary least squares
and principal components.

Dependent Variable _
Regression Statistics All Residues Fungicides Insecticides'
Observations 53 53 53
Regressors 30 30 30
Principal-Comp. retained 30 30 29
Degrees of freedom 22 22 23
F test for regression 2.374 b 2.S 12 0.8059
Signif. level for regression 0.0196 1..050i 0.7116
F test for restriction NA NA 0.2897
Signif. level for restriction NA NA 0.7513
R squared 11~ 0.8026 0.5040
Adjusted R squared 0.4423 0.5334 -0.1214
Dependent Variable
L.\planator \11 Residues Fungicides InsccTvcde '
\ ariable' (oef. P. 1 Tail C'oef. P. 1 tail (o d. P. I tall
FecrLltzr Jr'n -1 078 0 006 I (63 U 00t4 22: ';
Fertilizer, liquid -0.064 0.401 -0.071 0.380 -0.105 0.381
Fertilizer. foliar -0 848 0.090 -1.117 0029 0.522 0.265
Fertilizer, fertigation -0 48 0.055 -0.436 0 056 -0.323 0.114
Soil test frequency -0.167 0.286 -0.166 0.270 0.031 0.468
Plant tissue analysis -0.123 0.301 -0.061 0.389 -0.209 0.254
Organic ferniicr 0.344 0.103 0.246 0 1-' 0.602 0.u59
Scouting, inhouse -1.120 0.017 -1 063 014 -1 Ql2 0.034
Hrnefitalj -0.178 0.210 -0.210 0.150 -0.316 0.157
Bio-control 0.259 0.138 0 52 0 056 0.024 0.469
Pheromone monitoring -0.280 0.216 -0.412 0.106 0.066 0.404
Irrigation practices -0.138 0.307 -0.104 0.339 0.278 0.218
Mechanical cultivation 0.146 0.265 0.227 0.145 -0.055 0.424
Adjust planting date -0.132 0.269 -0.039 0.422 -0.177 0.264
Alternate pesticides -0 34! 0 056 -0.246 0.102 -0.325 0.137
Soil test for pests 0 811 0.015 1.052 0.002 -0.669 0.037
Calibrate sprayer -0.133 0.289 0.018 0.466 -0 402 0 06'
Drip irrigation -0 S2u 0.005 I 002 0001 0.307 0.222
Micro-Jet irrig. Cn.36i 0 068 0 398 0.040 0.273 0.211
Overhead irrig. -0.351 0.193 -0 612 0.u53 1 111 0.005
Crop rotation -0.047 0.424 -0.222 0.166 0.443 0.096
Days. plant harvest -0.119 0.254 -0.2 I 0.094 0.203 u 202
Distance between plants 0.263 0.195 0.077 1 ~'I 0.569r 0.084
\aricl\. Selva* 0.239 0.169 0420 0.037 -0412 0074
Variety. Sweat Charlie' -0.114 0.237 -0.038 0.398 -0.115 0.301
\ar~ct. Chandler 1.442 0.029 1.738 0.007 -0.973 0.070
,. temp. hjarei 0631 0.017 0.502 0.030 u 344 0 139
R.nfaii. har c.t 2.322 0.001 2.306 0.001 u. 710 0.032
\ temp. r.'% IT'- 1.862 0.007 1 787 0.005 i. 502 122
Rainfall ,ro% mn- 0934 0.007 0 866 n 006 1 4-, 10:

a. If the number of principal components retained equals the number of regressors then principal component restrictions
were not applied.
b. Shaded entries indicate statistical significance exceeding the 0.10 level.
c. Includes all fungicides, insecticides and three instances of DCPA herbicide residues.
d. Fungicides in strawberries include Captan, Chlorothalonil, Iprodione, Iprodione2, and Vinclozolin.
e. Insecticides in strawberries include Malathion, Methomyl, Mevinphos, Permethrin, Carbaryl, Diazinon, Dicofol, and
f. The numerical representation of each variable is described in Table 6.
g. Coefficients for varieties are relative to base type of Oso Grande.

Table 11. Tomato grower general practices model regression results, using ordinary least squares and principal

Dependent Variable
Regression Statistics All Residues Fungicides Insecticides
Observations 95 95 95
''.. .... 30 30 30
Principal-Comp. retained' 21 27 16
Degrees of freedom 73 67 78
F test for regression 3.9768 b 3.8213 7.3568
Signif. Level for regression 0.0000 0.0000 0.0000
F test for restriction 0.4118 0.2193 0.5373
,. '1 i Level for restriction 0.9243 0.9267 0.9095
R squared 0.5336 0.6063 0.6014
Adjusted R squared 0.3994 0.4476 0.5197
Dependent Variable
Explanatory All Residues Fungicides Insecticides
Variable Coef. P, I tail Coef. P. 1 tail Coef. P, 1 tail
Fertilizer, dry -0.037 0.337 0.021 0.419 -0.065 0.199
Fertilizer, liquid 0.010 0.457 -0.055 0.391 0.122 0.009
Fertilizer, foliar -0.105 0.169 -0.027 0.408 0.097 0.019
Fertilizer. fertigation -0.061 0.256 -0.207 0.032 -0.013 0.401
,.0i test frequency -0.178 0.024 -0.384 0.010 -0.091 0.045
Plant tissue analysis 0.148 0.061 0.166 0.124 0.008 0.453
Organic fertilizer 0.255 0.000 0.359 0.026 -0.037 0.231
Scouting, inhouse -0.065 0.288 0.239 0.061 -0.016 0.390
Benificials 0.080 0.267 -0.055 0.389 -0.064 0.071
Bio-control -0,171 0.083 -0.111 0.212 -0.OSO 0.116
Pheromone monitoring -0.030 0.405 -0.166 0.171 -0.108 0.006
Irrigation practices 0.093 0.176 0.016 0.457 0.066 0.151
Mechanical cultivation 0.026 0.393 -0.193 0.123 0.149 0.004
Adjust planting date -0.034 0.363 0.128 0.267 0.165 0.002
Alternate pesticides 0.055 0.311 0.033 0.397 0.071 0.163
Soil test for pests 0.360 0.000 0.027 0.444 0.066 0.050
Calibrate sprayer 0.365 0.003 0.080 0.302 0.199 0.001
Drip irrigation -0.271 0.026 -0.121 0.254 -0.045 0.085
Plastic mulch 0.055 0.249 0.115 0.295 0.001 0.491
Crop rotation -0.045 0.303 -0.451 0.013 0.084 0.112
Days, plant harvest -0.136 0.061 -0.003 0.487 -0.161 0.019
l):-.r.;:. between plants 0.038 0.306 -0.406 0.058 0.108 0.000
Variety. AgroSet -0.075 0.228 0.085 0.332 0.023 0.358
Variety, Sunny -0.006 0.469 0.152 0.183 0.022 0.355
Variety, Sunbeam -L 103 0)00 0.082 0.367 0.041 0.160
Variety, SolarSet 0 19, u.015 0.783 0.000 -0.010 0.401
Avg temp, harvest 0.133 U.u66 0.680 0.001 -0.061 0.032
R.irnlll harvest 0.069 0.256 -0.182 0.080 0.030 0.341
Avg temp, growing -0.340 0.004 -0.959 0.000 -0.031 0.284
ajn! :la. growing -0.093 0.162 -0.406 0.003 0.038 0.295

