|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 EXPLORING THE ASSOCIATIONS OF PHYSICAL, MANAGERIAL, AND SOCIAL SETTINGS WITH HIKER VOLUME FOR THE FLORIDA NATIONAL SCENIC TRAIL By JOSHUA CUCINELLA A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010
2 2010 Joshua Cucinella
3 To my supportive family
4 ACKNOWLEDGMENTS I present my foremost gratitude to my excellent advisor, Taylor Stein For my part, there could not have been a better match for me. He has given me the flexibility, trust, and constructive criticism I needed to complete this work. A lifetime of thanks for the unique opportunities he opened for me and allowed me to pursue. This thesis could not have been completed without a lot of technical assistance from my two committee members, Timothy Fik and Henry Hochmair. For their frequent assistance I am very thankful. During the transition period when I began to take over portions of the Florida National Scenic Trail project, I was very fortunate to have invaluable assistance and training from a very willing and thorough colleague, Rachel Albritton. I offer sincere thanks to Rachel, for bridging the gap between the beginning and t he future of this project. I would like to thank her most of all, however, for giving me the idea for this research (I may have never come up with it!). I extend big thanks to all the highly cooperative and engaged public land managers in Florida who helped this project run smoothly. Also, much appreciation goes out to Sunny Kil and Bin Wan for their analytical and logistical support. I would also like to thank James Colee at IFAS statistics for his much needed assistance. In addition, I thank all the people who have worked on this six plus year FNST project. Their enthusiasm and dedication at the lab and in the field made the analysis for this thesis possible. I would like to acknowledge my beloved mother, Robin Cucinella, for listening while I described my research and helping me get over stumbling blocks. Also, thanks to my Aunts Cindy and Margie for asking me about my program and research. Finally, I
5 give many thanks to my oldest and greatest friend, Travis, for keeping me sane and focused on the treats of life.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................. 4 LIST OF TABLES ............................................................................................................ 8 LIST OF FIGURES ........................................................................................................ 10 LIST OF ABBREVIATIONS ........................................................................................... 11 ABSTRACT ................................................................................................................... 12 CHAPTER 1 INTRODUCTION .................................................................................................... 14 Outdoor Recreation ................................................................................................ 14 Counting Visitors ..................................................................................................... 14 The Goal and Objectives for this Research ............................................................ 20 2 LITERATURE REVIEW .......................................................................................... 21 Overview ................................................................................................................. 21 Terminolog y: Environmental Correlates .................................................................. 22 Theoretical Background: Behavioral Ecological Models ......................................... 24 Environmental Correlates of Outdoor Physi cal Activity ........................................... 25 Potential Environmental Correlates Chosen for this Study ............................... 26 Measuring Potential Environmental Correlates Using G IS ............................... 29 Predicting Use ........................................................................................................ 31 Summary of Literature Review ................................................................................ 35 3 STUDY SITE AND METHODS ............................................................................... 40 Study Site ............................................................................................................... 40 Estimating Visitation to the FNST ........................................................................... 42 Total Annual FNST Visitation Estimation .......................................................... 45 Limitations and Adjustments Used for Estimating Visitation ............................. 47 Measuring Potent ial Environmental Correlates of Trail Use .................................... 48 Socio Demographic Variables .......................................................................... 49 Management and Environmental Variables ...................................................... 50 Descriptive Statistics of Trail Visitation ................................................................... 51 Identifying Significant Environmental Correlates ..................................................... 52 Creating a Predictive Model for Trail Visits ............................................................. 55
7 4 RESULTS ............................................................................................................... 64 Visitation Estimates ................................................................................................ 64 Total Annual Visitation ...................................................................................... 64 Monthly Visitation ............................................................................................. 65 Environmental Correlates of Visit ation on the FNST ............................................... 66 Prediction Model for Hiker Visits ............................................................................. 69 Summary of Results ................................................................................................ 71 5 DISCUSSION AND MANAGEMENT IMPLICATIONS ............................................ 90 Examining the Environmental Correlations of FNST Hiker Volume ........................ 90 Sup port for the Behavioral Ecological Model .................................................... 90 Examining Specific Environmental Correlates: Management Implications ....... 92 Limitations: Causation and Spatial Autocorrelation .......................................... 95 The Prediction Model .............................................................................................. 97 Conclusion .............................................................................................................. 98 APPENDIX: PREDICTION RESULTS FOR INDIVIDUAL MANAGING AGENCies ... 100 LIST OF REFERENCES ............................................................................................. 114 BIOGRAPHICAL SKETCH .......................................................................................... 121
8 LIST OF TABLES Table page 2 1 Potential environmental correlates of FNST hiker visitation that were obtained from prior studies. ................................................................................ 38 2 2 Variables that were chosen based on observations while collecting visitor counts and surveying at trailheads along the FNST. .......................................... 38 3 1 Types of FNST count data and the number of years the data was collected. ..... 58 3 2 Survey site classification for FNST. .................................................................... 61 3 3 Access poi nt classification for FNST. ................................................................. 61 3 4 Sources for spatial and managerial settings data. .............................................. 62 4 1 Estimated annual visitation to FNST. .................................................................. 74 4 2 Monthly visits to 43 access points along the FNST. ........................................... 75 4 3 Labels used during statistical analyses and to present results .......................... 76 4 4 Environmental correlates of hiker volume on FNST. .......................................... 77 4 5 Pairwise comparisons of managing agency variables mean predic tions for FNST monthly hiker visitation. ............................................................................ 78 4 6 Pairwise comparisons of prominent vegetation variables mean predictions for FNST monthly hiker visitation. ....................................................................... 80 4 7 Pairwise comparisons month variables mean response differences. ................. 81 4 8 Final prediction model for monthly hiker volume on FNST. ................................ 85 A 1 Prediction errors for state parks. ....................................................................... 100 A 2 Prediction errors for Cross Florida Greenway. .................................................. 101 A 3 Prediction errors for water management districts. ............................................ 102 A 4 Prediction errors for state forests. ..................................................................... 104 A 5 Prediction er rors for Florida Fish and Wildlife Conservation Commission. ....... 107 A 6 Prediction errors for national forests. ................................................................ 109
9 A 7 Prediction errors for The Department of the Interior. ........................................ 111 A 8 Prediction errors for US Air Force. .................................................................... 112
10 LIST OF FIGURES Figure page 3 1 Map of FNST, public lands, and major cities of Florida. ..................................... 58 3 2 Locations of infrared counters along FNST from 20032009. ............................. 59 3 3 Locations and names of personal observation locations on FNST. .................... 60 3 4 Trailheads where registration cards were collected to count visitors. ................. 61 3 5 Example of census data aggregation buffers. .................................................... 62 3 6 Example of Trail settings buffers ........................................................................ 63 3 7 Example of criteria used to determine presence of loop option for two FNST access points. ..................................................................................................... 63 4 1 Total annual traffic on the FNST. ........................................................................ 74 4 2 Average monthly visitation for 43 access points obtained from each access points most current monthly visitation data. ....................................................... 74 4 3 Scatterplot of prediction models predicted hiker visits versus the standardized residuals for those predictions. ..................................................... 87 4 4 Actual versus predicted counts for monthly hiker visitation for 43 access points on the FNST. ............................................................................................ 88 4 5 Scatterplot comparing the actual and predicted monthly hiker counts for 503 months on 43 FNST access points. .................................................................... 89
11 LIST OF ABBREVIATION S AF M Activities focused mana gement AIC Akaike information criterion AT Appalachian Trail BEM Behavioral Ecological Model BIC Bayesian information criterion FNST Florida National Scenic Trail FTA Florida Trail Association GIS Geographic Information System GLM Generalized Linear Model IRR Incidence Rate Ratio OFM Outcomes focused management TFC Total Foot Count TOC Total Other Count TVC Total Visitor Count USDA United States Department of Agricu lture USFS United State Forest Service VIF Variance Inflation Factor
12 Abstract of Thes is Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science EXPLORING THE ASSOCIATIONS OF PHYSICAL, MANAGERIAL, AND SOCIAL SETTINGS WITH HIKER VOLUME FOR T HE FLORIDA NATIONAL SCENIC TRAIL By Joshua Cucinella May 2010 Chair: Taylor V. Stein Major: Forest Resources and Conservation People choose to recreate in specific outdoor settings for a number of reasons, but particularly because they are seeking a pac kage of experiences that they believe they can obtain by engaging in a certain activity in specific outdoor settings Outdoor settings are comprised of physical, social, and managerial components that interact to influence what a visitor will experience. T he goal of this study is to add to the scientific understanding of hiker visitation patterns to natural trails and how visitation relates to setting characteristics and surrounding environments. The study site for this research was the Florida National Scenic Trail (FNST). First, hiker visitation was measured at 43 access points along the FNST using infrared counters, personal observations, and registration cards. Then, aspects of the environments around each access points were measured and defined using re motely gathered spatial data. Aspects of the environmental, managerial, and socio demographic settings were collected. Next, negative binomial regression was used to create a model to describe the relationships between the FNSTs settings and the number of visitors various Trail access points receive. Finally, a prediction model was
13 created using negative binomial regression to see if the known relationships between settings and visitation could be used to make accurate predictions for visitation. Results indicate that 28 measured aspects of the Trails environment were associated with how many visitors visited the Trail Some variables were positively associated with visitation such as population density, the proportion of unemployed, scrub as the predominant vegetation type, and the presence of loop options for hikers. V ariables with a negative association included the proportion of the population under age 18, the fee for entrance, and management by the US Air Force. The prediction model showed that these known relationships could be used to make predictions about visitation. Several implications for management arose from the results of this research. These include marketing and development implications. Some implications for marketing include emphasizing people under age 18 and promoting habitats that are unique to Florida. Implications for development include providing loop trails for hikers, and placing new sections of the FNST near residential population centers.
14 CHAPTER 1 INTRODUCTION Outdoor Recr eation Outdoor recreation is an experience defined as an intrinsically rewarding, voluntary engagement that occurs during nonobligated time (Driver & Tocher, 1970, p. 10). People choose to recreate in specific outdoor settings for a number of reasons, but particularly because they are seeking a package of positive psychophysiological experiences that they may only be able to attain by engaging in a certain activity in specific outdoor settings (Moore & Driver, 2005). Outdoor settings are comprised of p hysical, social, and managerial components that interact to influence what a visitor will experience. In addition to the settings, the sociodemographics, experience level, cultural background, and physical and psychological makeup of an individual will al so impact his or her experience. Knopf (1983) summed this up when he said the environments people see are, in part, created by the mind (p. 223). In other words, specific environments do not provide the same experiences to everyone; rather they provide opportunities for people to achieve different experiences. How recreation environments translate into experience depends on the social, physical, and psychological characteristics of the people using them. Thus, it is a combination of complex interacting factors that controls who will have different experiences at a particular outdoor recreation venue. Those who do have a positive experience will likely try to come back and thus visitat ion will increase to that area. Counting Visitors Outcomes focused management (OFM) is an established outdoor recreation management framework that focuses on identifying and measuring the positive and
15 negative outcomes that result from the provision of opportunities to engage in outdoor recreation. It was recently known as bene fits based management (BBM), but now the name reflects the consideration of negative as well as positive outcomes of outdoor recreation (Driver, 2009; Moore & Driver, 2005). Today, many agencies like the USFS, BLM and public agencies from other countries have adopted OFM (Booth et al., 2002), illustrating the acceptance among managers of its value as an effective management framework. OFM emphasizes collaborating with various stakeholders such as visitors, local tourism providers, and residents of surroundi ng communities to establish beneficial goals that should result from the provision of a particular outdoor leisure opportunity. In this way, it answers the question of why should any leisure service be provided? (Moore & Driver, 2005, p. 191). The OFM pr ocess is a right to left process that first identifies beneficial outcomes desired and negative outcomes that should be avoided. Then it moves left to see which combination of opportunities for positive experiences and activity settings can provide these benefits. So, on the left are the inputs of management such as facilities and visitors, and on the right are the outcomes from the provision of the leisure opportunity. Activities focused management (AFM) is the most basic component of OFM and it focuses on the left side of the OFM process. AFM is concerned with providing recreation settings for people to pursue outdoor activities in such as camping, canoeing, and hiking (Driver,199 3 ). The key measures of success and accountability for AFM is the number of v isitors over a given time period and the number and type of recreation facilities (Budruk, Virden, & Waskey, 2009). OFM is mostly interested in identifying the beneficial outcomes of outdoor recreation and the number of visitors can give an
16 indirect measur e of this if used in conjunction with surveys or other methods of assessing the quality of the visitors experience. Because AFM is the most basic part of OFM, measuring visitor counts is among the most important assessments that a practitioner of OFM should conduct. Other assessments include determining realized personal benefits, user settings preferences, potentiality for zoning and marketing, and conducting environmental inventories (Driver, 2008). Knowing how many visitors a location has or will have can also help a manager better assess resource impacts, plan for future facilities, and calculate the economic impact from providing a particular recreation opportunity. Without good use data, decisionmakers may overlook recreation areas when allocating budgets for maintenance and expansion. Also, being able to cite visitor counts can give added credibility to management decisions that favor the provision of recreation to other alternative uses for a resource (Loomis, 2000). A land manager cannot credibly cite the importance of recreation amongst multiple competing interests if no real use data exists. It is not enough to say there are a lot of people using an area. The number of visitors a recreation area receives can also be important when considering cr owding issues and potential impacts on a recreation resource. The Limits of Acceptable (LAC) change framework is an example of why it is important to know how many visitors a location is receiving. This is a ninestep process used by the USFS that is inten ded to manage impacts of visitor use and to preserve social and environmental settings for future recreational enjoyment. The fourth step of the process, after identifying the indicators of successful management, is to inventory the c urrent conditions at t he park. This often includes visitor use numbers (Stankey et al., 1985).
17 Ultimately, standards are set for different areas such as number of felled trees at a campsite or the number of encounters along the trail using these inventories as guidelines. Manag ers use the inventories to make judgments about what is an acceptable measure of contact or any other measure. The inventories help them set reasonable standards based on the areas current condition. For instance, if users of a backcountry trail within a national park encounter an average of five people per day, and people at a developed campground encounter an average of twenty people per day, reasonable standards might be six encounters per day on the backcountry trail and thirty at the campground. The i nventory may also show that conditions have already reached an unacceptable level. So again, collecting visitor use numbers is considered a basic requirement for effective management. There are several methods for counting recreation area visitors and vis its. Recreation visits measure how many times a human being enters a certain recreation area in a period time. This includes repeat visits by the same person during that same period of time. Measuring the number of visitors an area receives tells a manager how many distinct individuals have visited an area over some period of time (Moore & Driver, 2005). Some of the various methods to measure visits and visitors include visual observations, mechanical traffic counters, remote sensing techniques, and voluntary or mandatory registration stations. They each have positive and negative characteristics relating to effects on visitor experience, cost, and reliability. When choosing an appropriate method several issues should be taken into consideration including the size of the recreation area, the number of access points, and availability of resources like
18 time and money (see the Chapter 3 for an overview of the process researchers used to estimate the number of visits to the FNST for this study). Understanding how many visitors an outdoor recreation area receives has been shown to be a basic component of effective management. This study sought to assist Florida land managers by first, describing visitation to recreation sites on the Florida National Scenic Trail u sing a uniform methodology for long term monitoring. The study then moves further to attempt to identify which characteristics of the Trail and the communities surrounding it are related to varying levels of use along the Trail. Finally, the study sought t o help managers estimate visitation patterns for areas of the FNST where visitation has not been directly measured using modeling techniques based on settings and community characteristics. Hopefully, this information will help the land managers involved i n administration of the FNST provide beneficial outcomes to visitors, the natural communities along the Trail, and society at large. Because of the unique qualities of national scenic trails as units of public recreation land, measuring use has been probl ematic. The Appalachian Trail (AT), the first completed national scenic trail, is a good example of the problems facing a multi thousand mile recreation area regarding the measurement of visitation. As a unit of the National Park Service, the AT has a mandate to report use under a measure of the Government Performance and Results Act (GPRA). However, there has never been an objective, scientific evaluation of use on the AT and they therefore cannot meet these guidelines to report on their performance as a public provider of outdoor recreation. There simply has not been a widely used or accepted methodology for estimating use over such a large geographic scale. Many national parks only need to count the cars
19 coming in and out of manned entrances and exits to achieve a reasonable estimate of visitation. There are so many different agencies and even states involved in managing the AT it presents a significant logistical challenge to try to estimate use. Hikers may enter the AT at a national forest in one state and then exit the trail on private land in a different state. The National Park Service has recently joined with The Pioneering Research Unit of the USFS Southern Research Station in an attempt to address this challenge by devising a statistically valid met hodology for measuring visitation on the AT and potentially other national trails (USDA Forest Service). The USFS, as the managing agency in charge of administering the FNST, faces challenges similar to the AT for measuring visitation. Although it only exi sts in one state, the FNST crosses land owned by a wide variety of public and private parties. These parties often have very different styles of management and control of access. Also, many of the properties that the FNST runs through have various other co nnecting or spur trails besides the FNST. This can further complicate the measurement of strictly FNST use. Like the AT, the FNST is also geographically enormous at over 1,500 miles. This increases the logistical capacity needed to measure visitation. Although these realities make measuring visitation to the FNST challenging, it is not impossible. Researchers at the University of Florida have spent the years from 2003 to the present fine tuning methods that have produced consistent and reasonable measures o f visitation on the FNST. The results of their work are the foundation for the research presented in this thesis.
20 The Goal and Objectives for this Research The goal of this study is to add to the scientific understanding of hiker visitation patterns to nat ural trails and how visitation relates to setting characteristics and surrounding environments. In pursuit of this goal the following objectives were chosen. The objectives for this research are: 1. To describe the use patterns on the Florida National Scenic Trail (FNST). 2. To identify characteristics of the FNST and the communities surrounding it that are associated with hiker participation levels and to describe the nature of that association. 3. To predict hiker use levels on the FNST based on various character istics of the Trail itself and the communities near it. These objectives are based on the hypothesis that a statistically significant relationship exists between the level of use a trail receives and the trails environmental, so cial, and managerial backdr op. The following chapter is a review of the literature concerning the relationship between levels of outdoor physical activity and the social, physical, and managerial milieu in which the activity takes place. Chapter 3 describes the methods used to meet the objectives of this study. Chapter 4 lists the results from the analyses described in Chapter 3. Chapter 5 puts the results of this study into context, describes the limitations of the study, and suggests possibilities for future research.