a. If the number of principal components retained equals the number of regressors then principal component restrictions
were not applied.
b. Shaded entries indicate statistical significance exceeding the 0.10 level.
c. Fungicide residues in tomatoes included Chlorothalonil and Maneb.
d. Insecticide residues in tomatoes included Chlorpyrifos, Endosulfan, Methamidaphos, Methomyl, Parathion,
Permethrin, and Acephate.
e. The numerical representation of each variable is described in Table 6.

Organic fertilizers may foster the growth of a number of different kinds of
agronomic pests. A positive relationship for the use of these fertilizers was marginally
confirmed by the results. In tomatoes, aggregate pesticide and fungicide residues were both
significantly influenced by organic fertilizer. Organic fertilizer coefficients were uniformly
positive in the strawberry regressions, but only significant for insecticides.

Integrated Pest Management

Ten variables were included in the general practice models to represent various
facets of integrated pest management (IPM). These included the use of: scouting, beneficial
insects, biological control agents, pheromones, irrigation practices, mechanical cultivation,
adjusting planting dates, alternating pesticides, soil testing for pests, and calibrating spray
equipment to help control pests. All interviewed growers said they routinely scouted their
fields for insect and fungus pests. What varied among operations was who performed this
function. Some performed scouting "in-house", that is either doing it themselves or using
their own employees. Other growers hired outside consultants. The scouting variable tested
whether there was any relationship between residues and the use of "in-house" verses
outside consultant scouting. In the strawberry regressions, the coefficient for this variable
was uniformly negative and significant at the 0.05 level or better. Thus, owners and their
employees appear to be prescribing fewer pesticide applications than outside consultants,
thereby leaving fewer residues. This implication was not supported by the tomato
regressions where in-house scouting was significant but only positive for fungicides.

Consideration of beneficial insects in pesticide applications was significant for
insecticide residues in tomatoes. This coefficient was negative as hypothesized. The use of
biological control agents had a significant negative impact on aggregate residues in the
tomato regressions. Negative, but less than significant coefficients for fungicides and
insecticides in tomatoes supported this result. Bio-control was inexplicably positive for
fungicides in strawberries. When this survey was conducted, most bio-control practices were
directed at controlling insect pests, thus this result for strawberries is suspicious. Pheromone
monitoring was significantly negative for tomato insecticide residues. This confirmed the
hypotheses that monitoring of insect pest populations using pheromones could help reduce
pesticide applications and their consequent residues.

Adjusting irrigation practices as part of IPM was consistently insignificant for both
commodities. Mechanical cultivation was found significant and positive in the tomato
regressions for insecticides. This result indicates that disturbing the soil through cultivation
may be beneficial to the pupa stage of development for some insect pests in tomatoes,
ultimately resulting in the need for more insecticide applications. Adjusting planting dates
was also positive for tomato insecticide residues. This was counter to expectations that
shifting the growing period or season could help avoid certain seasonal pest pressures.
Alternating pesticides to avoid the development of resistant strains was found to negatively
influence aggregate residues in strawberries. Switching pesticides is supposed to help avoid
the need to increase application rates due to the development of resistant strains. It may also
help prevent a buildup of residue from the repeated use of one particular pesticide. Soil
testing for pests is typically directed at detecting nematodes. Since the outcomes of these

tests were not requested and no nematicide residues were reported in laboratory analysis of
tomato and strawberry samples, this variable may be functioning as a proxy for pest
management intensity. Soil testing for pests was generally positive in its relationship to
residues. One exception to this result was for insecticide residues in strawberries, which
was significantly negative. More frequent sprayer calibration was expected to be negatively
related to residues, since spray equipment will tend to broadcast too much product as it
becomes worn. This was not the case for tomato insecticides or aggregate pesticides. It did
hold true though for insecticides in strawberries although this regression itself was
insignificant. An alternative hypothesis for this variable is that the frequency of calibration
is positively correlated with the frequency of pesticide applications. In other words, those
growers who rely more heavily on pesticides, calibrate more frequently, and thus have more

Irrigation, Crop Rotation, Days from Plant to Harvest, and Plant Density

The form and use of irrigation is postulated to alter the immediate environment of a
field crop. Since some growers used more than one type of irrigation technology, the set of
binary variables used to represent this production practice was not mutually exclusive and
there was no need to choose a base variable. In the General Practices regressions, residue
levels were negatively associated with the use of drip irrigation technology in both tomatoes
and strawberries. Alternative forms of irrigation technologies used in strawberries included
micro-jet and overhead systems. Micro-jet had a positive influence on fungicide residues
while overhead irrigation positively impacted insecticide residues but negatively affected
fungicide residues. Gravity or seep irrigation was also used by tomato growers, but could
not be included in the regression due to a linear dependency.