21 CHAPTER 2 L ITERATURE REVIEW Overview The following literature review will examine how demographic characteristics such as education, income, and age affect participation in outdoor physical activity. It will also examine how managerial and environmental settings of outdoor recreation areas impact use levels. For example, wilderness areas generally receive lower use levels than visitor centers within national forests and this is an intentional result of managerial interventions. The review begins with a discussion of t erminology that will be used throughout the rest of this document. Next, a description of the theoretical framework this study is based on will be presented. Then, the review will investigate the characteristics of outdoor recreation areas and their visitors that have already been identified as influential to visitation levels. Finally, the review will examine the methods from previous studies that were used for this research. This study is founded in the discipline of leisure studies. Therefore, its focus is on correlates and predictors of outdoor recreation, specifically hiking. However, several reviews of studies that examine outdoor physical activity in more general terms and not specifically for recreation are included. This was done because the methods used in this study were adopted from several different fields that are all concerned with outdoor physical activity for different reasons. Studies that examine why people choose to engage in physical activities outdoors, and in which settings, come from publications representing several distinct scientific disciplines. This literature review will examine studies from four distinct disciplines to help illustrate the various approaches and points of consensus among the different fields. Researchers in the fi elds of health,
22 transportation, planning and urban design, and outdoor recreation management have all been concerned with knowing which variables impact where and how often people engage in physical activity (Giles Corti et al. 2005). Some of these studies have sought to predict use levels, others focus on determining which environmental characteristics impact use in which direction and to what extent, and some do both. Terminology: Environmental Correlates Throughout the remainder of this document, the term environmental correlates will be used to describe the various variables that can be used to explain levels of outdoor physical activity in different settings (Saelens, Sallis, & Lawrence 2003, p. 80). The term correlation is derived from statistics and it means a relationship existing between events or entities which tend to fluctuate or be associated together in a linear fashion. The relationship can either be direct or inverse (DeMaris, 2004). In other words, correlates of outdoor recreation are those variables that can be measured and shown to be significantly related to levels of outdoor physical activity. Significance means that the concurrent fluctuations most likely (with at least 95% certainty for most analyses) occur due to a real world connec tion between the variables and not just by chance. It is important to use the term correlate versus predictors or influences because it is difficult to prove causation on one side or the other for this correlation. Because causation is difficult to prove f or most variables, the consistent use of the term correlate will be applied in this thesis to avoid any unsound assumptions. This caution being noted, this study does seek to identify manipulable environmental correlates for hiking. Identifying correlates that cannot be manipulated to encourage or discourage use or visitation to an outdoor area may be useful for
23 prediction purposes. However, identifying correlates that can be manipulated to increase or decrease visitation is more valuable for outdoor recrea tion managers. Knowing what these variables are could assist land managers in providing positive outcomes by allowing them to encourage or discourage use depending upon their objectives. Judgments on causation of these correlates will be held until the dis cussion section of this study. In some instances, causation may be impossible to determine. However, some relatively safe hypotheses about the direction of influence can be made for some variables. For instance, say that levels of precipitation have been s hown to be correlated with use levels at an outdoor recreation area. It is unlikely that more people using the area causes less precipitation and more likely that the precipitation is the cause of the effect of declines in visitation. The word environmental is used because this study is examining the characteristics of an outdoor recreation area and the communities surrounding it. These characteristics are distinctly different from other types of correlates of outdoor physical activity such as individual ps ychological characteristics or individual demographics, which can be obtained by surveying visitors. The word environmental may be misleading because it is often associated with ecological. Here, however, the more general meaning of environmental is being used to represent variables that are measurements of the physical, social, or managerial settings of an outdoor area. The environment in this study also includes aggregated sociodemographic measurements for the human populations (communities) that surroun d the FNST and thus represent the population pool for potential visitors. These measurements include information on income, age, race, and more.
24 Theoretical Background: Behavioral Ecological Models The foundation for the methods in this study is the behavi oral ecological model (BEM). Behavioral Ecological Models are an extension of Ecological Systems Theory (Bronfenbrenner, 1986 ). This theory states that peoples behavior and thoughts can best be understood by examining the microsystem and macrosystem of t heir environments. Microsystems are the individual, unique aspects of a persons life like their roles (e.g. father, CEO, student) and the activities they have participated in during their life. Their macrosystem includes their socioeconomic status, gender, and cultural structures and hierarchies (Woodside, Caldwell, & Spurr, 2006). Behavioral Ecological Models are based on the premise that people are part of nature, and so our behavior is as lawful as the universe around us (Hovell, Wahlgren, & Gehrman, 2002, p. 348). In these models, a behavior (e.g. hiking) is the outcome, and aspects of the environment produce this outcome. This type of model intentionally avoids placing cognitive links in between the environment and the resulting behavior. For exam ple, suppose an individual encounters an aggressive mountain lion in a nature area, escapes unhurt, but never returns to that area because of the encounter. Cognitive models of behavior based in psychology might suggest that fear kept the individual from r eturning. The BEM would say that the actual encounter with the mountain lion kept the individual from returning. This avoids reification, which gives the power to control behavior to abstract concepts like the mind, which does not exist as a tangible o bject or event (Hovell, Wahlgren, & Gehrman, 2002, p. 348). This premise allows researchers to identify environmental variables that can be manipulated to control behavior. Manipulating environmental variables may be easier to do and to
25 understand for recr eation managers than manipulating peoples thoughts, feelings, or emotions in an attempt to control their behavior. BEMs illustrate that behaviors are influenced by four factors including intrapersonal, interpersonal, environmental, and policy variables (Saelens, Sallis, and Lawrence 2003, p 80). The intrapersonal factors are those personal attributes of an individual like health or income that may affect behavior choices. Interpersonal variables include characteristics of the people surrounding an individual such as the history of outdoor physical activity within an individuals family. Environmental variables include measures of the biophysical world like the availability of natural areas close by or the level of biodiversity within a recreation area. Policy variables include regulations or information campaigns that may affect outdoor physical activity such as a state program to encourage hands on environmental education for children. This type of model was born out of the health and psychology liter atures but can be used to explain a wide range of behaviors. The ecological model approach to predicting behavior states that these four dimensions of influence interact with one another and each impact behavior to a different degree (Sallis & Owen, 1997). Environmental Correlates of Outdoor Physical Activity The following is a synopsis of studies that have examined the environmental correlates of walking outdoors using regressionbased methods involving actual counts of walkers or self reported data on lev els of outdoor physical activity. The first section of this literature review examines studies that evaluated the environmental correlates of outdoor activity that were used in this study. The next section highlights studies that provided the methodologica l basis for the statistical analyses used in this study. The
26 studies come from the transportation, health, urban planning, and leisure studies disciplines. Potential Environmental Correlates Chosen for this Study The a priori environmental correlates chose n for this study (Table 2 1 ) came from the studies reviewed in this section. B ased on a multidisciplinary review of relative literature, only one study has specifically examined the environmental correlates of hiking ( Fesenmaier Goodchild, & Lieber, 1981 ) Therefore, variables were taken from the studies in this section where they were deemed transferrable and appropriate. Also, there were some limitations on which variables could be used from prior studies because of the different methods of data collection and data availability Reynolds, Wolch, Byrne, Chou, Feng, Weaver, & Jarrett (2007) examined the environmental correlates of use levels on three multiple use greenways in Chicago, Dallas, and Los Angeles. The authors measured physical attributes of the t rail using a standardized dataform called the Systematic Pedestrian and Cyclist Environmental Scan (SPACES). They split each trail into half mile segments, with a total of 102 segments. They completed the SPACES instrument for each segment. They also cond ucted personal observations over four days (two weekdays and two weekends) for each segment in order to get a measure of use. They observed use on each segment in 15 minute blocks. They conducted Spearmans Rho correlation analysis to identify significant environmental correlates of greenway use. Next, they conducted Poisson regression. In their regression analysis, they controlled for the city in which each trail was located and the population density surrounding each trail segment. Their regression and co rrelation analyses showed that litter, trail noise and having a drainage canal as the main physical feature were all associated with decreased use on the
27 multiple use urban greenways. Increased use was associated with the presence of street lights, increas ed vegetation density, the presence of a caf, and the number of trailside facilities. Their final Poisson regression model contained 11 significant independent variables. In another study of urban greenway use, Lindsey, Han, Wilson, and Yang (2006) exam ined environmental correlates of five multiple use paths in Indianapolis, Indiana. They counted visitors using 30 infrared traffic counters more or less evenly spaced along the five trails. They measured aspects of the communities surrounding the trails us ing census data aggregated using GIS. Using stepwise multivariate regression with log transformed daily visit counts as the dependent variable they found that trail use was positively associated with average median household income, percent of population over 25 with a college degree, percent African American, percent race other than white or black, and percent of surrounding land use designated as commercial. Trail use was negatively associated with the percentage of surrounding population that was over ag e 65 and under age 5. They also included variables representing weather and the month of the year. With all of the variables (15 total), they were able to explain 80% of the variation in trail use. In a study that examined self reported use of a multipleuse urban trail in Arlington, M assachusetts, Troped, Saunders, Pate, Reininger, Ureda, & Thompson (2001) examined potential correlates of use obtained from both survey data and objectively obtained environmental measures. They used simple logistic regressi on to look for significant differences in the odds of using the Minuteman Bikeway based on differences in individual sociodemographics and aspects of the respondents
28 neighborhoods. They found that men were almost twice as likely as woman to have used the Bikeway in the last four weeks. Also, the respondents distance to the Bikeway was inversely related to the odds that they used the Bikeway. Finally, having to climb a steep hill was negatively associated with the likelihood of use. Troped, Saunders, Pate Reininger, & Addy (2003) determined correlates of outdoor walking and cycling using a combination of survey data and GIS derived data. In their study, respondents were asked how many minutes per day on average they rode their bike or walked for transportat ion (in two separate questions). Respondents were also asked to describe their neighborhoods using survey items like presence of sidewalks, presence of streetlights, and amount of automobile traffic in the neighborhood. Distance from neighborhoods to a mul tiple use greenway was the objective GIS measure used in the analysis. These variables, along with demographic items and measures of physical limitation, were used as explanatory terms in simple bivariate regressions with the combined (both walking and bic ycle riding) humanpowered transportation measures as the continuous response variable. They found that age, distance to a multiple use greenway, and living in mostly residential neighborhoods were all negatively associated with using humanpowered modes o f transportation. Enjoyable scenery, presence of sidewalks, and presence of streetlights were shown to be positively associated with walking and biking for transportation. Giles Corti, Timperio, Bull, & Pikora (2005), in their review of the environmental correlates as predictors literature, recommend using environmental variables that change when measured for different parts of a recreation area. There is wide variation of environments, managing agencies, and use levels for most of the FNST trailheads
29 exam ined in this study. In addition to the a priori variables selected from the literature, variables that may be correlated to use levels were also chosen based on observations made during time spent surveying and collecting visitor counts from the FNST (Tabl e 22 ). The authors recommend exploring new variables impacts on use in order to add to the knowledge base on environmental correlates of recreation. Measuring Potential Environmental Correlates Using GIS Among the studies that examine environmental correlates of outdoor physical activity, there are several methods used for obtaining measures of the physical environment. Reynolds, Wolch, Byrne, Chou, Geng, Weaver, & Jarrett (2007) used the Systematic Pedestrian and Cyclist Environmental Scan (SPACES) instr ument to measure different aspects of the built environment such as slope and presence of litter along a greenway. They used the data to identify which variables were related to number of users on multipleuse trails. Wilson, Lindsey, and Liu (2008) used L IDAR and high resolution satellite imagery to produce measures of greenness and visual magnitude along a greenway. These methods produced high quality trail data, but LIDAR is costly to obtain even for small areas, and gathering on the ground data may only be appropriate for smaller trails unless funding or volunteer work can support manual data gathering over hundreds of miles. More widely available data may be more appropriate for obtaining trail data for long distance trails like national scenic trails or trails that are not completely connected like many national historic trails. Lindsey, Han, Wilson, & Yang (2006) used GIS based census and land use data to measure aspects of the adjacent neighborhoods such as race and percent of commercial land use. Br ainard, Bateman, and Lovett (2001) point out the positives and negatives associated with using GIS data. They accurately point out that imperfections
30 in these datasets may introduce bias into the study. However, the ease of accessing this information and t he fact that it is often universally available makes these types of data particularly useful for managers or researchers with limited funds to create more accurate forecasting models than they had previously. In a review of the health literature covering s pecific environmental attributes association with walking, Owen, Humpel, Leslie, Bauman, & Sallis ., (2004) showed that 13 studies used perceived measures of the environment and 12 also included at least one objectively measured attribute. They did not com pare the outcomes of the two methods, but others have. Troped, Saunders, Pate, Reininger, Ureda, & Thompson (2001) compared the outcomes of using GIS obtained trail data versus using recreationists perceptions of the same variables and found different r esults on their impact on the recreationists likelihood to visit a greenway. They looked at how having a steep hill or a busy intersection between respondents homes and the greenway affected their likelihood to use the path. They found respondents' self reported presence or absence of a busy street barrier to be associated with their likelihood to use the trail. The objective GIS measure of this barrier was not associated with use likelihood. For the hill barrier, an association was found for the GIS meas ure, but not the self reported measure. In another comparison study, Kirtland, Porter, Addy, Neet, Williams, Sharpe, Neff, Kimsey, & Ainsworth (2003) compared the associations of several self reported and objectively measured measurements of the neighborhood environment with level of physical activity. Some of the measures were street traffic, neighborhood crime, presence of sidewalks, and presence of public recreation facilities. They found fair to low agreement between the self reported measures and the objectively obtained
31 measures (Kirtland et al., 2003, p. 330). They propose that the poor agreement could be based on peoples general inability to estimate distance. They also point out that people perceive their environment in different ways due to differences in cultural background, past experiences, and expectations. This study used objectively measured environmental variables, as opposed to self reported measures. Other studies have used measurements of environmental correlates that are observed directly (Reynolds et al., 2007; Brainard, Bateman, & Lovett, 2001) or self reported by those participating in the outdoor activity (De Bourdeaudhui, Sallis, & Saelens 2003; Giles Corti & Donovan, 2002). This study followed the methods of other researchers that used GIS data to obtain objective measurements for all potential environmental correlates (Lindsey et al., 2007; Lindsey et al., 2006; Troped et al., 2003). The fact that the FNST is over 1500 miles in length dictated the necessity to collect data in this manner. Plus, this method is less time and resource intensive for those who may want to use the model in the future or adapt their own models. This type of data collection could be repeated by managers or planners who have access to GIS software, which is widely available for free or relatively cheap (e.g. FreeGIS.org ). Predicting Use Predicting the number of people who are likely to use an area in the future is referred to as forecasting in the transportation literature. Among the various methods for forecasting future use, regression based methods are considered the most complex (Loomis & Walsh, 1997). Current methods for predicting nonmotorized traffic are considered poor and the U S Department of Trans portation says that more research in this area is a high priority (U S Department of Transportation, 2000, p. 45). The
32 Federal Highway Administrations (FHWA) Guidebook on Methods to Estimate NonMotorized Travel (FHWA, 1999) describes five groups of meth ods that predict demand for trails or travel corridors. The methods chosen for this particular study share characteristics with the group of methods called pedestrian demand models. Studies in this group use sample pedestrian counts and surrounding land us e or physical characteristics to create a prediction model. The model used in this study uses surrounding land use, population data, and settings characteristics from the Trail itself to attempt to explain use levels. Pedestrian demand models can be used t o identify the factors that influence the overall number of pedestrians in a particular area, predict the change in number of users by changing one of these factors, and to predict the amount of users in an area based on data collected from different areas (FHWA, 1999). A classic example of a pedestrian demand model (FHWS, 1999) from the transportation literature comes from Behnam and Patel (1977). First, they measured pedestrian sidewalk volume using personal observations in the central business district of Milwaukee, Wisconsin. They obtained the volume measures by having observers travel to several city blocks along a loop within each hour and observe each block for six minutes. Then, they multiplied each count by ten to arrive at hourly pedestrian volum e for each block. They used stepwise regression to arrive at two final models for predicting pedestrian volume. One model used a log transformed hourly average for daily pedestrian volume for each city block as the dependent variable. The other model used the log transformed noontime pedestrian volume for each block as the dependent variable. Their independent variables consisted of measures of land use in the surrounding neighborhood including commercial space, office space, cultural and
33 entertainment space, vacant space, residential space, and storage space (all measured in square meters). Their noontime pedestrian volume model was able to explain 73.9% of the variation in the number of pedestrians. Their hourly average pedestrian model was able to explain 76.4% of the variation in pedestrian numbers. They tested both models on five city blocks and found that the second model had mostly lower percentage errors for the predictions. Pushkarev and Zupan (1971) used much the same methods to predict the number of pedestrians on any given block sector in midtown Manhattan at two times of day (lunch and rushhour). They obtained their dependent variable (counts of pedestrians) from aerial photography for each of 600 block sectors. Their independent variables were units of floor space (in square feet) for the offices, restaurants, and retail locations in buildings adjacent to the walking surfaces for each block sector. In this way they were determining to what extent different types of destinations surrounding the sidewalks impacted use of those sidewalks. Also, actual walking space (again measured in square feet) of sidewalks and plazas on each block sector was an explanatory variable. Their noontime model was able to explain 61% of the variation in the number of pedestrians for block sectors occurring on streets, and 36% of pedestrian volume on avenues. They attribute the difference to a wider spread in observed pedestrian volume on avenues caused by bunching which occurs when pedestrians have to stop and go at st reet lights. This occurs more often on avenues which have more lights and so the aerial photography was less capable of capturing consistent pedestrian volume measures for avenues. Their rush hour models were less effective and only explained 52% and 23% o f the variation in street and avenue pedestrian volume, respectively. These models did
34 not contain all of the predictors of the noontime models because pedestrian volume in the evenings was more highly correlated with the distance to transportation hubs t han with the types of uses in adjacent buildings. Therefore, distance to these hubs became an independent variable in their models and, because of the varying capacity and draw of different hubs, was not an effective explanatory variable and thus produced lower R2 values. In a follow up to their greenway study previously described in the environmental correlates section of this thesis, Lindsey, Wilson, Rubchinskaya, Yang, & Han (2007) tested a regression derived prediction equation on two sections of two ur ban, multipleuse greenways. Their best model contained 18 predictors for daily traffic on 30 segments of 5 multiple use greenways. The independent variables included temporal and weather variables, and measures of the environment of the trails and the nei ghborhoods surrounding them. They compared observed and predicted values of total daily traffic for seven days on two sections of trail, each from a different greenway. They calculated the percentage error for each days prediction of total traffic (pedest rian and bicycle). The average percentage error for their predictions was 21.9%. There have been a few studies that have attempted to discover the environmental correlates of outdoor recreation use and predict visitation at specific parks or types of park s. Brainard, Bateman, and Lovett (2001) modeled recreation demand for English woodlands. They obtained counts for visitation by counting cars in two different years with mechanical counters at 33 English Forestry Commission sites. Due to limited sample per iods and problems with the counters, they were only able to obtain estimates of annual use. Using multiple regression techniques, they explored the
35 ability of 25 variables relating to onsite conditions and facilities, presence of surrogate woodlands near t he locations, and population density within 120 minute driving time to explain annual visitation to the forests. They chose variables for the models based on theoretical decisions such as how easy it was to obtain the data (and thus how feasible it would be for managers or policy analysts to duplicate the work). Their strongest model contained four variables and explained about 83% of the variation in use among the sites. They did not actually test the ability of the model to predict use, stating that their sample size of 33 was too small to make any reliable predictions. In a unique design, Dwyer (1987) attempted to explain the variation in use and make predictions about use at a single recreation location. Use was measured by the number of vehicles enteri ng a oneway in, oneway out forest preserve near Chicago. They used stepwise regression to pare down a list of variables (including interactions) and arrive at a final model. The independent variables in the final model were day of the week and holiday st atus, monthly average noontime temperatures, and monthly averages of sunshine, rain, and snow depth. The final model had an R2 value of 0.90 and managers could calculate any days use using only hand calculations or a calculator. They say in their discuss ion that they tested the model at the park and got an average error of 19%. However, they do not describe how they tested the model or for how many days they tested it. Summary of Literature Review The results of the studies in this literature review show that peoples walking behavior (either for recreation or transportation) is clearly associated with specific characteristics of these individuals and their environments. Some key social variables that showed up in several of these studies were age, educat ion, race, income, and
36 gender. Younger individuals (Troped et al., 2001; Saelens, Sallis, & Lawrence 2003) with higher education levels and incomes (Giles Corti & Donovan, 2002) are more likely to walk outdoors. These social variables tend to be correlated to levels of outdoor walking whether they are measured for the user themselves, or aggregated for communities surrounding trails. For example, older populations typically produce fewer walkers, and communities with more college graduates and higher incom es tend t o produce more walkers (Lindsey et al., 2006). This seems to indicate that the individual personal characteristics likely to be associated with walking outdoors aggregate to whole communities. In other words, communities made up of individuals wit h personal characteristics that lend themselves to walking outdoors tend to produce more walkers per capita. This makes it possible to measure correlates of use from surrounding communities instead of just the user group. Beyond just the background of the users, some consistent associations between use levels and the physical walking environment were also found in the literature. In general, levels of outdoor physical activity are positively associated with better facilities (Fesenmaier, Goodchild, & Lieber 1981) and what people perceive to be aesthetically pleasing environments (Ball et al., 2001; Humpel et al., 2004; Troped, et al., 2003). What constitutes enjoyable scenery varies depending on the location but could include increased vegetation density more natural f eatures, mixed vistas (Reynolds et al., 2007; Lindsey et al., 2008), and clean surroundings (no litter). Some characteristics of the walking environment that tend to be associated with decreased use ar e heavy automobile traffic (Liu et al., 2 00 7), noise, and litter (Reynolds et al., 2007).
37 These environmental correlates for walking outdoors tend to remain significant whether people were walking for transportation or recreation. Also, the same consistent correlates were found in a variety of backgrounds such as urban sidewalks and multiple use greenways in the studies examined. This consistency seems to indicate that walking outdoors is well suited for examination through the conceptual framework of a behavioral ecological model because peoples walking behavior can be explained by measurable aspects of their social and environmental background. In addition, this consistency also indicates that this type of model can be adapted for examining different types of outdoor activity in a variety of se ttings. The following chapter explains the methods used to identify environmental correlates and model hiker volume on the FNST.
38 Table 2 1. Potential environmental correlates of FNST hiker visitation that were obtained from prior studies. Type of v ariabl e Variable Studies u sing this v ariable Impact on u se Socio Demographic Surrounding land use (%commercial) b Lindsey et al., 2007 + Population density a Lindsey et al., 2007; Reynolds et al., 2007 + % Pop. in urban areas a Lovett, Brainard, & Bateman, 199 7 + Employment status (% not in work force) a Lovett, Brainard, & Bateman, 1997 + % Married a Troped et al., 2003 % population with college degreea Lindsey et al., 20 07; Troped et al., 2003; Troped et al. 2001 + % African American a Lindsey et al., 2007 + % Males a Troped et al., 2001 + Management Presence of facilities at trailhead (# of total facilities [e.g. benches, kiosks etc. ] ) Reynolds et al., 2007 + Environmental Density of roads [ln( mile/mile 2 ) ] b, c Handy, 1996 + Month of year Lindse y, et al., 2006 Will vary a. Indicates that these measurements are for the populations within 30 mi les surrounding each trailhead. b Indicates that these measurements are for populations within 15 driving miles of each trailhead. c The natural logarithm of this variable was used because the software used to analyze this data (PASW 18) reported mathematical errors when the full measurements were used. Table 22. Variables that were chosen based on observations while collecting visitor counts and surveyi ng at trailheads along the FNST. Type of v ariable Variable Anticipated i mpact on u se Sociodemographic Surrounding land use (% residential) b + % Hispanic or Latino a % for five income brackets a Will vary % for five age brackets a Will vary a. Indica tes that these measurements are for the populations within 30 driving miles surrounding each trailhead. b Indicates that these measurements are for populations within 15 driving miles of each trailhead.
39 Table 22 Continued. Type of v ariable Variable An ticipated i mpact on u se Management Fee amount charged + Managing agency for the trailhead Will vary Presence of toilet + Presence of map at trailhead + Environmental Geographic variables (longitude, latitude) Will vary Prominent vegetation type W ill vary Measure of vegetation type variability (# of vegetation types that make up at least 10% of buffer area) + Loop options for trail + Area of water along trail ( ln( feet 2 /mile 2 ) ) c + c. The natural logarithm of this variable was used because the software used to analyze this data (PASW 18) reported mathematical errors when the full measurements were used.
40 CHAPTER 3 STUDY SITE AND METHO DS Study Site This study was conducted on the Florida National Scenic Trail (FNST) from 2003 to 2009. The FNST was established in 1983 by an amendment to the National Trails System Act which was passed by Congress in 1968. The Act was created to in order to promote the preservation of, public access to, travel within, and enjoyment and appreciation of the openai r, outdoor areas and historic resources of the Nation (USC, Vol. 16, Section 2). The Act established national scenic trails, national recreation trails, and national historic trails. Each type of national trail is intended to meet different objectives. The first FNST blaze was painted on a tree near the center of Florida in 1966. Years of continuous volunteer work laid the groundwork for consideration by Congress to grant the Trail national scenic status and thus protect it for posterity through funding and more holistic management. The FNST was the 13th Trail designated under the Act. In accordance with legislative mandate, every national scenic trail has a federal land management agency, and a nonprofit organization counterpart, associated with it. The F lorida Trail Association (FTA), a nonfor profit group, is largely responsible for the upkeep and promotion of the Trail. The actual administration of the Trail is the responsibility of the United States Department of Agricultures (USDA) Forest Service (a bbreviated USFS). In 2003, and again in 2009, the USFS contracted the University of Florida, School of Forest Resources and Conservation to perform a fiveyear visitor assessment for the entire Trail. This assessment has consisted of counting visits and es timating total use for the Trail and individual Trail segments, examining visitors
41 settings preferences, demographics, and motivations through the use of surveys, and conducting a marketing study. The study described in this document builds on the use est imates received from 2003 through 2009. The FNST travels from the Alabama border in the Florida panhandle down to Big Cypress National Preserve close to the southern tip of the State ( F igure 3 1 ). The FNST travels through most of Floridas diverse ecosys tems including sandhill, flatwoods, scrub, cypress swamps, prairie, and even coastal sand dunes. The Trail travels through land managed by 24 land management partners as well as some private lands (FTA, 2009). The USFS and the FTA work with these partners to manage the Trail as laid out in USC, Vol. 16, Section 7 which states that the Secretary of Agriculture (the head of the USDA and thus the USFS) shall enter written cooperative agreements with the States or their political subdivisions, landowners, priv ate organizations, or individuals to operate, develop, and maintain any portion of such a trail either within or outside a federally administered area. The FNST is primarily a footpath, but there are some sections that overlap other trails (such as the W ithlacoochee State Trail) that allow other nonmotorized uses such as bicycles and rollerblades (Friend, 2004). Survey data has shown that most hikers travel five miles or less per visit to the Trail. However, the whole Trail can be travelled in 8 to 12 weeks. A majority of hikers (68%) on the FNST reported having no children at home (Wan, Cucinella, & Stein, 2009). Most (91%) of hikers were white and represented an even spread of income brackets, with over $100,000 annually being the largest percentage group (16%). Most groups consisted of one or two people (81%) and
42 used the Trail seven or more times in the previous year (52%). The Trail is less than 120 miles from most Florida residents (Albritton & Stein, 2007) and is surrounded by varying degrees of development. Some sections are in the heart of wilderness areas and others travel directly through urban areas. Just over 1000 miles of the 1562 mile Trail have been built in protected corridors (Florida Trail Association, 2009). The remaining stretches are along roads (360 miles) or on private property. It is the goal of the FTA and USFS that ultimately the remaining road sections of the Trail will be moved into protected corridors. This is in accordance with the guidelines set forth in the National Trails Sy stem Act which require that national scenic trails be located to provide for maximum outdoor recreation potential and for the conservation and enjoyment of the nationally significant scenic, historic, natural, or cultural qualities of the areas through which such trails may pass (USC, Vol. 16, Section 3). Obviously, sections located along roads do not do this as well as sections in protected, natural areas. When selecting where new segments of the trail will run, the Secretary of Agriculture must obtain the advice and assistance of the states, local governments, private organizations, and landowners and land users concerned (USC, Vol. 16, Section 7). Being able to approximate the level of use that a proposed section of rerouted Trail will have could help all of the stakeholders involved make better informed preparations and land use choices during the planning phases, as well as justify budget requests (Loomis, 2000). Estimating Visitation to the FNST Three different methods were used to obtain visitation estimates for the FNST (Table 31) Infrared counters were used for most of the FNST where only hiking is
43 allowed. The exception to this is on Eglin Air Force Base. Counts for this section of the Trail were obtained from the managers of the base who administer mandatory registration cards to all FNST visitors. Personal observations were used on sections of the FNST that allow multiple modes of nonmotorized travel in order to distinguish between types of users (e.g. bicycle versus hiker). The registration cards and the counts from the infrared counters were used to measure monthly visitation. These month by month counts were used to derive annual totals for FNST hiker visitation. The personal observation data was used to measure total annual and seasonal us e for those portions of the FNST that allow multiple modes of nonmotorized transportation. Because monthly counts were only obtained for sections of the FNST that allow only hiking, these sections were the only portions of the Trail included in the environmental correlate analyses and the predictive modeling work. Monthly counts were obtained for 43 FNST access points. Four locations were missing one or months of data for a total of 13 missing months. Therefore, the total months with distinct count data was 503 ([12 x 43] 13). The infrared counters were placed solely on hiking only sections of the Florida Trail because of their inability to distinguish use on multipleuse sections of the Trail. There are six FNST sections that contain multiple use access points where visitation estimates required the use of trained observers in order to determine how many visits were hikers and how many were travelling by some other means (e.g. by bicycle). These six multiple use portions of the Trail have 34 access points (Figure 3 2 ). The multiple use sections include bicycle, rollerbladers, and pedestrian.
44 The personal observation data were used to report total annual visitation to the multiple use sections of the Trail. A stratified random sampling approach was used to assign personal observation work shifts to trained observers. The sampling framework contained two divisions: One Day type a a. Weekdays (Mon. Thurs.) b b. Weekends (Friday Sunday) Two Time of day a a. Morning b b. Afternoon Every personal observation day contained f our possible observing periods: (2) 3hour survey shifts in the morning and (2) 3hour shifts in the afternoon. During these periods of personal observation, surveyors kept a record of all individuals entering and exiting the FNST, the size of each group, the number of males, the number of females, the activity each individual was engaged in, and their direction of travel. These observation logs obtained during the above sampling blocks were used to develop seasonal and annual estimates of visits to the FNS T for areas where mechanical counters could not be installed or would not be effective. For each access point within each personal observation section of Trail, the following counts were recorded: 1. TFC = Total Foot Count. Total number of visitors that are c onsidered foot traffic (hikers, walkers, backpackers, runners) who were observed entering or exiting the FNST. 2. TOC = Total Other Count. Total number of bikers, horseback riders, ore roller bladers, who were observed entering or exiting the FNST. 3. TVC = Tota l Visitor Count. Total number of visitors, including all activities, who were observed entering or exiting the FNST. These counts were used to estimate seasonal averages of use for each multipleuse section of the FNST (Figure 3 3) Annual counts of TFC, T OC, and TVC were calculated
45 for each multipleuse Trail section using a four step process developed and reported by Sanborn, Belcher, Albritton, & Stein ( 2004) These steps were only used for personal observation sites to calculate the annual use for these sites, not individual monthly visitation. For some locations along the FNST, reliable measures of use were already available to researchers. For six access points within the site managed by Eglin Air Force Base, visitation was measured using mandatory reg istration cards (Figure 3 4) These cards were provided to researchers in 2003 and 2004. The cards are mandatory and are located at each trailhead. There is a fine for individuals caught hiking without registering. Therefore, the use estimates derived from these cards are considered to be valid. The annual total for use in that year was recorded at the time the data was collected ( Sanborn et al., 2004). The data from the registration cards was also entered into data base files at that time. Data for three o f the six access points was subsequently lost and therefore monthly use data could not be derived for those locations. Total Annual FNST Visitation Estimation In the beginning of this study, in 2003, UF researchers identified 27 Trail sections that corresp onded to public land management areas that could be accessed by visitors and surveyors. These Trail sections were divided into high, medium, and low use categories (Table 32 ). Each of these Trail sections contains one or more access points were visitors c an access the FNST. Every access point within theses 27 Trail sections was also assigned to a category based on use level. These divisions were accomplished at the beginning of the study by visiting each FNST access point and speaking with managers to get an idea of visitation (Table 3 3 ). As the study progressed
46 and actual count data was received for many of the access points, access points were moved to different use level groups as needed. This grouping was done to help estimate use where no actual count ing had been done. Each year, an average of visitation for each access point use type (see T able 33 for the use type groupings) was calculated based on counts from the infrared counters, personal observations, and registration cards (depending on which method(s) were used) using the most current data. An average of visitation for each of the twelve months of the year was calculated for each access point use type group (A,B, C, D, and E). For those access points where visitation had not yet been measured, t he corresponding access point average was applied. This allowed researchers to estimate annual use for the entire FNST, even when counts could not be obtained from every access point due to resource limitations. As the study years went by, many of these av erages were gradually replaced by actual count data while the counters cycled through the various access points of the FNST. Therefore, annual use estimates have increased in accuracy over the six years of the study (20032009). These estimates of visitation for each access point on the FNST were summed each year for each season and for the whole year to obtain the total annual and seasonal estimates for use of the FNST. The latest years data was used each year and replaced former years measurements for l ocations where visitation was measured for more than one year. Each study year is also divided into two hiking seasons. These two time periods differentiate two seasons for hiking in the very warm and humid state of Florida. The summer is the low season, and the fall/spring season is the season with higher visitation to the FNST (Albritton, Wan, & Stein, 2008). This breakdown was created in the
47 beginning of the study based on input from managers along the Trail and has since been validated by the counts measured during the last six years. These two seasons are defined as: 1. Summer season: June 1st to September 30th 2. Fall/Spring Season: October 1st to May 31st Limitations and Adjustments Used for Estimating Visitation The personal observation data was useful for calculating a more accurate estimate of annual visitation to the FNST. However, due to the cost of sending out trained observers, the sample size is relatively low for these locations when compared to the nearly constant monitoring of the infrared moni tors. The cost of sampling prevented researchers from collecting observations every month. Thus, the data could not be used to accurately describe monthly fluctuations in visitation the way the infrared counter data and registration cards could. Because of this, the personal observation data were only used to help estimate total FNST annual visitation. The data was not used in the analysis to identify environmental correlates or the analysis to predict use on the FNST. The infrared counters used for this s tudy, although cost effective and fairly accurate have some limitations that should be noted. These counters cannot distinguish between types of users or wildlife. Therefore, a wildlife crossing is recorded as a visit. Also, when large groups pass by the c ounters, they are often not all recorded, so some underestimation can occur. This can occur anytime two people pass side by side. In addition, the counters can malfunction during prescribed burns or extreme weather and unusually high counts can be triggered by heavy winds shaking trees which move the counters. Because these counters are unmanned, several were lost to vandalism over this study. Finally, some data were lost due to drained batteries.
48 To correct for many of the limitations of the counters, meth ods of adjustment were used in an effort to clean the data. In order to correct for over or under counting, which is a common problem with these types of infrared counters (Lindsey et al., 2006), researchers conducted manual test counts each month for each counter. These consisted of walking in front of the counter ten times on the trail and then reading the number of counts registered. A ratio was derived from this test, and applied to the prior months total count. Also, all counts obtained from 10 p. m t o 5 a m were eliminated to assure that nighttime wildlife crossings were not counted as hikers. In addition, any unusually large counts that were not reported by the FTA (such as an organized group hike) were eliminated. Measuring Potential Environmental Correlates of Trail Use As stated in the methods section of the literature review, a geographic information system (GIS) was used in this study to measure aspects of the environment surrounding the FNST. To accomplish this, first the 43 different access points that had monthly hiker use data were located on a geographic grid within ArcMap 9.2 GIS software. The geographic locations of each trail traffic monitor were determined either with GPS devices or from written descriptions of their locations together w ith aerial photography and other GIS data layers that could be used to accurately place them in geographic space (Figure 3 5 ). After locating the counters in the GIS software, pertinent data layers were imported. These were all free, public access datasets except for FTA data layers concerning the FNST which normally cost but were given to the researchers by the FTA. Then, measurements were made and recorded for 32 different characteristics of the FNST at each counter location and the communities surrounding that location (Tables 2 1 and 2 2). The higher number (32) comes from the fact that
49 differing levels of the same variables were measured. For instance, to measure age distribution, eight age brackets representing all ages were used. The results of these measurements were put into a Microsoft Excel table along with the monthly visitation data from the infrared counters and hiker registration cards. Five hundred three rows were created, each representing an FNST access point and the corresponding month when visits were counted. The columns represented the measurements for the potential environmental correlates. These tables were then imported into PASW 1 8 statistics software to begin the statistical analysis. Before describing the statistical analyses used on the resulting dataset, the following paragraphs will elucidate the process of measuring the potential environmental correlates. SocioDemographic Variables Giles Corti, Timperio, Bull, & Pikora (2005) discuss the size of the areabased neighborhood from which to measure environmental variables and conclude that little continuity exists in the literature as to what areas the measurements should come from. Measurements have come from road network distances to facilities (Lindsey et al., 2006; Troped et al. 2001) and also uniform buffer distances from individuals homes (Sallis, Bauman, & Pratt, 1998). This study used methodology adapted from Lindsey, Han, Wilson, & Yang (2006) by creating a 30 mile driving distance buffer around each trailhead using ArcGIS Network Analyst. The 30 mile buffer distance comes from recent research that showed about 87% of hikers on the FNST drive 30 miles or less to arrive at the respective trailheads (Wan, Cucinella, & Stein, 2009). Then, the sociodemographics for the census blocks that overlap or are adjacent to each buffer were aggregated into percentages, means, or counts depending on the variable being
50 measured. Census block groups were used for data (e.g. income) which is not available by census block ( F igure 35 ). Socio demographic data for the populations surrounding each trailhead were collected from the Census Bureau website ( http://factfinder.census.gov ) and then edited to make them compatible within GIS software. The lan d use data came from files created by the University of Florida GeoPlan Center. These land use files were created by GeoPlan from Florida Department of Transportation (FDOT) land use data. FDOTs original 99 land use types were collapsed into 15 generalized land use types. The file was created in 2007 and was downloaded from the UF GeoPlan website ( Florida Geographic Data Library ). Parcel level data was only available for selected counties in GA and it was cited from various years. Because of this lack of comprehensive land use GIS data available for GA, smaller buffers (15 miles instead of 30) were used for both land use variables (percent residential and percent commercial) in order to stay inside Floridas boundary. M anagement and Environmental Variables For environmental and managerial variables such as vegetation type and number of facilities, several different sources were used including Florida Fish and Wildlife Commission data, USGS data, and GIS data provided by the FTA. The spatial variables were measured within buffers that extended five miles in each direction (following the Trail) from each visitor counter location (they are generally located near the trailhead and parking areas) and 1000 feet out to each side. This buffer distance was again deri ved from survey data. Albritton, Wan, and Stein (2008) showed that 78.1% of respondents walked five miles or less on the FNST. The data within these spatial buffers represent what hikers should see, hear and experience while hiking on the Trail.