Crop rotation is believed to help reduce many types of pest problems. This
hypothesis was confirmed for fungicide residues in tomato production but refuted for
insecticides in strawberries. Strawberries and tomatoes which were sampled later in their
growing season were found to have fewer fungicide and insecticide residues, respectively.
This runs counter to the hypothesis that early season varieties or plantings might have fewer
pest problems and require fewer pesticide treatments. Other factors related to the timing of
crop production that might influence residues were presumed to be captured through the
temperature and rainfall variables. Spacing between plants or plant density could not be
completely represented in the models due to an improperly worded survey question
regarding row width. Consequently, only the distance between plants within the row could
be used in these regressions. Surprisingly, this variable was positive for insecticide residues
in both strawberries and tomatoes (wider spacings were linked with higher residues). More
in line with expectations was its significant negative relationship to fungicides in
strawberries. The rationale here was that higher plant densities (shorter distances between
plants) are more favorable to various of pest infestations, thus leading to more pesticide

Plant Varieties

The use of different varieties or cultivars of tomatoes and strawberries had
significant effects on residues. Oso Grande, the most frequently reported strawberry cultivar
in the survey, served as a base for these variables. Selva and Chandler strawberry cultivars
were positively associated with fungicide residues compared to Oso Grande. The
relationships were just the opposite for insecticides, where the Selva and Chandler varieties
had significant negative impacts. A single base variety could not be used for tomatoes. Six
of the ten tomato varieties used by growers had only one to six observations. To help avoid
problems with linear dependencies all six of these minor varieties were dropped from the
regressions. Thus the base for the tomato variety coefficients is a composite of these six
varieties. The Sunbeam tomato cultivar was negatively associated with aggregate residues,
but this was not supported by the fungicide or insecticide regressions. SolarSet was
positively associated with fungicide and aggregate residues. Given the difficulties
representing tomato varieties it is difficult to derive reliable implications from these results.


Results for the weather variables in the General Practices models were generally
consistent with those found in the Attributes Models. Higher temperatures and more rainfall
were again positively associated with fungicide residues in strawberries. This time, these
results carried over more consistently to aggregated residues although only harvest period
rainfall was significant in the insecticide regression. For tomatoes, the results were identical
to those of the Attribute model except that both harvest and growing season rainfall were
found to reduce fungicide residues. This was unexpected, particularly with respect to the
pre-harvest period, since fungus problems are often associated with humid or wet

Specific Practices Models

Specific Practices were modeled somewhat differently than Attributes and General
Practices. Background research indicated that the environmental fate of each type of
pesticide was so unique that aggregating residues for this purpose would be inappropriate.
To implement this strategy, the bulk brand-name chemical application rates as reported in
the specific chemical practices section of the survey were transformed into pounds of the
common-name active ingredient applied per acre. As discussed earlier, data on the rate,
number, timing, and chemical half-life for each pesticide application were used to derive
theoretical weighted active ingredient rates (equation (8c)). Estimates of the chemical half-
life of each pesticide were obtained from Hubble and Carlson (1993). The sign for this
derived rate variable was hypothesized to be positive, the same as the untransformed rate.
Individual data sets were then created for each common-name chemical used on each
commodity. Each one consisted of pertinent data on all reported applications of a particular
common-name pesticide matched to the corresponding pesticide residue levels found in the
laboratory samples. For example, in the regression for the fungicide Captan, each
observation of the explanatory variables (the right hand side of the regression) consisted of
pertinent data on the applications of Captan that a particular grower made to the field from

which his strawberries had been sampled. The dependent variable (the left hand side) of this
regression consisted of the level of Captan residues found in that grower's sample. There
were instances where growers did not report applying pesticides that were detected in the
laboratory samples. Conversely, there were also many cases where growers reported
applying pesticides and none showed up in the laboratory samples. Obviously, observations
of the former type could not be included in the analysis but those of the latter were.
Regression procedures were then applied to each common-name data set using the
specification given in equation (8c). Regressions were also performed on Specific Practices
aggregated as either fungicides or insecticides. This was accomplished by simply stacking
the appropriate data sets.

There were a variety of statistical problems encountered in implementing the
Specific Practices regressions. These included no or very few positive observations for a
particular dependent variable (residues), explanatory variables with zero variance,
multicollinearity and singularity in the design matrix, and low degrees of freedom. After
preliminary testing, it was decided to drop the Target Pest and Level of Infestation variables
due to their potential functional relationship with other explanatory variables in the models
and their relatively high proportion of missing values. Pre-harvest temperature and rainfall
variables were also dropped to help preserve degrees of freedom and reduce

The results of the Specific Practices regression are presented in Tables 12 and 13 for
strawberries, and Tables 14 and 15 for tomatoes. Pesticides are grouped by type with Tables
12 and 14 representing fungicides, and Tables 13 and 15, insecticides. For strawberries,
regressions were run on Captan, Iprodione, Vinclozolin, Methomyl, Diazinon, Mevinphos,
aggregate fungicides and aggregate insecticides. All but one (Methomyl) of these eight
regressions were significant at the 0.10 level. Adjusted R2 statistics ranged from 0.088 for
Methomyl to 0.997 for Mevinphos. The exceptional performance of the Mevinphos
regression is questionable given its small sample size. It seems likely that a near linear
dependency exists between the dependent and independent variables in this case.


Overall, there was little consistency in the results for different pesticide active
ingredients in strawberries. Not surprisingly, regressions on aggregated insecticides and
fungicides did not perform as well as those for the individual chemicals. The lack of
consistent signs and significance for the weighted application rate variables was particularly
discouraging. Although weighted rate was significantly positive for Iprodione and Diazinon
it was negative for Captan, which was the most frequently applied pesticide for strawberries.
Either this variable has been mis-specified or data on application rates and timing are
inaccurate, or both. Dry product form was positively related to residues for Vinclozolin but
negative for Diazinon. Significant results for method of spray application (broadcast verses
banded) were positive with the exception of Vinclozolin. While it seems obvious that
banded spray applications would introduce less pesticides into the environment, this does
not appear to carry over for residues in strawberries. The addition of non-adjuvants to
pesticide applications had a negative influence on Iprodione and Mevinphos residues.

Table 12. Strawberry grower specific practices regression results for fungicides, using ordinary least squares and principal components.
Dependent Variable
Regression Statistics Captan Iprodione Vinclozolin Fungicides
i1 4.i.,. 49 15 19 83
No. of non-zero residues 39 5 7 51
Regressors 10 9 8 10
Principal-Comp. Retained 9 9 6 8
Degrees of freedom 39 5 12 74
F test for regression 3 8574 8.4399 4.2621 2.9886
Signif. level for iL .'i i i 0 110015 0.0151 0.0157 0.0059
F test for restriction 0.1302 NA 0.3446 0.3177
inil level for restriction 0.8783 NA 0.7938 0.8125
R squared 0.4709 0.9382 0.6806 0.2142
Adjusted R squared 0.3489 0.8271 0.5209 0.1625
)-way. 0.05 critical value 1 '.X 2.0150 1.7823 1.6657
1-way, 0.10 critical value I '1 f 1.4759 1.3562 1.2931
Dependent Variable
Explanatory Captan Iprodione Vinclozolin Fungicides
Variable' Coe. P, tail Coef. P, I tail Coef. ', I tail Coef, P, tail
H eni lilch.d A, .1 IIn-i ,C .11* 328 11011 7.516 1) 11()2 l. 1) 11 117 11 i I fi l
Form 11.203 0.084 d 0.246 0.005 -0 1 71 0.061
Method 0 328 0.Uil -0. 110 0.443 -0.325 0.020 0.197 0.010
N.n \lii.iot 0.099 0.198 -7.390 0.002 d 0.146 0.088
Surfactant -0..143 0.002 1.129 0.064 0.341 0.038 -0.220 0.001
Sticker -0.091 0.265 -0.471 0.255 e -0.124 0.083
Post overhead irrigation 0.124 0.141 -0.556 0.300 -0.565 0 0u7 0.036 0.344
Decision type -1.32. 0.003 -0.162 0.379 -0.511 0.002 -0.221 0.007
Avg. temperature, harvest -0.027 0.410 0.515 0.036 0.595 0.005 0.002 0.491
Rain fall, harvest -0.190 0.116 1.218 0.003 0.382 0.019 0.148 0.056

a. If the number of principal components retained equals the number of regressors then principal component restrictions were not applied.
b. Shaded entries indicate statistical significance exceeding the 0.10 level.
c. The numerical representation of each variable is described in Table 6.
d. Variable had zero variance for this regression.
e. Variable was linearly dependent with other explanatory variables.

Table 13. Strawberry grower specific practices regression results for insecticides, using ordinary least squares and principal components.
Dependent Variable
Regression Statistics Methomyl Diazinon Mevinphos Insecticides
Observations 38 14 15 67
No. of non-zero residues 4 1 3 9
Regressors 10 8 8 10
Principal-Comp. Retained 10 4 8 10
Degrees of freedom 27 9 6 56
F test for regression 1.3584 3.5060 b 712.67 1.7663
Signif. level for regression 0.2514 0.0546 00000 0.0886
F test for restriction NA 0.1346 NA NA
Signif. level for rcsiriction NA 0.9769 NA NA
R squared 0.3347 0.6091 0.9989 0.2398
Adjusted R squared 0.0883 0.4354 0.9975 0.1040
1-way, 0.05 critical value 1.7033 1.8331 1.9432 1.6725
I-way, 0.10 critical value 1.3137 1.3830 1.4398 1.2969
'Dependent Variable
Explanatory Methomyl Diazinon Mevinphos Insecticides
Variable' Coef. P, 1 tail Coef. P, I tail Coef. P, I tail Coef. P, I tail
Weighted Act Ingr. Rate 0 076 .1367 0.314 0.019 0.033 0.280 (1052 0.157
Form 0.000 0 500 -0.355 0.029 d 0.055 0 351
Method -0.111 0.280 -0.101 0.243 0.558 0.000 -0.060 0.321
Non-Adjuvant -0.108 0.275 d -6.750 0.000 -0.066 0.303
Surfactant -0.037 0.436 0.232 0.051 -0.080 0.284 -0.048 0.387
Sticker 0.447 0.020 -0.077 0 228 1.794 0.000 0.438 0.002
Post overhead irrigation -0.139 0.224 0.278 0.021 6.948 0.000 -0 131 0 176
Decision type -0.226 0.127 d d -0.124 0.173
Avg. temperature, harvest 0.537 0.020 -0.124 0.169 0.014 0.357 0.426 0.008
Rain fall, haiHest 0 093 0.315 0.303 0.039 -0 22 0318 0.118 0 197
a. If the number of principal components retained equals the number of regressors then principal component restrictions were not applied.
b. Shaded entries indicate statistical significance exceeding the 0.10 level.
c. The numerical representation of each variable is described in Table 6.
d. Variable had zero variance for this regression.

Table 14. Tomato grower specific practices regression results for fungicides, using ordinary
least squares and principal components.

Dependent Variable
Regression Statistics Chlorithalonil Maneb Fungicides
Observations 81 144
No. of non-zero residues 20 20
Regressors 7 9
Principal-Comp. Retained b 6 9
Degrees of freedom 74 134
F test for regression 1 9433 c 2.0164
Signif. level for rcrsrcs-ion 0.084L) 0.0420
F test for restriction 0.4548 NA
Signif. level for restriction 0.6363 NA
R squared 0.1361 0.1193
Adjusted R squared 0.0661 0.0601
1- %a. 0.05 critical value 1.6657 1.6563
1-way. 0.10 critical value 1.2931 1.2879
Dependent Variable
Explanatory Chlorithalonil Maneb Fungicides
Variable d Coef. P, I tail Coef. P, 1 tail Coef. P, 1 tail
Weighed acit ingr. ratc i U 4S: -0 186 0.055
Form e -0.055 0.288
Method -0' 04 0.219 -0.079 0.193
Non-Adjuvant e 0.077 0.210
Surfactant -0.078 0.079 -0.235 0.021
Sticker -0.240 0.022 -0.170 0.059
Post overhead irri'cation e e
Decision tx pe -0.326 0 009 -0.305 0.002
.\ ,. temperature -0.211 0.049 -0.139 0.072
Rain fall 0.199 0.057 0.158 0.056

a Dependent variable had zero variation.
b. If the number of principal components retained equals the number of regressors then principal component
restrictions were not applied.
c. Shaded entries indicate statistical significance exceeding the 0.10 level.
d. The numerical representation of each variable is described in Table 6.
e. Variable had zero variance.