51 Once again, all of the data were aggregated and organized using ArcMap (v. 9.2) GIS software (ESRI, Redlands, California) for each of the 43 individual trailheads ( F igure 3 6 ). This same 5 mile distance was used to identify loop opportunities where hikers had the opportunity to return back to the trailhead (at least partially) along a different (trail) route than the one on which they hiked out (Figure 3 7 ). If no alternative return route was available to a hiker within five miles in either direction, then the access point received a 0 (no) for the yes or no variable presence of loop trail. This variable was included because survey data from the FNST has shown that FNST hikers, on average, prefere loop trails (60%) to linear (out and back) trails (31%) (Wan, Cucinella, &Stein, 2009). Descriptive Statistics of Trail Visitation In order to describe visitation along the FNST, descriptive statistics were generated (using PASW v. 18 ) from the corrected visit counts obtained from the infrared counters, personal observation data, and hiker registration cards. Total annual visitation was derived by combining counts from the locations that had infrared counters (40), those with registration cards (3), access points that were observed by t rained observers (34), and 42 access points that were calculated using the averages for their access point use type groups. Monthly use was estimated and reported for 43 access points from infrared counter and registration card data. For those trailheads where visits were counted for more than one year, only the most current years estimates were reported and used to estimate annual use. Total FNST visitation to all access points combined is also reported for each study year since 2004.
52 Identifying Signif icant Environmental Correlates As in most of the studies examining environmental correlates of outdoor physical activity, regression was used to build a model in this study to examine associations between hiker volume and the settings of the FNST. Count da ta will rarely produce a normal conditional distribution and so ordinary least squares regression often is not appropriate (DeMaris, 2004). A type of regression more suitable for counts is one in which the dependent variable (here monthly hiker visits) are assumed to have a Poisson distribution. Poisson regression assumes that the mean and variance of y are equal. For the count data used in this analysis, the variance was much higher than the mean. This indicates that the count data may be over dispersed an d thus be better suited to a type of regression known as negative binomial regression (DeMaris, 2004). Tests comparing the dispersion value of the full model assuming a Poisson distribution, and the full model assuming a negative binomial distribution, con firmed the data was over dispersed. The dispersion is defined by: Deviance/ degrees of freedom (D/df). Deviance is a measure used to compare the fit of two models (in this case the Poisson versus the negative binomial) and is defined by: D = 2 The traditional form of the negative binomial model is a combination of the Poisson and Gamma models, where the Gamma distribution is used to adjust the Poisson model when over dispersion of the count data exists (Hilbe, 2007). A probability function describes properties of the response variable, in this case monthly visits by hikers to FNST access points. The negative binomial model is derived directly from its probability function. The negative binomial probability density func tion can be expressed as:
53 f Where: negative binomial over dispersion parameter; y = 0, 1, 2, .; estimated mean of y (e.g. E(y)) 2 8 usin g the Generalized Linear Model function which uses the following algorithm: or 2. Compute the weights as W1 2, where is the derivative of the link function and V is the variance 3. Compute a working response, a one term Taylor linearization of the log likelihood function, with a standard form of (using no subscripts) 4. Regress z on predictors X1Xn with weights, W, to obtain updates on the vector of parameter estimates, 5. Compute or based on the regression estimates. 6. Compute or as g1 7. Compute the deviance or log likelihood function. 8. Iterate until the change in deviance or log likelihood between two iterations is below a specified level of tolerance, or threshold. This is the Iteratively Reweighted Least Squares (IRLS) algorithm described by Hilbe (2007, p. 26) that is used by most Generalized Linear Model (GLM) software applications. It estimates the negative binomial model and its over dispersion 2 to 1 (Hilbe, 2007, p 94) and thus finding the parameters that best fit the count data. The resulting regression equation can be 1 through Xn using the equation:
54 1X1 nXn). Before beginning the negative binomial regression analyses to look for statistical associations between hiker volume and FNST settings, tests for multicollinearity among the independent variables were conducted. According to DeMaris (2004, p. 241), problems from multicollinearity are an issue for all generalized linear models which includes negative binomi al models. Multicollinearity is a condition in which one or more of the independent variables is almost exactly determined by the other regr essors (DeMaris, 2004, p. 224). Multicollinearity can cause several problems when trying to interpret the results of correlation analyses. These include inflation of coefficient magnitudes (which tell us how influential each variables association with the independent variable is) and signs for these coefficients that are counterint uitive (DeMaris, 2004, p. 109). Th ere is more than one way to determine multicollinearity among independent variables in a regression model. However, according to DeMaris, (2004, p. 268), the best single indicator of collinearity problems is the VIF for each coefficient The VIF is the v ariance inflation factor, which is the ratio between SE2(bj) and the value SE2(bj) would have if Xj were uncorrelated with all other regressors (Darlington, 1990, p. 129). Variance Inflation Factors are produced for each independent variable in a regress ion analysis and variables with a value of ten or higher suggest problems with multicollinearity (DeMaris, 2004). A simple fix for multicollinearity problems is to remove the variable involved. Variables with the highest VIF values were removed one at a ti me until all the remaining variables had values below ten.
55 All of the remaining independent variables were run in a negative binomial regression with a log link function using the GLM command in PASW 1 8 The monthly count data for 43 FNST access points was the dependent variable in the regression. This regression model allowed researchers to identify settings characteristics that were significantly associated with hiker volume after controlling for all other variables. Each independent variables The IRR is calculated for each independent variable (Xi) in a negative binomial regression by: IRR(Xi) = e. The IRR is the expected count for Xi + 1, divided by the expected count for Xi (I ncidence Rate Ratio, n.d.). In other words, the IRR was used to identify the impact on the predicted monthly count of hikers for a one unit change in each of the environmental correlates in the model. As an example, consider the variable entrance fee amoun fee, the expected change in hiker volume for a one dollar increase in entrance fee for any given FNST trailhead would be: IRR(fee)=e fee. For those independent variables that are measured in the same units (e. g. the sociodemographics measured as percents of the whole population), the IRRs can be compared directly to rank the independent variables by their influence on the expected number of monthly hikers (Hilbe, 2007). In order to explore the differences in the impact on the mean between levels of nominal variables, pairwise comparisons of means were executed. This provided significance values for the null hypothesis that there were no differences between mean predictions among the various levels of nominal var iables. Creating a Predictive Model for Trail Visits Predictions or forecasts of future events based on human judgment have been shown to be influenced by the personality of the person as well as the mood of the
56 individual (see Loye, 1980). Mechanical meth ods of prediction have been shown to be more effective than human judgment (Wiggins, 1973). Regression analysis based on user count data still involves human judgment, but much less than human forecasting alone. Negative binomial regression allows for the use of both continuous and nominal data to be used as predictors for a count based response variable (see Hilbe, 2007, p. 129 for an example). A significant result still does not imply causation, but it does imply a mathematical relationship which can be used to make informed predictions about the true value of Y (e.g. FNST monthly visitation levels). The saturated model was used as the base model from which the final prediction model was derived. In order to resolve the insignificant variables in the saturated model, first, possible interactions were examined. No significant interactions were identified. Following the search for interactions, variables that were not significantly associated with Trail use in the saturated model were removed and a second regression was run with the remaining variables. This process iterated until all variables in the model were significant. Two iterations were necessary and four variables were removed. Akaike's information criterion (AIC), and Bayesian information criterion (BIC) were reported as goodness of fit statistics and compared between models. Lower values indicated a better fit. This model building process in which insignificant variables are dropped from the full model is illustrated by Hilbe (2007, pg. 113 and 114) In order to look for individual observations that were outliers, standardized deviance residuals were generated for each prediction. None of the values were beyond plus or minus 2.0, which would indicate an outlier (Hilbe, 2007).
57 To test the predictive c apabilities of the final model, predicted monthly counts were generated for all months for which actual counts existed for each of the 43 FNST access points used in the analysis and the predictions were recorded. This produced 43 sets of predictions for a total of 503. After obtaining these predictions for monthly hiker visitation, the percent errors were obtained for each prediction. This was done by finding the difference of the actual counts and the predicted counts for each trailhead, and then dividing by the actual count. This proportion was then multiplied by 100 to get the percentage error. In addition, the overall mean and median percentage error was calculated to obtain an overall measure of final models reliability for predicting FNST trailhead us e. This methodology was adopted from Lindsey, Wilson, Rubchinskaya, Yang, & Han (2007). Mean absolute errors were also calculated and reported. In addition, a line graph was generated comparing the actual and predicted monthly counts to visually examine the goodness of fit of the model. Finally, a scatterplot comparing the predicted and observed counts was also created as a means of looking for ranges of count values that performed better than others.
58 Table 31. Types of FNST count data and the num ber of years the data was collected. Visitor counting method Type of data produced # Years data collected # Access points measured Infrared counters Monthly visitation 6 40 Personal observations Annual & seasonal visitation 5 34 Registration cards Mont hly visitation 1 3 Figure 3 1 Map of FNST, public lands, and major cities of Florida.
59 Figure 32 Locations of infrared counters along FNST from 20032009.
60 Figure 3 3 Locations and names of personal observation locations on FNST.
61 Figur e 3 4 Trailheads where registration cards were collected to count visitors. Table 3 2 Survey site classification for FNST. Site u se t ype Annual n umber of v isits High 1000 or more Medium 366 999 Low 0 365 Table 3 3 Access point classification fo r FNST. Access p oint u se t ype Monthly a verage of v isits A 500 or more B 100 499 C 50 99 D 15 49 E 15 or less
62 Figure 3 5 Example of census data aggregation buffers. Table 3 4 Sources for spatial and managerial settings data. Data Source Tra ilhead facilities Florida Trail Association: Map on CD 2008 Roads USGS 1:24,000 ROADS: Compiled by GeoPlan at UF, 1998 Loop (trail) options FloridaTrailsNetwork.com: Compiled by GeoPlan at UF, 2008 Fees Florida Trail Association: Map on CD 2008 Veget ation Florida Fish and Wildlife Conservation Commission, March 2004 Water USGS 1:24,000 HYDROGRAPHY: Compiled by GeoPlan at UF, 2002
63 Figure 3 6 Example of Trail settings buffers Figure 3 7 Example of criteria used to determine presence of loop option for two FNST access points.
64 CHAPTER 4 RESULTS The results in this chapter are organized in the order they were presented in the objectives from Chapter 1 as well as the order in which the analyses were completed. First, visitation estimates are pr esented for the FNST with an emphasis on monthly visitation because that data were used for subsequent analyses. The results from the environmental correlate analyses follow. The final section in this chapter focuses on the efficacy and prediction capabili ties of the final prediction model including a comparison of how the model predicts use among the various managing agencies for FNST trailheads. Visitation Estimates The first part of this study was to measure and describe the use patterns on the FNST over the last six years (20032009). The first few years of the study showed some relatively large fluctuations in annual use, but this was likely the result of methodological inconsistencies that existed as a result of the learning curve researchers had to go through to measure visitation along a 1500plus mile trail. The latter three years of the study showed very consistent visitation with modest increases. The monthly visitation data have remained consistent since the beginning of the study due to little change in tactics for collecting or analyzing these data. Total Annual Visitation In 2008 and 2009, the FNST received an estimated 349,701 visits. Annual use estimates have been increasing since University of Florida researchers began estimating use five yea rs ago (Wan, Cucinella, & Stein, 2009). The 2005 to 2006 hiking season was an exception, which saw a 38% decline in estimated use (Table 41 and
65 Figure 4 1 ) However, this exception occurred due to a change in methodology which separated Lake Okeechobee into its own highest use category from the other Trail sections and thus lowered the estimates, which were based on the access point average that contained Lake Okeechobee. Similarly, the overall trend in increasing visitation likely reflects an increase in the accuracy of the estimation methods and an increase in available visitation data that occurred as the study proceeded. Total annual visitation for individual Trail sections ranged from 298 at Rice Creek Conservation Area to over 200,000 for Lake Okeecho bee. For the past four years of the study, visitation has been nearly equally split between hikers and other traffic. This other group includes nonmotorized uses like cycling and rollerblading. Monthly Visitation For the 43 access points with monthly vi sitation data, the average monthly use was calculated from the most current years counts for each access point. The month with highest average use was March, and the lowest average use levels were in June. The monthly averages generally increase (with small declines in Sept. and Dec.) beginning with the low point in June and continuing to March. After March, as spring transitions into summer and the heat of the Florida summer sets in, use levels fall rapidly each month until June (Figure 4 1 and Table 42 ) For individual trailheads, there was a wide range in hiker visitation. Several locations received zero visitors for entire months, while others never dipped below 100 hikers even in the middle of the summer. The highest single month of visitation occurr ed at Little Big Econ State Forest near Orlando with 585 hikers during March of 2004.
66 Environmental Correlates of Visitation on the FNST In total, 33 variables describing characteristics of the FNST and the communities surrounding the Trail were examined t o see how they are related to monthly visits of hikers. These 33 variables were the independent variables in a correlation analysis and are represented in this section using the shortened names that they were represented with while working in PASW 1 8 stat istical software (Table 4 3 shows these shortened names). This was done to simplify the display of the results. As mentioned in Chapter 3, this environmental correlation analysis began by determining if there was multicollinearity among the 33 independent variables. After removing the variables responsible for the greatest inflation of coefficient variances (e.g. those with the highest VIF values), 22 variables remained. These variables all showed VIF values less than 10. These variables were then moved for ward into the model building process. The model building process for examining environmental correlates of hiker visitation to the FNST began by testing to see if the monthly count data had a normal conditional distribution, which would allow for the use of ordinary least squares regression (OLS). The data failed the Wilks Shapiro test for normalcy and thus OLS regression was deemed inappropriate. Following this test, the data was tested for suitability for analysis using standard Poisson regression. To test this possibility, a test regression on the saturated model was run using the Poisson distribution with a log link function and the dispersion (e.g. Deviance/degrees of freedom) value was recorded at 20.4. This value should be close to one for count data that are not over dispersed (Hilbe, 2004). The high dispersion value indicated over dispersion and so a regression was run for the complete model that assumed a negative binomial distribution (with a
67 log link function) and the dispersion was recorded at 1.2. This value was much closer to one and thus the problem of over dispersion was adequately addressed. Therefore, negative binomial regression was selected as an appropriate method to examine potential environmental correlates of hiker volume. The full (s aturated) model contained 22 independent variables and an intercept. Seven of the variables were not significant in the saturated model. The seven nonsignificant variables were longitude of the access points, percent commercial land use surrounding the T rail, amount of roads and water along the Trail, the number of facilities at each trailhead, the presence of a bathroom at the trailhead, and the proportion of households earning between 50 and 99 thousand dollars annually in the communities around the FNS T. In addition, several levels within the nominal variables month, vegetation type, and managing agency were also not significantly associated with hiker volume when compared to their referents. Incidence rate ratios (IRRs) were calculated because they o ffer an easily interpretable means of comparing the associations of the different independent variables with hiker volume (Table 44) Continuous variables and nominal variables with only two levels are examined first. A pairwise examination of the nominal variables with more than two levels follows this examination of the IRRs of the continuous variables and dichotomous nominal regressors. Among the sociodemographic variables, which were measured in percents, the percentage of males showed the highest IR R at 1.32. This means that a one percent rise in the proportion of males in communities surrounding the Trail is equivalent to a roughly 32% rise in predicted hiker volume for the corresponding trailhead. The proportion of surrounding populations with
68 an a nnual household income of 200 thousand or more showed the strongest negative association with hiker volume among the sociodemographic variables. Controlling for the other variables in the model, a one percent increase this income group is associated with a 70% decrease in Trail use. Among the continuous variables representing environmental aspects of the FNST, only the vegetative diversity along the Trail was associated with hiker volume. Vegetative diversity was measured as the number of vegetation types (as measured by the Florida Fish and Wildlife Conservation Commission) that make up at least ten percent of the FNST spatial buffer. An increase in this measure by one unit is associated with a 35% decrease in hiker volume. Several of the continuous managerial variables were significantly associated with hiker volume on the FNST, indicating that how public recreation areas are managed can encourage or discourage visitation. The presence of a loop for hikers to travel upon at along the Trail was associated with a doubling in hiker visits after controlling for all of the other variables in the model. Interestingly, the presence of a map at the trailhead was associated with decreased visitation. This indicates that FNST hikers are not discouraged from hiking by the absence of a map at the access point. However, the number of total facilities, (e.g. trash cans, kiosks, and trail registers) is positively correlated with Trail use. Adding one more facility is associated with a seven percent increase in visits. Pa irwise comparisons for the models nominal variables various levels were run to examine whether true differences in their associations with hiker volume exist ( T ables 4 5, 4 6, and 47 ). Among the managing agencies, the Cross Florida Greenway and
69 Departm ent of the Interior managed access points showed significantly higher visitation than the other agencies. The two agencies associated with the lowest visitation, controlling for population density and the other variables in the model, were national forests of Florida and the US Air Force. For the prominent vegetation category, scrub was associated with significantly more hikers than all of the rest of the vegetation types. Locations with hardwood hammock as the majority vegetation type were associated with fewer hikers than all of the other types except pasture (there was no significant difference in use between the two). Months of the year were considered as an environmental variable and a proxy for weather data in this analysis. June and July both received significantly lower visitation than the ten months from August through May. March received the most hikers with significantly higher visits over eight other months of the year. January, February, March and April showed no significant differences in use a mongst each other. These results follow closely the monthly averages of Trail wide use which are the lowest in June and July and have an apex in March ( Figure 4 2 ). Prediction Model for Hiker Visits The same variables representing settings characteristics that were used in the environmental correlate analysis were also used to create a prediction model for monthly use on the FNST. This analysis was based on the premise that the relationship between use and the social, environmental, and managerial aspects o f the FNST is strong enough to give reasonable predictions for use. According to Mason and Perreault (1991, p. 268), when correlation among predictors in a regression model occurs, overall prediction is not affected, but interpretation of and conclusions based on the size of the regression coefficients, their
70 standard errors, or the associated t t ests may be misleading. Because of this fact, two versions of the prediction model were created. One addressed multicollinearity in the same way it was addresse d in the environmental correlate analysis (by dropping those variables with the largest VIF values one at a time), and the other ignored multicollinearity. The results from the two were compared and the model that included the correlated predictors perform ed significantly better than the model without these predictors. The mean prediction error for the model that addressed multicollinearity was 113% while the mean error for the model that ignored multicollinearity among the predictors was 76%. The following paragraph describes the results from the model that performed better by ignoring multicollinearity. In order to achieve the final prediction model, three insignificant variables from the saturated model were dropped. These were percent residential housi ng, percent Hispanics and the longitude of each FNST access point. In the resulting model two more variables lost their significant contribution for predicting monthly hiker volume. These variables, percent married and total number of facilities, were also dropped. The final prediction model contained 28 significant predictors using robust standard errors ( Table 4 8 ). Predictions for monthly hiker visits for 43 access points along the FNST were automatically generated using PASW 18 statistical software. The final prediction model was used to make predictions for each of the 43 trailheads on the FNST for which monthly hiker counts were available. The model made predictions with errors that ranged from zero percent to 1850% for individual months. The prediction model produced an overall mean error of 76% and median error of 37%. These percent errors should be kept in perspective. For example, if actual use was 1
71 visitor, and predicted use was 6, this results in a 500% error, even though the prediction is relatively close to the actual value. Because of this inflation of the percent errors toward the lower side of the prediction scale, the absolute errors in visitation were also calculated. The absolute error ranged from zero to 261 with a mean of 24 and a median of 13. Percent errors could not be calculated for actual counts with a value of zero due to the mathematical problems of dividing by zero. To correct for this, one count was added to both the actual and predicted counts in those instances. Overall, there were slightly more underestimates than overestimates for the monthly visitation predictions. In addition to examining the overall predictive power of the model, errors were also examined for individual managing agencies. This was done in order to look for a particular agency that could potentially benefit the most from this model. The Cross Florida Greenway produced both the lowest mean and median percent error at 35% and 22%, respectively. A scatterplot of the predicted counts versus the standardized res iduals was created as a visual check for heteroscedacity of the errors. This was done to see if any range of values for the number of monthly visitors performed better or worse using the prediction model. The results ( Figure 4 3 ) indicate that the variance of the errors may not be constant. The standardized residuals funnel down as the mean of the responses increase. Summary of Results Objectives 2 and 3 of this study were concerned with describing the relationship with remotely measurable aspects of the FN ST and using that relationship to predict use on the Trail. The total number of independent variables examined included 33
72 variables representing characteristics of the social, environmental, and managerial settings of the FNST. For the environmental corr elate analysis, 11 of these were dropped due to their multicollinearity and resulting impacts on the models beta coefficients. Fifteen of the remaining 22 independent variables showed significant associations with FNST hiker volume. Population density, t he percentage of males, the percentage of unemployed, and the proportion of residential land use were all correlated with increased Trail visits. The percentage of African Americans, citizens aged 18 and under, and household incomes above 200 thousand dol lars annually were all negatively correlated with FNST visitation. These results show that the sociodemographic settings in which hiking trails are found have specific associations with how much use the trails receive. Several variables describing how dif ferent parts of the FNST are managed were significantly associated with the number of monthly hiker visits. Among the managers of the 43 trailheads on the FNST examined in the correlation analysis, the Department of the Interior and the Cross Florida Greenway drew in more hikers than all of the other agencies after controlling for all of the other influences in the model. Access points for the FNST that run through Floridas national forests and Eglin Air Force Base received the fewest hikers each month. How managers develop a site also was related to use. Sections of the FNST where loops were connected to the Trail showed higher visitation. Also, more facilities, and lower fees were relat ed to increases in visitation. T he presence of a map at the trailhead had a negative association with Trail visits, indicating that this is not a necessity for FNST hikers.
73 A few of the environmental characteristics of the FNST that were examined also displayed relationships with the number of hikers. Among the many plant c ommunities represented along the FNST, scrub received significantly higher use than areas dominated by any other vegetation while hardwood hammocks received fewer hikers. A higher diversity of vegetation types was associated with less use. The amount of roads or water encountered by hikers did not impact visitation levels. GIS data concerning weather was not freely available, and so month of the year was used as a surrogate for weather fluctuations. T railheads received significantly fewer hikers in June and July than most of the other months of the year. These months correspond with the peak of summer in the subtropical state of Florida. The predictions from the reduced model showed rather high percent errors, however, these numbers do not tell the whole story. Visual representations comparing the actual and predicted hiker volume show that the predictions follow the actual monthly visitation well (Figures 4 4 and 45) A prediction model that allowed multicollinearity produced more accurate predictions than one that corrected for multicollinearity by removing correlated predictors.
74 Table 4 1 Estimated annual visitation to FNST. Study year Total annual foot traffic Total annual other traffic Total annual traffic 2003 2004 173,138 71,763 244,9 01 2004 2005 234,473 253,319 487,778 2005 2006 163,261 166,496 300,536 2006 2007 175,435 168,556 343,991 2007 2008 180,302 169,335 349,637 2008 2009 180,366 169,335 349,701 *Other traffic includes bicyclists and rollerbladers. Figure 4 1 Total annual traffic on the FNST. Figure 4 2 Average monthly visitation for 43 access points obtained from each access points most current monthly visitation data.
75 Table 4 2 Monthly visits to 43 access points along the FNST. Access p oint Monthly v isits (for most current year) JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Gold 71 52 82 109 64 31 21 19 44 61 57 74 Etoniah 51 21 36 27 9 8 8 17 12 13 41 17 LB 377 379 366 315 216 159 134 204 182 296 361 302 Santos 244 310 310 238 209 114 119 185 165 312 322 270 Rod_E 56 42 61 116 102 79 28 108 13 39 56 71 Rod_W 21 33 38 24 10 10 7 22 17 20 9 16 Juniper 381 280 83 88 67 39 22 148 56 75 205 280 SR_19 214 199 136 66 98 91 91 78 81 75 110 144 Lake_D 33 62 18 29 23 4 2 5 4 6 26 14 Clear 21 135 164 12 1 56 35 91 78 81 75 17 17 Rice_C 48 58 40 27 31 4 7 6 2 10 43 24 SR_44 22 60 47 23 18 15 5 12 0 0 0 15 SR_46 177 230 184 151 106 53 51 59 57 76 133 101 Big_N 288 279 266 118 114 47 63 82 77 79 219 78 Big_S 54 69 104 162 74 12 7 5 7 4 2 16 Bluff 70 8 8 84 36 29 48 99 108 109 28 40 71 Hickory 80 42 47 11 27 17 29 12 11 10 48 20 US_192 11 13 27 34 25 13 4 10 9 11 4 9 Crab 65 49 126 70 66 46 25 27 65 83 114 93 3_Lake 84 51 32 33 18 16 35 17 41 67 28 69 River a a 56 30 31 2 5 3 a 9 a a Rock 43 84 11 39 19 31 49 72 81 a a a SR_471 a 33 23 33 19 a a 18 40 28 78 a Hog_I 76 69 133 97 30 15 31 10 42 37 68 55 Rich 192 145 138 138 82 46 100 174 102 105 355 283 LBE 477 561 585 413 381 315 209 146 135 233 415 259 Toshat 29 45 40 61 61 44 11 32 90 29 45 40 Kicco 42 20 43 32 20 15 47 46 44 23 28 67 Pine 38 103 87 101 108 19 19 14 20 75 92 58 Aucilla 52 29 73 40 4 33 44 25 35 35 30 22 StMarks 105 77 146 90 45 26 29 10 75 114 99 83 Sopchop 33 35 76 16 11 7 14 23 7 18 24 19 Camel 10 25 28 37 9 0 5 2 3 14 6 7 Buck_P 2 5 4 12 13 4 2 0 10 2 4 3 Alaqua 2 1 4 3 7 2 0 2 1 1 4 3 SR_85 11 20 17 12 4 6 3 4 8 17 13 10 Battle 31 17 32 16 11 2 8 6 41 6 30 25 Turkey 51 37 45 40 22 13 6 20 2 27 24 23 Black 64 149 88 112 36 36 32 133 41 131 70 88 Twin_R 69 78 93 91 87 8 47 78 87 70 34 76 Stephen 52 85 392 128 141 27 23 48 10 18 142 71 Suwan. 75 67 47 23 18 13 5 20 5 5 28 52 Econfina 96 57 127 153 73 a 21 39 9 51 40 81 a Measurements were missing and were not included in analyses.
76 Table 4 3 Labels used during statistical analyses and to present results. Variable Variable labels (shortened names) Population density a POP_DENS % Males a MALES % Married a MARIED %African American a BLACK % Hispanic or Latino a HISPAN % population with a college degree a AA _GRAD % aged 18 and younger a UNDER_18 % aged 18 to 2 9 a 18_2 9 % aged 30 to 4 9 a 30_ 4 9 % aged 5 0 to 64 a 50 64 % aged 65 and older a 65_UP % households with annual income 25K or less a 25K_LESS % households with annual income 25 to 49 K a 25_49K % househo lds with annual income 50 to 99 K a 50_99K % households with annual income 100 to 199 K a 100_199K % households with annual income 200K or more a 200K_UP % eligible population not in work force UNEMPLOY Surrounding land use (% commercial) b COMM Surroundi ng land use (% residential) b RES Surrounding land use (% urban [vs. rural]) a URBAN Prominent vegetation: pine trees PINES Prominent vegetation: sand hill SAND Prominent vegetation: scrub SCRUB Prominent vegetation: swamp SWAMP Prominent vegetation: h ardwood hammock HARD Prominent vegetation: prairie PRAIRIE Prominent vegetation: pasture PASTURE Vegetation diversity (# of vegetation types that make up at least 10% of buffer area) VEG_DIV Density of water (ln(feet 2 /mile 2 )) LNH2O Latitude (in miles from an arbitrary point of origin) LAT Longitude (in miles from an arbitrary point of origin) LONG Managing agency: state parks STATE_P Managing agency: Cross Florida Greenway CFGW Managing agency: water management district WATER Managing agency: stat e forests SF a Measurements come from communities within 30 mile dri ving distance to each trailhead. b. Measurements are for populations within 15 driving miles of each trailhead.
77 Table 4 3 Continued. Variable Variable labels (shortened names) Manag ing agency: Florida Fish and Wildlife Conservation Commission FWC Managing agency: USFS NF Managing agency: US Department of the Interior INTERIOR Managing agency: US Air Force AIR_FOR Fee charged for access to FNST FEE_AMT Presence of map at trailhea d MAP Presence of toilet BR Loop options for trail LOOP_T Density of roads (ln (mile/mile2)) LNROADS Number of facilities FACILS Table 4 4 Environmental correlates of hiker volume on FNST. Parameter Std. e rror 95% Profile l ikelihood Confidence i nterval Hypothesis t est IRR Lower Upper Wald ChiSquare Sig. (Intercept) 2.514 2.2001 1.798 6.826 1.306 .253 POP_DENS .005 .0007 .003 .006 43.819 .000 1.005 MALES .279 .0515 .178 .380 29.265 .000 1.322 BLACK .091 .0129 .117 .066 50.189 .000 0.913 UNDER_18 .163 .0345 .230 .095 22.183 .000 0.850 50_64 .715 .0784 .868 .561 83.143 .000 0.489 50_99K .028 .0475 .122 .065 .358 .550 0.972 200K_UP 1.204 .2300 1.655 .753 27.413 .000 0.300 UNEMPLOY .118 .0159 .087 .149 55.226 .000 1.125 COMM .083 .0598 .200 .034 1.915 .166 0.920 RES .064 .0156 .034 .095 17.013 .000 1.066 PINES .947 .6227 2.167 .273 2.313 .128 0.388 SAND 1.279 .9564 .596 3.153 1.787 .181 3.593 SCRUB 3.985 .8950 2.231 5.739 19.829 .000 53.785 SWAMP .766 .6394 2 .019 .488 1.434 .231 0.465 HARD 2.145 .4593 3.045 1.245 21.813 .000 0.117 PRAIRIE 1.216 .5649 2.323 .109 4.634 .031 0.296 PASTURE 0 a VEG_DIV .433 .1091 .646 .219 15.722 .000 0.649 LNH2O .042 .0357 .028 .112 1.388 .239 1.043 LONG .001 .0008 .002 .001 .900 .343 0.999 a. Set to zero because this parameter is redundant. AIC: 4745.1 BIC: 4935.0
78 Table 4 4 Continued. Parameter Std. e rror 95% Profile l ikelihood Confidence i nterval Hypothesis t est IRR Lower Upper Wald ChiSquare Si g. STATE_P 2.766 .3848 2.011 3.520 51.640 .000 15.895 CFGW 4.384 .5328 3.340 5.429 67.709 .000 80.158 WATER 2.319 .5045 1.330 3.308 21.132 .000 10.166 SF 3.312 .5302 2.273 4.351 39.014 .000 27.440 FWC 2.180 .5278 1.145 3.214 17.057 .000 8.846 NF .0 35 .4004 .820 .749 .008 .930 0.966 INTERIOR 5.290 .4847 4.340 6.240 119.111 .000 198.343 AIR_FOR 0 a . FACILS .065 .0680 .068 .198 .920 .337 1.067 FEE_AMT .231 .0440 .317 .145 27.677 .000 0.794 MAP .660 .1659 .985 .335 15.811 .000 0.51 7 BR .137 .1310 .393 .120 1.088 .297 0.872 LOOP_T .701 .1409 .424 .977 24.727 .000 2.016 LNROADS .121 .1209 .358 .116 1.000 317 0.886 JAN .987 .1364 .720 1.254 52.341 .000 2.683 FEB 1.041 .1314 .784 1.299 62.794 .000 2.832 MARCH 1.197 .1422 .918 1.475 70.801 .000 3.310 APRIL .954 .1324 .695 1.214 51.938 .000 2.596 MAY .591 .1379 .320 .861 18.354 .000 1.806 JUNE .077 .1416 .355 .200 .298 .585 0.926 JULY 0 a . AUG .314 .1476 .024 .603 4.513 .034 1.369 SEPT .333 .1538 .031 .634 4.677 .031 1.395 OCT .463 .1436 .181 .744 10.383 .001 1.589 NOV .850 .1452 .565 1.134 34.230 .000 2.340 DEC .736 .1348 .472 1.000 29.810 .000 2.088 (Scale) 1 b (Negative binomial) .389 .0261 .341 .444 a. Set to zero because this parameter is redundant. AIC: 4745.1 ; BIC: 4935.0 T able 4 5 Pairwise comparisons of managing agency variables mean predictions for FNST monthly hiker visitation. (I) MNGMT (J) MNGMT Mean d ifference (I J) Std. e rror Sig. STATE_P CFGW 479.61 a 147.946 .001 a. The mean difference is significant at the 0.05 level.
79 Table 45 Continued. (I) MNGMT (J) MNGMT Mean d ifference (I J) Std. e rror Sig. STATE_P WATER 42.69 25.379 .093 SF 86.14 44.161 .051 FWC 52.55 a 25.724 .041 NF 111.32 a 20.109 .000 INTERIOR 13 61.35 a 548.150 .013 AIR_FOR 111.06 a 19.507 .000 CFG STATE_P 479.61 a 147.946 .001 WATER 522.30 a 137.583 .000 SF 393.48 a 128.210 .002 FWC 532.16 a 141.729 .000 NF 590.94 a 143.621 .000 INTERIOR 881.73 584.384 .131 AIR_FOR 590.68 a 143.806 .000 WATER STATE_P 42.69 25.379 .093 CFGW 522.30 a 137.583 .000 SF 128.82 a 34.685 .000 FWC 9.86 14.304 .491 NF 68.64 a 12.004 .000 INTERIOR 1404.03 a 553.541 .011 AIR_FOR 68.38 a 13.480 .000 SF STATE_P 86.14 44.161 .051 CFGW 393.48 a 128.210 .002 WATER 128.82 a 34.685 .000 FWC 138.68 a 34.118 .000 NF 197.46 a 37.437 .000 INTERIOR 1275.21 a 554.401 .021 AIR_FOR 197.20 a 37.669 .000 FWC STATE_P 52.55 a 25.724 .041 CFGW 532.16 a 141.729 .000 WATER 9.86 14.304 .491 SF 138.68 a 34.118 .00 0 a. The mean difference is significant at the 0.05 level.
80 Table 45. Continued. (I) MNGMT (J) MNGMT Mean d ifference (I J) Std. e rror Sig. FWC NF 58.78 a 12.367 .000 INTERIOR 1413.90 a 552.444 .010 AIR_FOR 58.52 a 13.365 .000 NF STATE_P 111. 32 a 20.109 .000 CFGW 590.94 a 143.621 .000 WATER 68.64 a 12.004 .000 SF 197.46 a 37.437 .000 FWC 58.78 a 12.367 .000 INTERIOR 1472.67 a 552.717 .008 AIR_FOR .26 2.968 .930 INTERIOR STATE_P 1361.35 a 548.150 .013 CFGW 881.73 584.384 .131 WA TER 1404.03 a 553.541 .011 SF 1275.21 a 554.401 .021 FWC 1413.90 a 552.444 .010 NF 1472.67 a 552.717 .008 AIR_FOR 1472.41 a 551.672 .008 a. The mean difference is significant at the .05 level. Table 4 6 Pairwise comparisons of prominent vegetation v ariables mean predictions for FNST monthly hiker visitation. (I) VEG_CAT (J) VEG_CAT Mean d ifference (I J) Std. e rror Sig. PINES SAND 291.06 a 129.487 .025 SCRUB 4852.68 a 2100.522 .021 SWAMP 7.01 9.053 .439 HARD 24.61 a 5.252 .000 PRAIRIE 8.32 6.603 .208 PASTURE 55.61 56.472 .325 SAND PINES 291.06 a 129.487 .025 SCRUB 4561.62 a 2003.331 .023 SWAMP 284.05 a 132.508 .032 HARD 315.67 a 130.513 .016 a. The mean difference is significant at the 0 .05 level.
81 Table 46. Continued. (I) VEG_CAT (J) VEG_CAT Mean d ifference (I J) Std. e rror Sig. SAND PRAIRIE 299.38 a 130.719 .022 PASTURE 235.45 175.077 .179 SCRUB PINES 4852.68 a 2100.522 .021 SAND 4561.62 a 2003.331 .023 SWAMP 4845.67 a 2103.085 .021 HARD 4877.29 a 2101.565 .020 PRAIRIE 4860.99 a 2101.782 .021 PASTURE 4797.07 a 2124.390 .024 SWAMP PINES 7.01 9.053 .439 SAND 284.05 a 132.508 .032 SCRUB 4845.67 a 2103.085 .021 HARD 31.62 a 10.321 .002 PRAIRIE 15.33 11.625 .187 PASTURE 48.60 56.382 .389 HARD PINES 24. 61 a 5.252 .000 SAND 315.67 a 130.513 .016 SCRUB 4877.29 a 2101.565 .020 SWAMP 31.62 a 10.321 .002 PRAIRIE 16.29 a 5.261 .002 PASTURE 80.22 54.560 .141 PRAIRIE PINES 8.32 6.603 .208 SAND 299.38 a 130.719 .022 SCRUB 4860.99 a 2101.782 .021 SWAMP 15.33 11.625 .187 HARD 16.29 a 5.261 .002 PASTURE 63.93 54.529 .241 a. The mean difference is significant at the 0 .05 level. Table 4 7 Pairwise comparisons month variables mean response differences. (I) MONTH (J) MONTH Mean Difference (I J) Std. Error Sig. AUG SEP 1.33 11.055 .904 OCT 11.09 11.222 .323 NOV 48.95 a 15.128 .001 DEC 36.28 a 12.539 .004 JAN 66.33 a 15.488 .000 a. The mean difference is significant at the 0 .05 level.