Table 15. Tomato grower specific practices regression results for insecticides, using ordinary least squares and principal components.

Dependent Variable
Regression Slatislics Endosulfan Methomyl Methamidaphos Chlorpyrifos Insecticides
No of noi-/ero residue. 20 1 44 20 85
Regressors 9 9 8 5 9
Principal-Comp. RetaineJ' 5 6 8 4 7
Degrees of freedom 51 24 59 27 180
I lekl for rLrc-rLsion 2.0524 b 0.7543 5.2542 4 1596 6.5018
Siniiif. clcl fur regression 0.0868 0.6123 0.0001 0.0094 0.0000
F test for restriction 0.2638 0.0813 NA 0.1021 0.3597
Sunif lek-l for re-_ri tln 0.9306 0.9872 NA 0.9033 0.7822
R squared 0,1675 0.1587 0.4160 0.3813 0.2018
Adjusted R squared 0.0859 -0.0517 0.3369 0.2896 0.1708
1-way, 0.05 critical value 1.6753 1.7109 1.6711 1.7033 1.6534
I-way, 0.10 critical value 1.2984 1.3178 1.2961 1.3137 1.2863
_____ Dependent Variable
Explanatory Endosulfan Methomyl Methamidaphos Chlorpyrifos Insecticides
Variable' Coef. P, I tail Coef. P, I tail Ce P, I ail Coe. P, tail Coef. P, I tail
W\ei'hied A\ci Iner R.1t- 0.248 0.019 -0.085 0 3*;7 .(0 18 0 229 0.3.I'l il .1 0322 0 000
Form -0.117 0.054 -0.376 0.048 d e- -0.177 0001
Method -0.009 0.443 0.039 0.331 -0 31 0.004 d -0.061 0.172
Non-Adjuvant -0.010 0.441 0.029 0.374 -0777 0.001 d -0.038 0.214
Surfactant -0.177 0.005 -0.076 0.195 0.441 0.002 0.546 0.012 0.072 0.086
Sticker 0.027 0.392 -0.203 0.079 -0.701 0.001 e -0.055 0.153
Post overhead irrigation d d d d d
Decision type 0 129 0056 -0.018 0.439 0.042 0.378 0.071 0.370 0.025 0.337
Avg. temperature -0.069 0.118 0.153 0.147 -0.198 0.081 -0.056 0.365 -0.206 0.001
Rain fall -0.171 0.066 -0.107 0.284 0.171 0.054 -0.113 0.200 -0.080 0.116
a. If the number of principal components retained equals the number of regressors then principal component restrictions were not applied.
b. Shaded entries indicate statistical significance exceeding the 0.10 level.
c. The numerical representation of each variable is described in Table 6.
d. Variable had zero variance for this regression.
e. Variable was perfectly collinear with surfactant with accompanying sign.

Surfactants had a significant influence in all fungicide regressions for strawberries, but its
sign flip-flopped between chemicals. Sticker had a positive effect on insecticide residues
with the exception of Diazinon. Stickers appear to negatively influence fungicide residues
although this was only significant in the aggregate. Overhead irrigation applied post-
application was significantly negative for Vinclozolin, but was positive for both Diazinon
and Mevinphos. These last two positive signs were counter to expected results, as it was
postulated that the use of this type of irrigation after pesticide applications would wash off
residues. Significant results for using economic thresholds to decide when to apply
pesticides were consistently negative. This provides some confirmation that basing spray
decisions on IPM criteria can help reduce residues. Significant coefficients for harvest
temperature and rainfall were all positive which was consistent with the results for these
weather variables in the attributes and general practice regressions for strawberries.


Tables 14 and 15 present the Specific Practices results for tomatoes. These include
regressions for Chlorothalonil, Endosulfan, Methomyl, Methamidaphos, Chlopyrifos,
aggregate fungicides including Maneb, and aggregate insecticides. Maneb could not be
implemented alone because there were no Maneb residues detected in samples collected
from the growers who reported applying this product. All tomato Specific Practices
regressions were significant above the 0.10 level with the exception of Methomyl, which
was also the only insignificant regression for strawberries. This was probably due to fact
that only one of the 31 observations that had Methomyl applied to them, contained any
residues of this pesticide. For the remaining significant regressions, adjusted R2 statistics
ran from 0.060 for aggregate fungicides to 0.337 for Methamidaphos. Weighted active
ingredient rate was significant and positive for Endosulfan and aggregate insecticides, but
negative for aggregate fungicides. This counter intuitive result for fungicides could be due
to Maneb data set having all zero residue levels. Dry product form was significantly
negative for Endosulfan, Methomyl and insecticides. Use of banded spraying methods was
negative for Methamidaphos. The Methamidaphos regression was also the only one for
which Non-adjuvant was significant. Its sign was also negative. Significant negative
coefficients for Surfactant were generated in the fungicide and Endosulfan regressions, but
positive estimates were obtained in the regressions for Methamidaphos and Chlopyrifos.
Chlorothalonil and Methamidaphos regressions both produced negative coefficients for
Stickers. No tomato growers indicated using overhead irrigation after pesticide treatments
so this variable was dropped from these models. The use of economic thresholds as a
decision criteria for pesticide applications was significantly negative for Chlorothalonil and
fungicides but marginally positive for Endosulfan. Signs for significant weather variables
were not wholly consistent with prior results. Higher temperatures during the harvest month
were found to reduce residues, which was consistent with the prior results. Harvest period
rainfall was found to contribute to Chlorothalonil and fungicide residues. This was in
contrast to insignificant results for this variable in the other tomato models. Temperature
during the harvest season was significantly negative for Methamidaphos and aggregate
insecticides which was consistent with prior results.