82 Table 47. Continued. (I) MONTH (J) MONTH Mean d ifference (I J) Std. e rror Sig. AUG FEB 73.87 a 15.543 .000 MAR 97.87 a 20.168 .000 APR 61.98 a 15.037 .000 MAY 22.03 11.652 .059 JUN 22.34 a 9.131 .014 JUL 18.58 a 9.199 .043 SEP AUG 1.33 11.055 .904 OCT 9.76 11.7 25 .405 NOV 47.62 a 15.585 .002 DEC 34.95 a 12.976 .007 JAN 65.00 a 15.806 .000 FEB 72.54 a 15.904 .000 MAR 96.55 a 20.365 .000 APR 60.65 a 15.306 .000 MAY 20.70 12.145 .088 JUN 23.67 a 9.717 .015 JUL 19.91 a 9.873 .044 OCT AUG 1 1.09 11.222 .323 SEP 9.76 11.725 .405 NOV 37.85 a 15.203 .013 DEC 25.19 a 12.721 .048 JAN 55.23 a 15.339 .000 FEB 62.78 a 15.342 .000 MAR 86.78 a 19.706 .000 APR 50.89 a 14.745 .001 MAY 10.94 11.902 .358 JUN 33.43 a 10.224 .001 JUL 29.6 8 a 10.216 .004 NOV AUG 48.95 a 15.128 .001 SEP 47.62 a 15.585 .002 OCT 37.85 a 15.203 .013 DEC 12.67 15.717 .420 JAN 17.38 17.606 .324 FEB 24.92 17.271 .149 MAR 48.93 a 21.363 .022 APR 13.03 16.710 .435 MAY 26.92 15.155 .076 JUN 71.28 a 1 4.825 .000 JUL 67.53 a 14.792 .000 a. The mean difference is significant at the 0 .05 level
83 Table 4 7 Continued. (I) MONTH (J) MONTH Mean d ifference (I J) Std. e rror Sig. DEC AUG 36.28 a 12.539 .004 SEP 34.95 a 12.976 .007 OCT 25. 19 a 12.721 .048 NOV 12.67 15.717 .420 JAN 30.05 15.991 .060 FEB 37.59 a 15.726 .017 MAR 61.59 a 19.927 .002 APR 25.70 15.047 .088 MAY 14.25 12.737 .263 JUN 58.62 a 11.849 .000 JUL 54.86 a 11.829 .000 JAN AUG 66.33 a 15.488 .000 SEP 65.00 a 15.806 .000 OCT 55.23 a 15.339 .000 NOV 17.38 17.606 .324 DEC 30.05 15.991 .060 FEB 7.54 17.584 .668 MAR 31.55 21.056 .134 APR 4.35 16.776 .796 MAY 44.30 a 15.277 .004 JUN 88.66 a 15.133 .000 JUL 84.91 a 15.085 .000 F EB AUG 73.87 a 15.543 .000 SEP 72.54 a 15.904 .000 OCT 62.78 a 15.342 .000 NOV 24.92 17.271 .149 DEC 37.59 a 15.726 .017 JAN 7.54 17.584 .668 MAR 24.00 20.594 .244 APR 11.89 16.171 .462 MAY 51.84 a 15.099 .001 JUN 96.21 a 15.485 .000 JUL 92.45 a 15.479 .000 a. The mean difference is significant at the 0 .05 level.
84 Table 47. Continued. (I) MONTH (J) MONTH Mean d ifference (I J) Std. e rror Sig. MAR AUG 97.87 a 20.168 .000 SEP 96.55 a 20.365 .000 OCT 86.78 a 19.706 .000 NOV 48.93 a 21.363 .022 DEC 61.59 a 19.927 .002 JAN 31.55 21.056 .134 FEB 24.00 20.594 .244 APR 35.89 19.867 .071 MAY 75.85 a 19.674 .000 JUN 120.21 a 20.278 .000 JUL 116.46 a 20.186 .000 APR AUG 61.98 a 15.037 .000 SEP 60.65 a 15.306 .000 OCT 50.89 a 14.745 .001 NOV 13.03 16.710 .435 DEC 25.70 15.047 .088 JAN 4.35 16.776 .796 FEB 11.89 16.171 .462 MAR 35.89 19.867 .071 MAY 39.95 a 14.382 .005 JUN 84.32 a 14.959 .000 JUL 80.56 a 14.903 .000 MAY AUG 22.03 11 .652 .059 SEP 20.70 12.145 .088 OCT 10.94 11.902 .358 NOV 26.92 15.155 .076 DEC 14.25 12.737 .263 JAN 44.30 a 15.277 .004 FEB 51.84 a 15.099 .001 MAR 75.85 a 19.674 .000 APR 39.95 a 14.382 .005 JUN 44.37 a 10.772 .000 JUL 40. 61 a 10.782 .000 a. The mean difference is significant at the .05 level.
85 Table 47. Continued. (I) MONTH (J) MONTH Mean d ifference (I J) Std. e rror Sig. JUN AUG 22.34 a 9.131 .014 SEP 23.67 a 9.717 .015 OCT 33.43 a 10.224 .001 NOV 71. 28 a 14.825 .000 DEC 58.62 a 11.849 .000 JAN 88.66 a 15.133 .000 FEB 96.21 a 15.485 .000 MAR 120.21 a 20.278 .000 APR 84.32 a 14.959 .000 MAY 44.37 a 10.772 .000 JUL 3.75 6.889 .586 a. The mean difference is significant at the .05 lev el. Table 4 8. Final prediction model for monthly hiker volume on FNST. Parameter Std. e rror 95% Wald c onfidence i nterval Hypothesis t est Lower Upper Wald Chi Square Sig. (Intercept) 13093.85 1141.524 10856.5 15331.2 131.57 .000 POP_DENS 0.01 0.0 01 0.01 0.02 99.09 .000 MALES 2.34 0.239 1.9 2.8 95.30 .000 BLACK 0.11 0.022 0.07 0.16 27.91 .000 AAGRAD 0.44 0.034 0.5 0.4 173.46 .000 Under_18 7.65 2.864 13.3 2.0 7.14 .008 18_29 8.46 2.858 14.1 2.9 8.77 .003 30_49 10.99 2.820 16.5 5.5 15.18 .000 50_64 9.40 2.877 15.0 3.8 10.67 .001 65UP 7.80 2.848 13.4 2.2 7.50 .006 25K_LESS 123.72 11.352 146.0 101.5 118.77 .000 25_49K 122.70 11.315 144.9 100.5 117.58 .000 50_99K 124.07 11.397 146.4 101.7 118.52 .000 100_200K 119.5 9 11.060 141.3 97.9 116.92 .000 200K_UP 130.21 11.900 153.5 106.9 119.73 .000 a. Set to zero because this parameter is redundant.
86 Table 4 8. Continued. Parameter Std. e rror 95% Wald c onfidence i nterval Hypothesis t est Lower Upper Wald Chi S quare Sig. UNEMPLOY 0.24 0.050 0.34 0.14 22.90 .000 COMM 0.14 0.056 0.03 0.25 6.31 .012 URBAN 0.07 0.011 0.05 0.09 38.43 .000 PINES 2.91 0.525 1.9 3.9 30.64 .000 SAND 3.30 0.767 1.8 4.8 18.54 .000 SCRUB 0.21 0.741 1.7 1.2 0.08 .777 SWAMP 4.08 0 .572 3.0 5.2 50.73 .000 HARD 2.47 0.453 3.4 1.6 29.68 .000 PRAIRIE 2.95 0.505 2.0 3.9 34.14 .000 PASTURE 0 a . VEG_DIV 0.41 0.087 0.58 0.24 22.22 .000 LNH2O 0.32 0.041 0.24 0.40 63.52 .000 LAT 0.10 0.007 0.09 0.12 182.52 .000 State_P 10 .41 0.653 9.1 11.7 253.87 .000 CFGW 13.17 0.912 11.4 15.0 208.72 .000 WATER 11.42 0.864 9.7 13.1 174.67 .000 SF 12.41 0.869 10.7 14.1 203.91 .000 FWC 18.87 1.467 16.0 21.7 165.36 .000 NF 8.40 0.673 7.1 9.7 155.50 .000 INTERIOR 14.71 1.021 12.7 16.7 2 07.38 .000 AIR_FOR 0 a . FEE_AMT 1.04 0.095 0.85 1.23 118.95 .000 MAP 3.96 0.353 3.3 4.7 126.16 .000 BR 1.32 0.129 1.1 1.6 105.14 .000 LOOP_T 2.13 0.262 2.6 1.6 66.11 .000 LNROADS 1.68 0.149 2.0 1.4 126.77 .000 AUG 0.29 0.126 0.05 0. 54 5.39 .020 SEPT 0.33 0.140 0.06 0.61 5.73 .017 OCT 0.44 0.121 0.20 0.67 12.96 .000 NOV 0.80 0.120 0.56 1.03 43.96 .000 a. Set to zero because this parameter is redundant.
87 Table 4 8. Continued. Parameter Std. e rror 95% Wald c onfidence i nterval Hypo thesis t est Lower Upper Wald Chi Square Sig. DEC 0.72 0.107 0.51 0.93 44.85 .000 JAN 1.00 0.115 0.78 1.23 75.64 .000 FEB 1.09 0.115 0.86 1.31 88.89 .000 MARCH 1.22 0.118 0.99 1.46 107.54 .000 APRIL 1.02 0.122 0.78 1.26 69.81 .000 MAY 0.67 0.128 0 .42 0.92 27.39 .000 JUNE 0.03 0.121 0.26 0.21 0.04 .832 JULY 0 a . (Scale) 1 b (Negative binomial) 1 a. Set to zero because this parameter is redundant. b. Fixed at the displayed value. AIC:4856.8; BIC:5067.8 Figure 4 3 Scatterplot of prediction models predicted hiker visits versus the standardized residuals for those predictions.
88 Figure 4 4 Actual versus predicted counts for monthly hiker visitation for 43 access points on the FNST.
89 Figure 4 5 Scatterpl ot comparing the actual and predicted monthly hiker counts for 503 months on 43 FNST access points.
90 CHAPTER 5 DISCUSSION AND MANAG EMENT IMPLICATIONS This chapter begins by returning to the theoretical underpinnings of this study which were discussed in C hapter 2. Specifically, it examines how the research in this thesis supports a Behavioral Ecological Model for hiking on trails. The discussion moves on to examine the individual results of the correlation analysis and how the results can inform recreation management in Florida. In addition, the results from these correlation studies are checked against qualitative understandings of the phenomenon in question. The discussion then examines the question of causation in relation to environmental correlation analyses. Next, the prediction model is examined for its usefulness both to hiking trail managers and as a way to validate the associations established in the correlation analysis. Suggestions for future research are presented throughout the chapter. Examini ng the Environmental Correlations of FNST Hiker Volume The information provided in Chapter 4 helps to build new lines of evidence about the environmental correlates of hiking in natural settings. The results can help researchers and managers better underst and how the social, environmental, and managerial backdrops for hiking trails are associated with how many visitors hiking trails will receive. This work builds on the work of other researchers who have examined these relationships for other forms of outdoor physical activity in different settings. Support for the Behavioral Ecological Model The benefits of outdoor recreation have been understood for decades and as a result, entire management constructs for maximizing the benefits from outdoor recreation have been established (e.g. Outcomes Focused Management). If it is within
91 the power of recreation managers to increase use to their areas, then doing so is likely to benefit whole segments of society, beyond just the user group themselves. If increases in ou tdoor recreation are deemed desirable by recreation managers and society, then research is needed to determine whether environmental changesdo increase the likelihood of more active behavioral choices (Owen et al., 2004, p. 67). This study helped add to the mounting evidence from the environmental correlate literature that there are strong connections between peoples habitual physical activity behavior and the settings in which they live, work, and play. Bargh and Chartrand (1999, p. 462) present the t hesis that most of a persons everyday life is determined not by their conscious intentions and deliberate choices but by mental processes that are put into motion by features of the environment. This is the foundation of the Behavioral Ecological Model (BEM), and this is why measuring the relationships between human behavior and the environment is important. This research identifies the mechanisms by which to steer human behavior on an aggregate scale toward desirable outcomes. Raymore (2002, p. 39) call s these mechanisms leisure facilitators which can encourage or enhance participation in specific leisure activities. Equally important are the leisure constraints which prevent or discourage individuals from engaging in particular recreation pursuits The research in this thesis was the means toward identifying the aspects of the environment that increases or decreases hiker participation. The particular environment examined was the Florida National Scenic Trail, but by association, the results from t his study can be extended to all hiking trails in Florida and perhaps beyond. Future research examining environmental
92 correlates of hiking trails specifically are needed to see if these results are unique to Florida, national scenic trails, or this specifi c Trail only. Examining Specific Environmental Correlates: Management Implications Population density was consistent with prior studies by being associated with increases in Trail use (Lovett, Brainard, & Bateman, 1997; Lindsey et al., 2007). The Florida N ational Scenic Trail can provide a remote and primitive experience, however, sections of the Trail that are closer to dense populations will receive higher use. Also, trails located near populations where men make up higher percentages of the population may also receive higher use. Prior studies showed that being a male made a survey respondent more likely to engage in outdoor physical activities (Troped et al., 2001; Reynolds et al., 2007). This could be the reason communities with more men are associate w ith higher visitation. Exposure to nature and how it can impact well being and environmental attitudes have received a lot of attention since Richard Louvs Last Child in the Woods (200 5 ) was published. In a review of sources for human environmental sensi tivity, Chawla (1998, p. 381) found that contact with natural areas has emerged as one of the most significant influences i n all of the studies reviewed on peoples environmental sensitivity. In other words, the amount of time people spend in natural areas is a key determinate of how much they care about the environment. The negative correlation between the proportion of children 18 and under in communities around the Trail and Trail visitation is consistent with Louvs concerns about children recreating outside. Survey data from the FNST shows that 65% of Trail visitors have no children at home (Albritton, Wan, & Stein, 2008). This indicates that children are not coming to the Trail.
93 Trail managers should do more to attract and market to this age group t o help change the direction of this correlation. The proportion of African Americans as a correlate of visitation was not consiste nt with prior research (Lindsey et al., 2007). For the FNST, populations with more African Americans per capita produce fewer hikers In a study of an urban greenway in Indianapolis, the as sociation was positive (Lindsey et al., 2007). Further research specific to primitive hiking trails is required to see if this association is consistent among different trails. The proportion of surrounding commercial land use did not show a significant association with hiker volume. However, the proportion of residential land use around the Trail was positively associated. This has indications for Trail placement in the future. Placing re rou tes of the FNST near residential areas is likely to increase visitation, however, commercial land use is unlikely to have an impact. This indicates that having places of work near the Trail is not important for impacting use. The positive correlation with unemployment seems to support this notion. Whether or not there are jobs near the Trail, people will visit it as long as it is within a reasonable distance from their homes. Looking at the environmental variables, the area of water features along the Trai l had no significant association with visitation. This is contradictory to survey data from the Trail which has shown that 72% of hikers said that seeing water features was an important reason for their visit to the Trail (Albritton, Wan, & Stein 2008). T his could be a case where the unit of measure (area of water) could be flawed. Perhaps Trail visitors just like seeing water along the Trail and the amount they encounter is not what is important. Future research could examine waters association with visi tation by using
94 simply the presence of a water feature instead of the area. Another result was that scrub w as the vegetation type associated with the highest Trail use. Scrub is a dense, dry habitat where visibility is usually limited to strictly the trail path itself. Peoples attraction to these sections of the Trail may be induced by the fact that these habitats are found mostly in Florida and are one of the most endangered ecosystem types in Florida (Scrub, n.d.). In other words, they are uniquely Flori dian and thus offer a unique experience that hiking trails in other states cannot offer. FNST and other Florida recreation managers should emphasize and promote Floridas unique habitats and endemic species to attract visitors. Several managerial variables showed significant associations with hiker volume after controlling for the other variables in the final environmental correlate model. The presence of loop trails indicates a preference for variety for hikers. This correlation is consistent with survey data. For the years when hikers were asked if loop trails were important to them, they consistently rated them as more important than not and also preferable to linear trails (see Albritton, Denny, & Stein, 2006; Albritton Denny, & Stein, 2007; Albritton, Wan, & Stein, 2008; and Wan, Cucinella, & Stein, 2009). Facilities were another managerial characteristic correlated with higher use. This too was consistent with what researchers already know about FNST hikers. When hikers were asked about improvements t hey would like to see on the FNST, they consistently have suggested either additional or improved facilities. Although FNST hikers are mostly engaging in an activity in primitive settings, offering facilities like benches, kiosks, and trash cans does not s eem to increase or decrease visitation. Also the presence of bathrooms at the trailhead was not significantly associated with use. This could be because when there
95 were not bathrooms at the trailhead, there may have been bathrooms nearby that hikers can q uickly access by car. This information may be useful to managers because among all the facilities at trailheads, bathrooms are the most expensive to construct but apparently not a necessity for Trail visitors. Limitations: Causation and Spatial Autocorrela tion In modern statistical work, a probabilistic view of causation is often favored over the deterministic. The difference is that rather than events arising necessarily from certain causes, the causes increase the probability of events occurring (Goldthor pe, 2001). To move from an association to causation, the relationship between X and Y must be shown to be authentic (and not spurious) by showing that Xs influence on Y is not weakened by adding new variables into the model and thus controlling for their effect. In this research all possible variables that impact FNST hiker visitation over time could not be examined due to resource and time limitations and is probably not possible. Weather data, temporary Trail closures, and information on how each access point is advertised were all variables that may have had an impact on visitation, but the cost and logistical requirements for obtaining this information prevented their input into the analysis. However, an important rejoinder to this view is that all o f the potential influences could never really be examined. In observational studies, there is no way for a researcher to ever truly know and control for all potential influences on the phenomenon being studied. Therefore, in this discussion, the significant correlates of hiker volume are treated and discussed as influences of hiker numbers. This allows for much more meaningful discussion that can better inform management and guide future research than removing all terms that imply causation.
96 In this study, researchers had no control over the FNST or its surrounding communities. This study instead, in the words of Freedman (1999, p. 247) relied on observations of existing circumstances where nature has conducted an experiment, assigning different treatment s to different areas. In this study, an experiment that changed aspects of the Trail or the communities around it in order to examine the impact on use would be impractical. However, some of the variables in this study could be manipulated and the results examined. For the sake of future research, the effect of changes in the following variables could be examined and their impact on visitation measured and compared to the coefficients in this study: managing agency, fee charged, presence of a map at the tr ailhead, and the presence of loop trails. Spatial autocorrelation is defined as the property of random variables taking values, at pairs of locations a certain distance apart, that are more similar (positive autocorrelation) or less similar (negative auto correlation) than expected for randomly associated pairs of observations (Legendre, 1993, p. 1659). This is a violation of the statistical assumption for regression that observations for variables are independent. In this research, the values for some of the observations for independent variables representing social characteristics of the FNST were measured using population data from overlapping census blocks. This could potentially produce spatial autocorrelation affects such as inflated confidence interv als. One method of correcting for spatial autocorrelation among the continuous variables in the model would be to use clustering analyses to create homogeneous subsets of observations and thus control for the geographic similarities (Legendre, 1993).
97 The Prediction Model As mentioned by Freedman (1999), being able to compare regression based models to reality provides a way for scientists to evaluate the models reliability. The prediction portion of this study is potentially useful for managers, but is al so useful for helping to validate the reliability of the environmental correlates analysis. This is true because the fact that the prediction model was relatively good at predicting use on the FNST, there must be an association between use levels and the s ignificant variables within the model. As stated in Chapter 3 of this thesis, some predictions were exactly correct while others were orders of magnitude off for monthly trailhead visitation. Looking at a graph of the actual versus predicted monthly visits ( Figure s 4 4 and 45 ), it is apparent that the model does well for predicting within an order of magnitude. This prediction model could potentially serve as one tool in the decisionmaking process for rerouting sections of the FNST that are along roads i nto natural areas. Areas with higher predicted use could be chosen over areas that predict far lower use. The results of the prediction evaluation indicated that the Cross Florida Greenways predictions performed the best among the eight managing agencies that manage the individual access points of the FNST. It would be interesting to test the model on a trail within the Cross Florida Greenway other than the FNST and then compare the results to actual measurements of visitation. However, as of the writing of this thesis, no reliable measurement of visitation is occurring on these trails. Any professional interested in adapting this model for other trails in Florida would need very limited equipment and close to no investment costs. All that is needed is a simple spreadsheet software program, and a GIS program, both of which can be
98 downloaded for free (e.g. FreeGIS.org ). The efficacy of the model could be tested through personal observations of visitors or mandatory t rail registers. Conclusion This work was an extension of the growing body of literature supporting Behavioral Ecological Models for outdoor physical activity. Specifically, the research was designed to strengthen the understanding of the connections between the social, environmental, and managerial settings that recreational trails are located in and the number of visits these trails receive. In other words, it sought to discern which characteristics of the environment promote or discourage hiking. Towards that goal, an environmental correlate analysis of hiking on the FNST was conducted that identified fifteen aspects of the Trail and the communities around it that had either a positive or negative association with the number of hikers visiting the Trail ea ch month. Aspects of management, the biophysical environment, and the communities around the Trail all showed significant relationships with Trail visit numbers. By comparing the results of this study with what is already known about the motivations, demographics, and preferences of FNST hikers, some important management implications were derived. For instance, although hikers often cite bathrooms or added facilities as potential improvements to the Trail, the presence or absence of these does not actually impact visitation. Also, it is apparent that children under 18 are not using the FNST. These implications have the potential to extend beyond the FNST to other hiking trails. The management implications include ways for trail managers to increase use, pro vide better or more targeted marketing for their trails, and decide where to put new sections of trail.