Summary, Conclusions, and Recommendations


The results of the empirical analysis in this investigation was generally encouraging.
Of a total of 27 regressions, involving three types of models, 23 were significant at the 0.10
level and all but nine were significant above the 0.02 level (F tests). More aggregated
residue data were used in the Attribute and General Practices models, consequently their
implications can be more general. All regressions from these two categories performed
well, except those for insecticide residues found in strawberries. Conversely, the regression
modeling insecticide residues in tomatoes out-performed its fungicide counterpart. This
may explain why regressions on combined insecticide and fungicide residues did not
perform as well as those for each type of pesticide individually. The empirical models of the
relationships between specific practices and residues were more difficult to implement and
interpret. Although the majority of Specific Practices regressions were statistically
significant, data and implementation related problems hindered the derivation of meaningful
implications from these models.

Temperature and rainfall were the only explanatory variables common to all models
in this analysis. These variables were included to account for the random influences of
environmental factors on residues that might otherwise have been attributed to grower
attributes or practices. By and large, their influence on residues was consistent between
models. Temperature and rainfall had a uniformly positive impact on strawberry residues,
particularly fungicides. Weather impacts on tomato residues were more varied, but higher
temperatures invariably reduced insecticide residues. The signs of significant coefficients for
these weather variables are summarized in Table 16.

Table 16. Significant temperature and rainfall empirical results.

Strawberries Tomatoes
Attri- General Specific Attri- General Specific
butes Practices Practices butes Practices Practices
Temp Fung. + + + + + -
Insect. + -
Rain Fung. + + + + +
Insect. + + +


Some important statistical relationships between attributes, practices, and residues
were identified by the empirical analysis in this study. Many of these associations are
unique to the commodity, type of pesticides, or active ingredient. In some cases, empirical
results contradict the hypothesized relationships. For a substantial number of other
variables, the null hypothesis could not be rejected. Certainly, insufficient and/or inaccurate

data had deleterious effects on the outcome of these regressions. None-the-less, a significant
portion of the variation of pesticide residues among the samples was explained in the
models. Conclusions regarding these results along with some insights into the methodology
used for this research are elaborated below.

Some interesting relationships between demographic and organizational Attributes
and pesticide residues were identified in the analysis. Contrary to expectations, education of
decision-makers and certification of employees in strawberry operations contributed to
pesticide residues. This refuted the hypothesis that firms with higher skilled personnel
would be more likely to adopt the latest technologies in pest monitoring and control. The
same rationale was applied to the expected relationship between firm size and residues, but
this was discredited in the tomato regression. Obviously some other relationship exists
between these attributes and pesticide use or pest management.

More aligned with prior hypotheses, both strawberry and tomato operations that were
affiliated with downstream market stages and used greater proportions of rented crop land
had fewer residues. One possible inference from these results relates risk exposure to
pesticide use. Firms that have more equity exposure (i.e., own greater proportions of
cropland, may face greater down-side financial risk than growers who rent most of their
cropland). To help reduce the risk of loosing their higher equity stake due to pest damage,
such firms use more pesticides. Firms that are affiliated with downstream market stages
should face less uncertainty or reduced risk in marketing their output, either through better
market access or through less volatile prices. Since pesticide applications are also a risk or
yield-variance reducing practice, affiliation is a substitute form of risk reduction which may
allow producers to be more judicious in their use of pesticides. The negative sign for
affiliation could also be construed to imply that improved coordination between market
stages can help reduce residues, particularly for insecticides.

The other notable implication from the Attribute models comes from the strong
association between residues and soil types. Compared to strawberries grown on sand type
soils, sandy loam and loam soils appears to be better suited for producing berries with lower
residues. For tomatoes, the reverse is partially true. Samples taken from fields with sand
type soils were associated with lower residues than those grown on sandy loam. Thus, with
respect to pesticide residues, if growers have a choice of soil types in their area, then sand
appears to be preferable for tomatoes and soils containing loam are better suited for
strawberries. This turns out to be more practical for strawberry growers. A review of the
sample data revealed that nearly 80 percent of the sampled tomatoes that were grown on
sandy-loam soils originated from Dade county. In contrast, samples taken from the largest
single strawberry producing county in the state came from all three soil types. The
preponderance of tomato samples grown on sandy-loam soils in Dade county makes it
difficult to preclude the possibility that some other geographically related factor may be
ultimately driving this statistical relationship. On the other hand it seems reasonable to
argue that pests have no way of perceiving their geographic location other than through
various natural and manmade environmental conditions such as temperature, rainfall, soil
type, irrigation, plant density, etc. Since these variables have been included in the

regressions, there would seem to be very few locational influences remaining for the soil-
type variable to pick up.

The General Practices regressions produced significant results in virtually all areas of
cultural practice, including fertilization, irrigation, integrated pest management, and general
crop management. Significant coefficients were more common for insecticides in tomatoes
and fungicides in strawberries. Different forms of fertilizer used in cultural production were
found to be influential on residue levels. Distributing fertilizer through the irrigation
system, or fertigation, was significantly associated with lower fungicide residues in both
tomatoes and strawberries. This implies that applying fertilizer in small amounts on a more
frequent and regular basis, which is typical of fertigation, helps reduce fungus problems. It
is not known for certain why liquid and foliar fertilizers would be associated with higher
residues in tomatoes or dry and foliar fertilizers would lower residues in strawberries.
Organic fertilizers were found to contribute to insecticide residues in strawberries and
fungicides in tomatoes. The application of plant nutrients from organic forms may contribute
to micro-environments more favorable for certain pest infestations. If so, this would
necessitate the use of more chemical pesticides. Since organic fertilizers have both
agronomic and environmental benefits, it would be inappropriate to recommend reduced
reliance on this input strictly on the basis of these results.

Integrated pest management practices were most influential on insecticide residues in
tomatoes. The consideration of beneficial insects in pesticide treatments, the use of
pheromones for pest monitoring, and the application of biological agents for pest control
were all significant in reducing residues in tomatoes. Other forms of IPM practices such as
adjusting planting dates, soil testing for pests, and sprayer calibration confounded
expectations by being positively associated with residues. Instead of representing
substitutes for chemical pest control activities, these activities may actually be serving as
indicators of pest management intensity within an operation. Thus a positive relationship
was identified. For example, growers who rely more heavily on chemical pesticides for
pest management, apply pesticides more often and therefore need to calibrate their
equipment more frequently.