99 Understanding these relationships can help researchers and managers identify manipulable settings to encourage outdoor recreation. Previous research has illustrated the importance of identifying correlates of specific types o f outdoor activity (Giles Corti et al., 2005). Therefore, more research is needed to strengthen the findings in this study and to determine if the correlates identified from this res earch remain consistent for different types of hiking trails, trails in different geographic locations, or for trails surrounded by human populations with different cultural backgrounds. Continued investigations into the correlates of hiking trail use will further solidify the BEM for hiking and the concept of Ecological Systems Theory in general as a theoretical foundation for understanding human behavior.
100 APPENDIX PREDICTION RESULTS FOR INDIVIDUAL MANAGING AGENC IES Table A 1. Prediction errors for st ate parks FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error GoldEnt JAN 76 71 5 7.04 GoldEnt FEB 83 52 31 59.62 GoldEnt MAR 95 82 13 15.85 GoldEnt APR 77 109 32 29.36 GoldEnt MAY 55 64 9 14.06 GoldEnt JUN 27 31 4 12.90 GoldEnt JUL 28 21 7 33.33 GoldEnt AUG 37 19 18 94.74 GoldEnt SEP 39 44 5 11.36 GoldEnt OCT 43 61 18 29.51 GoldEnt NOV 62 57 5 8.77 GoldEnt DEC 57 74 17 22.97 Stephen JAN 105 52 53 101.92 Stephen FEB 114 85 29 34.12 Stephen MAR 131 392 261 66.58 Stephen APR 107 128 21 16.41 Stephen MAY 75 141 66 46.81 Stephen JUN 38 27 11 40.74 Stephen JUL 38 23 15 65.22 Stephen AUG 52 48 4 8.33 Stephen SEP 54 10 44 440.00 Stephen OCT 60 18 42 233.33 Stephen NOV 85 142 57 40.14 Step hen DEC 79 71 8 11.27 Big_Oak JAN 39 75 36 48.00 Big_Oak FEB 42 67 25 37.31 Big_Oak MAR 49 47 2 4.26 Big_Oak APR 40 23 17 73.91 Big_Oak MAY 28 18 10 55.56 Big_Oak JUN 14 13 1 7.69 Big_Oak JUL 14 5 9 180.00 Big_Oak AUG 19 20 1 5.00 Big_Oak SEP 20 5 15 300.00 Big_Oak OCT 22 5 17 340.00 Big_Oak NOV 32 28 4 14.29 Big_Oak DEC 29 52 23 44.23 Mean 26.0 71.0 Median 16.0 35.7
101 Table A 2. Prediction errors for Cross Florida Greenway. FNST Access Point Month # of Hikers Predicted Actual # of Hi kers Absolute Error Percent Error LB JAN 409 377 32 8.49 LB FEB 444 379 65 17.15 LB MAR 510 366 144 39.34 LB APR 415 315 100 31.75 LB MAY 292 216 76 35.19 LB JUN 146 159 13 8.18 LB JUL 150 134 16 11.94 LB AUG 201 204 3 1.47 LB SEP 209 182 27 14.84 LB OCT 232 296 64 21.62 LB NOV 333 361 28 7.76 LB DEC 308 302 6 1.99 Santos JAN 304 244 60 24.59 Santos FEB 330 310 20 6.45 Santos MAR 380 310 70 22.58 Santos APR 309 238 71 29.83 Santos MAY 218 209 9 4.31 Santos JUN 109 114 5 4.39 Santos JUL 11 2 119 7 5.88 Santos AUG 149 185 36 19.46 Santos SEP 156 165 9 5.45 Santos OCT 172 312 140 44.87 Santos NOV 247 322 75 23.29 Santos DEC 229 270 41 15.19 Rod_E JAN 89 56 33 58.93 Rod_E FEB 97 42 55 130.95 Rod_E MAR 112 61 51 83.61 Rod_E APR 91 116 2 5 21.55 Rod_E MAY 64 102 38 37.25 Rod_E JUN 32 79 47 59.49 Rod_E JUL 33 28 5 17.86 Rod_E AUG 44 108 64 59.26 Rod_E SEP 46 13 33 253.85 Rod_E OCT 51 39 12 30.77 Rod_E NOV 73 56 17 30.36 Rod_E DEC 67 71 4 5.63 Rod_W JAN 29 21 8 38.10 Rod_W FEB 32 3 3 1 3.03
102 Table A 2. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Rod_W MAR 37 38 1 2.63 Rod_W APR 30 24 6 25.00 Rod_W MAY 21 10 11 110.00 Rod_W JUN 10 10 0 0.00 Rod_W JUL 11 7 4 57.14 Rod_W AUG 14 22 8 36.36 Rod_W SEP 15 17 2 11.76 Rod_W OCT 17 20 3 15.00 Rod_W NOV 24 9 15 166.67 Rod_W DEC 22 16 6 37.50 Mean 32.6 35.39 Median 18.5 22.10 Table A 3. Prediction errors for w ater management districts. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Rice_C JAN 29 48 19 39.58 Rice_C FEB 32 58 26 44.83 Rice_C MAR 36 40 4 10.00 Rice_C APR 30 27 3 11.11 Rice_C MAY 21 31 10 32.26 Rice_C JUN 10 4 6 150.00 Rice_C JUL 11 7 4 57.14 Rice_ C AUG 14 6 8 133.33 Rice_C SEP 15 2 13 650.00 Rice_C OCT 17 10 7 70.00 Rice_C NOV 24 43 19 44.19 Rice_C DEC 22 24 2 8.33 Bluff JAN 110 70 40 57.14 Bluff FEB 119 88 31 35.23 Bluff MAR 137 84 53 63.10 Bluff APR 112 36 76 211.11 Bluff MAY 79 29 50 172.41 Bluff JUN 39 48 9 18.75 Bluff JUL 40 99 59 59.60 Bluff AUG 54 108 54 50.00 Bluff SEP 56 109 53 48.62 Bluff OCT 62 28 34 121.43 Bluff NOV 89 40 49 122.50 Bluff DEC 83 71 12 16.90
103 Table A 3. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Hickory JAN 40 80 40 50.00 Hickory FEB 43 42 1 2.38 Hickory MAR 50 47 3 6.38 Hickory APR 40 11 29 263.64 Hickory MAY 28 27 1 3.70 Hickory JUN 14 17 3 17.65 Hickory JUL 15 29 14 48.28 Hickory AUG 20 12 8 66.67 Hickory SEP 20 11 9 81.82 Hickory OCT 23 10 13 130.00 Hickory NOV 32 48 16 33.33 Hickory DEC 30 20 10 50.00 River JAN 28 56 28 50.00 River FEB 23 30 7 23.33 River MAR 16 31 15 48.39 River APR 8 2 6 300.00 River MAY 8 5 3 60.00 River JUN 11 3 8 266.67 River JUL 13 9 4 44.44 Rock AUG 78 43 35 81.40 Rock SEP 85 84 1 1.19 Rock OCT 98 11 87 790.91 Rock NOV 79 39 40 102.56 Rock DEC 56 19 37 194.74 Rock JAN 28 31 3 9.68 Rock FEB 29 49 20 40.82 Rock MAR 38 72 34 47.22 Rock APR 40 81 41 50.62 SR_471 MAY 48 33 15 45.45 SR_471 JUN 55 23 32 139.13 SR_471 JUL 45 33 12 36.36 SR_471 AUG 32 19 13 68.42 SR_471 SEP 22 18 4 22.22 SR_471 OCT 23 40 17 42.50 SR_471 NOV 25 28 3 10.71 SR_471 DEC 36 78 42 53.85 Kicco JAN 58 42 16 38.10 Kicco FEB 63 20 43 215.00
104 Table A 3. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Kicco MAR 73 43 30 69.77 Kicco APR 59 32 27 84.38 Kicco MAY 42 20 22 110.00 Kicco JUN 21 15 6 40.00 Kicco JUL 21 47 26 55.32 Kicco AUG 29 46 17 36.96 Kicco SEP 30 44 14 31.82 Kicco OCT 33 23 10 43.48 Kicco NOV 47 28 19 67.86 Kicco DEC 44 67 23 34.33 Econfina JAN 84 96 12 12.50 Econfina FEB 92 57 35 61.40 Econfina MAR 105 127 22 17.32 Econfina APR 86 153 67 43.79 Econfina MAY 60 73 13 17.81 Econfina JUN 31 21 10 47.62 Econfina JUL 42 39 3 7.69 Econfina AUG 43 9 34 377.78 Econfina SEP 48 51 3 5.88 Econfina OCT 69 40 29 72.50 Econfina NOV 64 81 17 2 0.99 Mean 21.6 84.6 Median 16.0 48.4 Table A 4. Prediction errors for s tate forests. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Etoniah JAN 29 51 22 43.14 Etoniah FEB 32 21 11 52.38 Etoniah MA R 36 36 0 0.00 Etoniah APR 30 27 3 11.11 Etoniah MAY 21 9 12 133.33 Etoniah JUN 10 8 2 25.00 Etoniah JUL 11 8 3 37.50 Etoniah AUG 14 17 3 17.65 Etoniah SEP 15 12 3 25.00 Etoniah OCT 16 13 3 23.08 Etoniah NOV 24 41 17 41.46 Etoniah DEC 22 17 5 29.4 1 Bluff JAN 110 70 40 57.14
105 Table A 4. C ontinued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error SR_44 JAN 22 22 0 0.00 SR_44 FEB 24 60 36 60.00 SR_44 MAR 27 47 20 42.55 SR_44 APR 22 23 1 4.35 SR_44 MAY 16 18 2 11.11 SR_44 JUN 8 15 7 46.67 SR_44 JUL 8 5 3 60.00 SR_44 AUG 11 12 1 8.33 SR_44 SEP 12 1 11 1100.00 SR_44 OCT 13 1 12 1200.00 SR_44 NOV 19 1 18 1800.00 SR_44 DEC 16 15 1 6.67 SR_46 JAN 149 177 28 15.82 SR_46 FEB 162 230 68 29.57 SR_46 MAR 186 184 2 1.09 SR_46 APR 151 151 0 0.00 SR_46 MAY 107 106 1 0.94 SR_46 JUN 53 53 0 0.00 SR_46 JUL 55 51 4 7.84 SR_46 AUG 73 59 14 23.73 SR_46 SEP 76 57 19 33.33 SR_46 OCT 85 76 9 11.84 SR_46 NOV 121 133 12 9.02 SR_46 DEC 112 101 11 10.89 Hog _I JAN 72 76 4 5.26 Hog_I FEB 78 69 9 13.04 Hog_I MAR 90 133 43 32.33 Hog_I APR 73 97 24 24.74 Hog_I MAY 51 30 21 70.00 Hog_I JUN 26 15 11 73.33 Hog_I JUL 26 31 5 16.13 Hog_I AUG 35 10 25 250.00 Hog_I SEP 37 42 5 11.90 Hog_I OCT 41 37 4 10.81 Hog _I NOV 58 68 10 14.71 Hog_I DEC 54 55 1 1.82 Rich JAN 219 192 27 14.06 Rich FEB 238 145 93 64.14 Rich MAR 273 138 135 97.83
106 Table A 4. C ontinued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Rich APR 222 138 84 60.87 Rich MAY 157 82 75 91.46 Rich JUN 78 46 32 69.57 Rich JUL 80 100 20 20.00 Rich AUG 108 174 66 37.93 Rich SEP 112 102 10 9.80 Rich OCT 124 105 19 18.10 Rich NOV 178 355 177 49.86 Rich DEC 165 283 118 41.70 LBE JAN 486 477 9 1.89 LBE FEB 528 561 33 5.88 LBE MAR 607 585 22 3.76 LBE APR 494 413 81 19.61 LBE MAY 348 381 33 8.66 LBE JUN 174 315 141 44.76 LBE JUL 178 209 31 14.83 LBE AUG 239 146 93 63.70 LBE SEP 249 135 114 84.44 LBE OCT 276 233 43 18.45 LBE NOV 396 415 19 4.58 LBE DEC 366 259 107 41.31 Pine JAN 77 38 39 102.63 Pine FEB 84 103 19 18.45 Pine MAR 96 87 9 10.34 Pine APR 78 101 23 22.77 Pine MAY 55 108 53 49.07 Pine JUN 28 19 9 47.37 Pine JUL 28 19 9 47.37 Pine AUG 38 14 24 171.43 Pine SEP 40 2 0 20 100.00 Pine OCT 44 75 31 41.33 Pine NOV 63 92 29 31.52 Pine DEC 58 58 0 0.00 Black JAN 119 64 55 85.94 Black FEB 129 149 20 13.42 Black MAR 149 88 61 69.32 Black APR 121 112 9 8.04 Black MAY 85 36 49 136.11 Black JUN 43 36 7 19.44
107 Table A 4 C ontinued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Black JUL 44 32 12 37.50 Black AUG 59 133 74 55.64 Black SEP 61 41 20 48.78 Black OCT 68 131 63 48.09 Black NOV 97 70 27 38.57 Black DEC 90 88 2 2.27 Twin_R JAN 91 69 22 31.88 Twin_R FEB 98 78 20 25.64 Twin_R MAR 113 93 20 21.51 Twin_R APR 92 91 1 1.10 Twin_R MAY 65 87 22 25.29 Twin_R JUN 32 8 24 300.00 Twin_R JUL 33 47 14 29.79 Twin_R AUG 45 78 33 42.31 Twin_R SEP 46 87 41 47.13 Twin_ R OCT 51 70 19 27.14 Twin_R NOV 74 34 40 117.65 Twin_R DEC 68 76 8 10.53 Mean 28.1 76.1 Median 19.0 26.4 Table A 5. Prediction errors for Florida Fish and Wildlife Conservation Commission. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error US_192 JAN 18 11 7 63.64 US_192 FEB 19 13 6 46.15 US_192 MAR 22 27 5 18.52 US_192 APR 18 34 16 47.06 US_192 MAY 13 25 12 48.00 US_192 JUN 6 13 7 53.85 US_192 JUL 7 4 3 75.00 US_192 AUG 9 10 1 10.00 US_192 SE P 9 9 0 0.00 US_192 OCT 10 11 1 9.09 US_192 NOV 14 4 10 250.00 US_192 DEC 13 9 4 44.44 Crab JAN 113 65 48 73.85 Crab FEB 122 49 73 148.98 Crab MAR 141 126 15 11.90
108 Table A 5. C ontinued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Crab APR 114 70 44 62.86 Crab MAY 81 66 15 22.73 Crab JUN 40 46 6 13.04 Crab JUL 41 25 16 64.00 Crab AUG 55 27 28 103.70 Crab SEP 58 65 7 10.77 Crab OCT 64 83 19 22.89 Crab NOV 92 114 22 19.30 Crab DEC 85 93 8 8.60 3 _Lake JAN 54 84 30 35.71 3 _Lake FEB 59 51 8 15.69 3 _Lake MAR 68 32 36 112.50 3 _Lake APR 55 33 22 66.67 3 _Lake MAY 39 18 21 116.67 3 _Lake JUN 19 16 3 18.75 3 _Lake JUL 20 35 15 42.86 3 _Lake AUG 27 17 10 58.82 3 _Lake SEP 28 41 13 31.71 3 _Lake OCT 31 67 36 53.73 3 _Lake NOV 44 28 16 57.14 3 _Lake DEC 41 69 28 40.58 Toshat JAN 68 29 39 134.48 Toshat FEB 73 45 28 62.22 Toshat MAR 84 40 44 110.00 Toshat APR 69 61 8 13.11 Toshat MAY 48 61 13 21.31 Toshat JUN 24 44 20 45.45 Toshat JUL 25 11 14 127.27 Toshat AUG 33 32 1 3.13 Toshat SEP 35 90 55 61.11 Toshat OCT 38 29 9 31.03 Toshat NOV 55 45 10 22.22 Toshat DEC 51 40 11 27.50 Aucilla JAN 54 52 2 3.85 Aucilla FEB 59 29 30 103.45 Aucilla MAR 68 73 5 6.85 Aucilla APR 5 5 40 15 37.50 Aucilla MAY 39 4 35 875.00
109 Table A 5. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Aucilla AUG 27 25 2 8.00 Aucilla SEP 28 35 7 20.00 Aucilla OCT 31 35 4 11.43 Aucilla NOV 4 4 30 14 46.67 Aucilla DEC 41 22 19 86.36 Mean 17.2 63.9 Median 14.0 43.7 Table A 6. Prediction errors for n ational forests FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Juniper JAN 183 381 198 51.97 Juniper FEB 199 280 81 28.93 Juniper MAR 228 83 145 174.70 Juniper APR 186 88 98 111.36 Juniper MAY 131 67 64 95.52 Juniper JUN 65 39 26 66.67 Juniper JUL 67 22 45 204.55 Juniper AUG 90 148 58 39.19 Juniper SEP 94 56 38 67.86 Juniper OCT 104 75 29 38.67 Juniper NOV 149 205 56 27.32 Juniper DEC 138 280 142 50.71 SR_19 JAN 167 214 47 21.96 SR_19 FEB 182 199 17 8.54 SR_19 MAR 209 136 73 53.68 SR_19 APR 170 66 104 157.58 SR_19 MAY 120 98 22 22.45 SR_19 JUN 60 91 31 34.07 SR_19 JUL 61 91 30 32.97 SR_19 AUG 82 78 4 5.13 SR_19 SEP 86 81 5 6.17 SR_19 OCT 95 75 20 26.67 SR_19 NOV 136 110 26 23.64 SR_19 DEC 126 144 18 12.50 Lake_D JAN 23 33 10 30.30 Lake_D FEB 25 62 37 59.68 Lake_D MAR 29 18 11 61.11 Lake_D APR 24 29 5 17.24 Lake_D MA Y 17 23 6 26.09
110 Table A 6. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Lake_D JUN 8 4 4 100.00 Lake_D JUL 9 2 7 350.00 Lake_D AUG 11 5 6 120.00 Lake_D SEP 12 4 8 200.00 Lake_D OCT 13 6 7 116.67 Lake_D NOV 19 26 7 26.92 Lake_D DEC 18 14 4 28.57 Clear JAN 105 21 84 400.00 Clear FEB 114 135 21 15.56 Clear MAR 131 164 33 20.12 Clear APR 107 121 14 11.57 Clear MAY 75 56 19 33.93 Clear JUN 38 35 3 8.57 Clear JUL 39 91 52 57.14 Clear AU G 52 78 26 33.33 Clear SEP 54 81 27 33.33 Clear OCT 60 75 15 20.00 Clear NOV 85 17 68 400.00 Clear DEC 79 17 62 364.71 Sopchop JAN 30 33 3 9.09 Sopchop FEB 33 35 2 5.71 Sopchop MAR 38 76 38 50.00 Sopchop APR 31 16 15 93.75 Sopchop MAY 22 11 11 100.00 Sopchop JUN 11 7 4 57.14 Sopchop JUL 11 14 3 21.43 Sopchop AUG 15 23 8 34.78 Sopchop SEP 15 7 8 114.29 Sopchop OCT 17 18 1 5.56 Sopchop NOV 25 24 1 4.17 Sopchop DEC 23 19 4 21.05 Camel JAN 14 10 4 40.00 Camel FEB 16 25 9 36.00 Camel MAR 18 28 10 35.71 Camel APR 15 37 22 59.46 Camel MAY 10 9 1 11.11 Camel JUN 6 1 5 500.00 Camel JUL 5 5 0 0.00
111 Table A 6. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Camel AUG 7 2 5 250.00 Camel SEP 7 3 4 133.33 Camel OCT 8 14 6 42.86 Camel NOV 12 6 6 100.00 Camel DEC 11 7 4 57.14 Battle JAN 26 31 5 16.13 Battle FEB 28 17 11 64.71 Battle MAR 32 32 0 0.00 Battle APR 26 16 10 62.50 Battle MAY 18 11 7 63.64 Battle JUN 9 2 7 350.00 Battle J UL 9 8 1 12.50 Battle AUG 13 6 7 116.67 Battle SEP 13 41 28 68.29 Battle OCT 15 6 9 150.00 Battle NOV 21 30 9 30.00 Battle DEC 19 25 6 24.00 Turkey JAN 32 51 19 37.25 Turkey FEB 35 37 2 5.41 Turkey MAR 40 45 5 11.11 Turkey APR 32 40 8 20.00 Turke y MAY 23 22 1 4.55 Turkey JUN 11 13 2 15.38 Turkey JUL 12 6 6 100.00 Turkey AUG 16 20 4 20.00 Turkey SEP 16 2 14 700.00 Turkey OCT 18 27 9 33.33 Turkey NOV 26 24 2 8.33 Turkey DEC 24 23 1 4.35 Mean 23.5 79.1 Median 9.0 35.2 Table A 7. Pr ediction errors for The Department of the Interior FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Big_N JAN 181 288 107 37.15 Big_N FEB 196 279 83 29.75 Big_N MAR 226 266 40 15.04 Big_N APR 183 118 65 55. 08 Big_N MAY 129 114 15 13.16
112 Table A 7. Continued. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Big_N JUN 65 47 18 38.30 Big_N JUL 66 63 3 4.76 Big_N AUG 89 82 7 8.54 Big_N SEP 93 77 16 20.78 Big_N OCT 103 79 24 30.38 Big_N NOV 147 219 72 32.88 Big_N DEC 136 78 58 74.36 Big_S JAN 48 54 6 11.11 Big_S FEB 52 69 17 24.64 Big_S MAR 60 104 44 42.31 Big_S APR 48 162 114 70.37 Big_S MAY 34 74 40 54.05 Big_S JUN 17 12 5 41.67 Big_S JUL 18 7 11 157. 14 Big_S AUG 23 5 18 360.00 Big_S SEP 24 7 17 242.86 Big_S OCT 27 4 23 575.00 Big_S NOV 39 2 37 1850.00 Big_S DEC 36 16 20 125.00 StMarks JAN 100 105 5 4.76 StMarks FEB 109 77 32 41.56 StMarks MAR 125 146 21 14.38 StMarks APR 102 90 12 13.33 StMa rks MAY 72 45 27 60.00 StMarks JUN 36 26 10 38.46 StMarks JUL 37 29 8 27.59 StMarks AUG 49 10 39 390.00 StMarks SEP 51 75 24 32.00 StMarks OCT 57 114 57 50.00 StMarks NOV 81 99 18 18.18 StMarks DEC 75 83 8 9.64 Mean 31.1 128.2 Median 20.5 37.7 Table A 8. Prediction errors for US Air Force. FNST Access Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Buck_P JAN 7 2 5 250.00 Buck_P FEB 8 5 3 60.00 Buck_P MAR 9 4 5 125.00