Although there was no indication that modifying irrigation practices as part of an
integrated pest management program had any influence on residue levels, the use of
different irrigation technologies apparently does. This may be a consequence of how
different technologies impact the micro-environment of the plant or field, or more directly,
how they impact the environmental fate of pesticides after they are applied. Overhead and
micro-jet irrigation technologies wet all or part of the leaves, stem, and fruit of the plant.
Consequently the use of such systems may encourage pest infestations in crops, especially
fungi. On the other hand, overhead and micro-jet irrigation systems could wash off any
residues already on the plants at the time the water is applied. Drip irrigation places water
only near the roots of the plant and the rate of application is more gradual. In the empirical
analysis, drip irrigation was found to negatively influence residue levels in both strawberries
and tomatoes. Micro-jet which was only used on strawberries, was a positive factor for
fungicide residues. Overhead irrigation had a positive influence on insecticides but
negatively impacted fungicides in strawberries. This contradicted initial expectations and

counteracted the implications of the micro-jet coefficient. Despite the ambiguity of these
last results, the implication of a beneficial influence of drip irrigation on residues appears
quite strong.

Higher residues were associated with shorter growing seasons in both tomato and
strawberry General Practice models. This refuted the hypothesis that earlier maturing
varieties and earlier harvests would help avoid pest infestations. It had also been reasoned
that the longer a crop was in production prior to testing, the more pesticide treatments it
would have likely received, thus increasing potential residues. Apparently, late season
growing conditions are less conducive for pest infestations and/or more conducive to the
breakdown of residues.

Distance between plants within the row was significantly associated with residues for
both commodities. A negative relationship with respect to fungicide residues was expected,
but this was confirmed only for tomatoes. The positive relationship between distance and
insecticides in both tomatoes and strawberries was not anticipated. Possibly the greater
spacing allows for better coverage with insecticides increasing the amount of residues on the
fruit. These results should be viewed cautiously since distance between rows could not be
included in the analysis.

Variety or cultivar selection is an important component in pest management and
residue levels in strawberries and tomatoes. Both fungicide and insecticide residues were
significantly influenced by variety in strawberries. The use of Selva and Chandler cultivars
were positively associated with fungicide residues but negatively related to insecticide
residues compared to the base variety of Oso Grande. A statistical base could not be
established for cultivars of tomatoes. SolarSet tomatoes were positively related to aggregate
and fungicide residues, while the Sunbeam variety showed reduced aggregate residues. The
SolarSet relationship may be due to its susceptibility to Alternaria stem canker and early
blight. This disease is often treated with fungicides applied to the soil. Generally, any
relationship between variety and residues would probably be due to specific genetic
resistance, or lack thereof, to certain pathogens and insects. Also, characteristics related to
the timing of fruit set and maturity of different varieties may influence pest infestations and
the environmental fate of residues once pesticides are applied.

There is very little that can be generalized about the results of the Specific Practices
regressions. While more than half of these 15 regressions were significant at the 0.05 level,
each experienced problems that raise doubts about its results and implications. Seven of
these regressions could not be completely specified due to zero variances or linear
dependencies among the explanatory variables (this does not include the exclusion of
overhead irrigation in tomato regressions). Perhaps the most disappointing outcome was the
poor performance of the weighted active ingredient rate variable. The regression coefficient
for this variable was significant in only six of the 15 specific practices regressions and only
four of these six significant coefficients had the expected positive sign. The overall
mediocre performance of the Specific Practices models is believed to be due in large part to
inaccurate and insufficient data. This particularly appears to be the case for the rate and
timing of pesticide applications. Since the half-life of many pesticides can be a matter of

days, even small discrepancies in the reported dates of application could jeopardize any
possibility of identifying an empirical relationship.

Only three results from the Specific Practices models were consistent enough to
warrant general implications. First, in four out of six insecticide regressions where product
form was included and significant, the signs of these variables were all negative. Thus
insecticides which are available and used in a dry product form tend to be generating fewer
residues than those used in liquid forms. This result may be due to inherent characteristics
of the dry form of pesticide carrier or to the characteristics of the pesticides which are
available in this form. Second, in three of the seven regressions for individual pesticides
that included it, the nonadjuvants variable was significantly negative. In the survey data
nonadjuvants consisted of either macro or micro-nutrients (fertilizers) added to pesticide
spray mixes. There was no specific theory based hypothesis for this variable, so this result
could only be used to indicate a need for further research. Third, the use of economic
thresholds as a decision criterion for making fungicide applications was significantly
negative in five out of six individual fungicide active ingredient regressions. Here, the
implication is that growers who apply fungicides on a preventative or pre-determined
calendar schedule (i.e., not using economic thresholds as a decision criterion) end up having
more of these residues in their produce. This implication might have been confirmed in the
General Practice regressions but the comparable variable was dropped from the analysis due
to its low variance. This last result represents a strong endorsement for IPM type strategies
in pest control.


Two types of recommendations are discussed in this section. The first deals with the
issues related to the objectives of the analysis, that being the relationships between
attributes, cultural practices and residues. These are derived from the empirical analysis.
Methodological insights brought to light during the implementation of this research are the
basis of the second type of recommendations.

Topical Recommendations

Most of the relationships identified and measured in this study were found to be
specific to a commodity (strawberries or tomatoes) and type of pesticide (fungicides or
insecticides). Consequently, any topical recommendations or related policy decisions based
on the results of this study are and should be limited to their particular context.

*Socio-demographic analysis of residue levels indicate that education, certification, and
firm size are positively associated with higher residues. This leads to the conclusion that
current educational and certification programs may not be properly or adequately
addressing issues related to the occurrence of pesticide residues in produce.
Consequently, it is recommended that government agencies with duties and
responsibilities related to health, the environment, and agriculture, re-evaluate their
existing programs or initiate new programs to educate producers on integrated pest

management practices and the proper use and handling of chemical pesticides,
particularly as they relate to reducing pesticide residues.

Where practical, growers may be able to reduce pest pressures and residue levels by
avoiding certain soil types. Results indicate that sand type soils tend to yield tomatoes
with lower residues than those grown on sandy loams, while loam and sandy-loam soils
tend to produce strawberries with lower residues than those grown on sand.

Growers should not be discouraged from engaging in cooperative behavior or
arrangements with downstream market stages, as such arrangements are linked to
reduced levels of insecticide residues in both strawberries and tomatoes.

The effectiveness of integrated pest management practices involving: (1) the protection
of naturally occurring beneficial insects; (2) the use of bio-control products; (3) the
monitoring of insect populations using pheromone traps; and (4) the use of economic
thresholds as the decision criterion for pesticide applications, in reducing pesticide
residues was empirically supported in this research. Hence such practices should
continue to be encouraged through publicly funded research and education.

The use of drip irrigation technology should be encouraged. Also, the application of
fertilizers through such irrigation systems may indirectly help reduce residues in
strawberries and tomatoes.

* Strawberries and tomatoes produced through longer growing seasons (i.e. more days
between planting and harvest) were associated with lower residues. Consequently,
producers may be able to reduce residue levels by using later maturing varieties, or by
more carefully managing pest control practices on early maturing varieties or during
growing seasons which are favorable to early maturity.

* The empirical analysis supports the importance of considering the characteristics of pest
resistance and timing of fruit-set and maturity in the selection of crop cultivar or variety.

Methodological Recommendations

* In retrospect, it was very unlikely that the quantity and quality of data necessary to
evaluate the agronomic fate of specific pesticides after they are applied to crops could
have been obtained from growers on a historical and voluntary basis. If a future study
with similar objectives were to be conducted, it would be recommended that this type of
data be collected in a more direct and contemporaneous manner. This could be done
either through controlled field experiments conducted at agricultural research stations, or
by recruiting private growers to supply detailed information on their pesticide use and
related practices as they take place.

* Significant improvements in the depth and comprehensiveness of this type of analysis
could be gained by broadening its objectives and adopting a more controlled approach
toward data acquisition and modeling. It would be interesting to explore and model the
process by which producers make real-life pest control decisions. This would involve
substantially greater economic information on the expected values and variances of costs
and returns for alternative pest management practices and strategies, as well as growers
risk preferences. Nothing was done in this study to account for the costs of residues to
growers from either regulatory sanctions or diminished consumer demand. Second, as
suggested above, alternative procedures could be devised to obtain the quantity and
quality of data needed to properly conduct this type of modeling. Sufficient data on
inputs, prices, and the probability distribution of outcomes conditional on alternative
cultural and pest management choices would permit a more comprehensive evaluation of
the relationships between cultural practices and inputs, and various dimensions of
cultural outcomes. This would help account for the role that risk and uncertainty play in
the pest control/management problem.

* In this study a segmented approach was used in investigating pesticide use and residues
at the different stages of the market channel. It is recommended that future studies take
a more integrated systems approach in evaluating practices and residues within the food
marketing system. With respect to human health, pesticide residues in food do no harm
until they are actually consumed. Thus it would seem pertinent to determine the fate of
these residues as they move all the way through the market channel. This would be
accomplished by tracking specific lots of produce through the system and sampling their
residue levels at each market stage, down to and including the dinner table.


Chatterjee, Samprit and Bertram Price, "Regression Analysis by Example.", John Wiley &
Sons, New York, 1977.

EPA, FDA USDA., Joint Release on Food Safety/Pesticides
September 21, 1993. Release No. 0815.93 "Clinton
Strengthening the Nation's Pesticide and Food Safety Laws."

Administration Proposes

Hubbell, Bryan J. and Gerald A. Carlson., "Fruit and Vegetable Pesticide Characteristics",
1993, Draft.

Just, E. Richard. and Rulond D. Pope, "Production Function Estimation and Related Risk
Considerations." American Journal of Agricultural Economics 1979, 61(2):276-284.

Maddala, G. S., "Introduction to Econometrics." Collier Macmillan Publishers, New York,

Appendix A

Derivation of a Weighting Function for Pesticide Applications

The basic presumption of the concept of "half-life" is that the rate of decay or
break down of a substance will be some proportion, p, to the amount of that substance, Q,
present at any given time, t. That is

Q) = p Q(t), where p <0.

This implies that an exponential function identical to the familiar continuous
compounding or in this case discounting function should apply.

Defining the initial amount of the substance to be Qo the continuous discounting
function at rate r is

(1) Qt = Q0e-"

Now let h represent the half-life of the substance, which is the time required for the
ending quantity to equal one-half of the initial quantity, or

Qh = = Qoe-. Then

S=e-rh or 2=erh

Taking logarithms of both sides and rearranging gives


Therefore equation 1 can now be written as

(2) Qt =Q0e h

Equation 2 can be simplified by recalling that e n(x) = x, consequently

(3) Q, = Q02 = O
If a series of pesticide applications were made at regular intervals on a commodity prior
to sampling at time t and this pesticide degraded according to the above assumptions,
then using equation 3, the proper weighting function for these applications would

resemble a power series of the form

Qt-p Q -p-i Q -p-2i Qt-p-3i t-p-(n-l)i
(4) Q + + -p-2i + + + where
St p P+i p+2i pr3i p+(n-l)i
2h 2h 2h 2h 2 h

Q, = residual quantity of pesticide remaining on the commodity at time t (its
sampling date);
Qt-p-ki = quantity of pesticide applied on day (t p ki), prior to its being sampled;
p = pre-harvest interval, i.e., the number of days between the last pesticide
application and the sampling of the commodity;
i = the number of days between regular pesticide applications;
n = the total number of pesticide applications.

Equation 4 can be simplified if all the applications are made at the same rate of active
ingredient per acre, as is the case for much of our data. Consequently the weighting
function, equation 4, becomes

n-I 1
Q = Qa pki where
k=0 2 h

Qa = the constant application rate, and
k = indexes pesticide applications in reverse order that they were applied. That
is, k = 0 represents the last application and k = n 1 represents the first.

Florida Agricultural Experiment Station, Institute of Food and Agricultural Sciences. University o Florica, Richard Jones Dean for
Research, publishes this information to further programs and related activites, available to all persons regardless of race, color age
sex, handicap or national origin. Information about alternate formats is available from Educational Media and Services. Univers:v cf
Florida, PO Box 110810, Gainesville, FL 32611-0810 This information was published February 1999 as Bulietin 329. Florida
Agricultural Experiment Station.
ISSN 0096-6-7X

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