113 Table A 8. Continued. FNST Acce ss Point Month # of Hikers Predicted Actual # of Hikers Absolute Error Percent Error Buck_P APR 7 12 5 41.67 Buck_P MAY 5 13 8 61.54 Buck_P JUN 3 4 1 25.00 Buck_P JUL 3 2 1 50.00 Buck_P AUG 4 1 3 300.00 Buck_P SEP 4 10 6 60.00 Buck_P OCT 4 2 2 100.00 Buck_P NOV 6 4 2 50.00 Buck_P DEC 5 3 2 66.67 Alaqua JAN 4 2 2 100.00 Alaqua FEB 4 1 3 300.00 Alaqua MAR 4 4 0 0.00 Alaqua APR 4 3 1 33.33 Alaqua MAY 3 7 4 57.14 Alaqua JUN 1 2 1 50.00 Alaqua JUL 2 1 1 100.00 Alaqua AUG 2 2 0 0.00 Alaqua SEP 2 1 1 100.00 Alaqua OCT 2 1 1 100.00 Alaqua NOV 3 4 1 25.00 Alaqua DEC 3 3 0 0.00 SR_85 JAN 14 11 3 27.27 SR_86 FEB 15 20 5 25.00 SR_87 MAR 18 17 1 5.88 SR_88 APR 14 12 2 16.67 SR_89 MAY 10 4 6 150.00 SR_90 JUN 5 6 1 16.67 SR_91 JUL 5 3 2 66.67 S R_92 AUG 7 4 3 75.00 SR_93 SEP 7 8 1 12.50 SR_94 OCT 8 17 9 52.94 SR_95 NOV 11 13 2 15.38 SR_96 DEC 11 10 1 10.00 Mean 2.6 70.3 Median 2.0 51.5
114 LIST OF REFERENCES Agresti, A., & Finlay, B. (1997). Statistical Methods for the Social Sciences Upper S addle River, NJ: Prentice Hall. Albritton, R., Denny, C., & Stein. V. (2006). Florida National Scenic Trail visitor assessment. Presented to: National Forests in Florida, U.S.D.A. Forest Service, Florida Trail Association. Albritton, R., Denny, C. & Stein. V. (2007). Florida National Scenic Trail visitor assessment. Presented to: National Forests in Florida, U.S.D.A. Forest Service, Florida Trail Association. Albritton, R., & Stein, T. (2007). Exploring marketing opportunities for the Florida National Scenic Trail. Presented to: U.S.D.A. National Forests in Florida, Florida Trail Association. Albritton, R. Stein, T., & Thompson, T. (2005). Florida National Scenic Trail visitor assessment. Presented to: National Forests in Florida, U.S.D.A. Forest S ervice, Florida Trail Association. Albritton, R., Wan, B., & Stein, T. (2008). Florida National Scenic Trail visitor assessment. Presented to: National Forests in Florida, U.S.D.A. Forest Service, Florida Trail Association. Ball K., Bauman A., Leslie E., & Owen, N. (2001). Perceived environmental aesthetics and convenience, and company are associated with walking for exercise among Australian adults. Preventive Medicine, 33 434 440. Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of being. American Psychologist. 54 (7), 462 479. Behnam, J., & Patel, B. G. (1977). A method for estimating pedestrian volume in a central business district. Transportation Research Record, 629, 22 26. Booth, K. L., Driver, B. L., Espiner, S. R., & Kappelle, R. J. (2002). Managing public conservation lands by the Beneficial Outcomes Approach with emphasis on social outcomes. Science Internal Series, 52. New Zealand Department of Conservation. Bourdeaudhuij, I. D., Sallis, J. F., & Saelens, B. E. (2003). Environm ental correlates of physical activity in a sample of Belgian adults. American Journal of Health Promotion. 18(1), 83 92 Bowker, J. M., English, D. B. K., Cordell, H. K. (1999). Projections of outdoor recreation participation to 2050. In H. K. Cordell, C. Betz, J. M. Bowker, and others, Outdoor recreation in American life: A national assessment of demand and supply trend (pp. 323351). Champaign, IL: Sagamore Publishing.
115 Brainard, J., Bateman, I., & Lovett, A. (2001). Modeling demand for recreation in Engli sh woodlands. Forestry, 74 (5), 423 438. Bronfenbrenner, U. (1986). Recent advances in research on the ecology of human development. In R. K. Silbereisen, K. Eyferth, & G. Rudinger (Eds.), Development as action in context: Problem behavior and normal youth development (pp. 287309), New York: Springer. Budruk, M., Virden, R., & Waskey M.T. (2009). Community conversations around public lands: Exploring beneficial outcomes in the Wood River Valley. Journal of Park and Recreation Administration, 27(1), 102 118. Cervero, R. (1996). Mixed landuses and commuting: Evidence from the American Housing Survey. Transportation Research A, 30, 361377. Chawla, L. (1998). Significant life experiences revisited: A review of research on sources of environmental sensitivity. Environmental Education Research. 4(4), 369382. Chen, R. J. C., Bloomfield, P., & Fu, J. S. (2003). An evaluation of alternative forecasting methods to recreation visitation. Journal of Leisure Research, 35(4), 441454. Cordell, H. K., Betz, C. J., & Gr een, G. T. (2008). Naturebased outdoor recreation trends and wilderness. International Journal of Wilderness, 14(2), 7 13. Darlington, R. B. (1990). Regression and linear models. USA: McGraw Hill, Inc. DeMaris, A. (2004). Regression with social data: Modeling continuous and limited response variables. Hoboken, NJ: John Wiley & Sons, Inc. Driver, B. L. (Ed.) (2009). Managing to Optimize the Beneficial Outcomes of Recreation. State College, PA: Venture Publishing, Inc. Driver, B. L. (1993, November). The benefits based approach to amenity resource policy analysis and management. Proceedings, Scandinavian Conference on Recreation: Benefit and Other Positive Effects the Norwegian Institute of Natural Studies, Lillehammer, Norway. Driver, B. L., Bruns D., & Vir den, R. J. (2008). Case study evaluation of the pilot test application of Benefits Based Management on the McInnis Canyons National Conservation Area. Final report presented to: Division of Recreation and Visitor Services, BLM, Washington, D.C. Driver, B. L., & Moore, R. L. (2005). Introduction to outdoor recreation. State College, PA: Venture Publishing.
116 Driver, B. L., & Tocher, S. (1970). Toward a behavioral interpretation of recreational engagements, with implications for planning. In B. L. Driver (Ed.), Elements of outdoor recreation planning (pp. 931). Ann Arbor, MI: The University of Michigan Press. Dwyer, J. F. (1987). Predicting daily use of urban forest recreation sites. Landscape and Urban Planning, 15, 127 138. Ewing, R., & Cervero, R. (2001). Tr avel and the built environment synthesis. Transportation Research Record. 1780, 87 114. Federal Highway Administration. (1999). Guidebook on methods to estimate nonmotorized traffic: Supporting documentation (FHWA RD98166). McLean, VA: U.S. Department of Transportation. Fesenmaier, D. R., Goodchild M. F., & Lieber, S. R. (1981). The importance of urban milieu in predicting recreation participation: The case of day hiking. Leisure Sciences 4 (4), 459476. Florida Trail Association. (n.d.) Fast facts Retrieved October, 24, 2009 from Florida Trail Association Web site: http://www.floridatrail.org/ Frank, L. D., & Pivo, G. (1994). Impact of mixed use and density on utilization of three modes of travel: Single oc cupant vehicle, transit, and walking. Transportation Research Record, 1466, 44 52. Freedman, D. (1999). From association to causation: Some remarks on the history of statistics. Statistical Science, 14(3), 243258. Friend, S. (2004). The Florida Trail the official hiking guide. Englewood, CO: Westcliffe Publishers. FOSSGIS E. V. (2010). Latest updates Retrieved February 19, 2010, from the freegis.org Web site: FreeGIS.org Giles Corti, B., & Donovan, R. J. (2002). So cioeconomic status differences in recreational physical activity levels and real and perceived access to a supportive physical environment. Preventive Medicine 35 601 611. Giles Corti, B., Timperio, A., Bull, F., & Pikora, T. (2005). Understanding physic al activity environmental correlates: Increased specificity for ecological models. Exercise and Sport Sciences Reviews 33 (4), 175 181. Goldthorpe, J. H. (2001). Causation, statistics, and sociology. European Sociological Review, 17(1), 1 20. Handy, S. (1996). Urban form and pedestrian choices: Study of Austin neighborhoods. Transportation Research Record, 1552, 135 44.
117 Hilbe, J.M. (2007). Negative binomial regression. New York: Cambridge University Press. Hovell, M.F., Wahlgren, D.R., & Gehrman, C.A. (200 2). The Behavioral Ecological Model:Integrating public health and behavioral science. In R. J. DiClemente, R. A. Crosby, & M. C. Kegler (Eds.), Emerging theories in health promotion practice and research : Strategies for improving public health (pp. 347 38 5). San Francisco, CA: John Wiley & Sons, Inc. Humpel, N., Owen, N., Iverson, D., Leslie, E., & Bauman, A. (2004). Perceived environment attributes, residential location, and walking for particular purposes. American Journal of Preventive Medicine 26 11 9 125. Kirtland, K. A., Porter, D. E., Addy, C. L., Neet, M. J., Williams, J. E., Sharpe, P. A., Neff, L. J., Kimsey, D. C., & Ainsworth, B. E. (2003). Environmental measures of physical activity supports perception versus reality. American Journal of Prev entative Medicine, 26, 323 331. Kitamura, R., Mokhtarian, P. L., & Laidet, L. (1997). A microanalysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation, 24, 125158. Knopf, R. C. (1983). Recreational needs and behavior in natural settings. In I. Altman, & J. F. Wohlwill (Eds.), Behavior and the natural environment (pp. 205 240). New York: Plenum. Legendre, P. (1993). Spatial autocorrelation: Trouble or new paradigm? Ecology, 74(6), 16591673. Lindsey, G., Han, Y., Wilson J., & Yang, J. (2006). Neighborhood correlates of urban trail use. Journal of Physical Activity and Health. 3 (Suppl. 1), S139S157. Lindsey, G., Wilson, J., Jihui, A. Y., & Alexa, C. (2008). Urban greenways, trail characteristics and trail use: Im plications for design. Journal of Urban Design. 13( 1), 53 79. Lindsey, G., Wilson, J., Rubchinskaya, E., Yang J., & Han, Y. (2007). Estimating urban trail traffic: Methods for existing and proposed trails. Landscape and Urban Planning, 81, 189 215. Liu, G. C., Colbert, J. T., Wilson, J. S., Yamada, I., & Hoch, S. C. (2007). Examining urban environment correlates of childhood physical activity and walkability perception with GIS and remote sensing. In Geo Spatial technologies in urban environments (pp. 1211 39). Berlin Heidelberg: Springer Loomis, J. B. (2000) Counting on recreation use data: A call for long term monitoring Journal of Leisure Research, 21, 9396
118 Loomis, J. & Walsh, R. (1997). Recreation economic decisions: Comparing benefits and costs. St ate College, PA: Venture Publishing, Inc. Louv, R. (2005). Last child in the woods Saving our children from naturedeficit disorder. Oxford, Oxfordshire: Oxford University Press. Lovett, A. A., Brainard J. S., & Bateman, I. J. (1997). Improving benefit tr ansfer demand functions: A GIS approach. Journal of Environmental Management 51 373 389. Loye, D. (1980). Personality and prediction. Technological Forecasting and Social Change, 16, 93104. Mason, C. H., & Perreault, W. D. (1991). Collinearity, power, and interpretation of multiple regression analysis. Journal of Marketing Research, 28, 268 280. Owen, N., Humpel, N., Leslie, E., Bauman, A., & Sallis, J. F. (2004). Understanding environmental influences on walking. American Journal of Preventive Medici ne. 27(1), 67 76. Owen, N., Leslie E., Salmon J., & Fotheringham, M. J. (2000). Environmental determinants of physical activity and sedentary behavior. Exercise and Sport Science Reviews 28 153158. Pergams, O. R. W, & Zaradic, P. A. (2008). Evidence for a fundamental and pervasive shift away from naturebased recreation. Proceedings of the National Academy of Sciences, 105(7), 22952300. Pushkarev, B., & Zupan, J. M. (1971). Pedestrian travel demand. Highway Research Record. 355, 3753. Raymore, L. A. (2 002). Facilitators to leisure. Journal of Leisure Research. 34(1), 37 51. Reynolds, K. D., Wolch J., Byrne J., Chou C. P., Feng G., Weaver S., & Jarrett, M. (2007). Trail characteristics as correlates of urban trail use. American Journal of Health Promotion, 24 (1), 335 345. Saelens, B. E., Sallis, J. F., & Lawrence, F. D. (2003). Environmental correlates of walking and cycling: Findings from the transportation, urban design, and planning literatures. Annals of Behavioral Medicine, 25 (2), 80 9 1. Sallis, J. F., Bauman, A., & Pratt, M., (1998). Environmental and policy interventions to promote physical activity. American Journal of Preventive Medicine, 15(4), 379 397. Sallis, J. F., & Owen, N. (1997). Ecological models. In K. Glanz, F. M. Lewis, & B. K. Rimer, (Eds.), Health behavior and health education: Theory, research, and practice (pp. 403 424). San Francisco, CA: Jossey Bass Inc.
119 Sanborn, A., Belcher, A., Albritton, R., & Stein, T. (2004). Florida National S cenic Trail visitor assessment. P resented to: National Forests in Florida, U.S.D.A. Forest Service, Florida Trail Association. Stankey, G. H, Cole, D. N., Lucas, R. C., Peterson, E. M., & Frissell, S. S. (1985). The Limits of Acceptable Change (LAC) system for wilderness planning. General Technical Report INT 176. United States Department of Agriculture, Forest Service. Troped, P. J., Saunders, R. P., Pate, R. R., Reininger, B., Ureda, J. R., & Thompson, S. J. (2001). Associations between self reported and objective physical environmental factors and use of a community rail trail. Preventive Medicine, 32, 191200. Troped, P. J., Saunders, R .P., Pate R. P., Reininger, B., & Addy, C. L. (2003). Correlates of recreational and transportation physical activity among adults in a New England com munity. Preventative Medicine, 37 304310. The National Trails System Act Title 16 U.S.Code, Sections 2 1251 UCLA: Academic Technology Services, Statistical Consulting Group. (n.d.). Incidence Rate Ratio interpretation. Retrieved February 5, 2010, fro m UCLA Academic Technology Services Web site: http://www.ats.ucla.edu/stat/Stata/output/stata_nbreg_output.htm U nive rsity of Florida, School of Forest Resources and Conservation. (n.d.). Scrub Retrieved February 25, 2010, from the University of Florida, School of Forest Resources and Conservation Web site: http://sfrc.ufl.edu/4h/Ecosystems/Scrub/scrub.html U.S. Department of Agriculture, Forest Service. (n.d.) Appalachian Trail visitor use study Retrieved January 15, 2010, from the USDA Southern Research Station Web site: http://www.srs.fs.usda.g ov/trends/AT.html U.S. Department of Transportation, Bureau of Transportation Statistics. (2000). Bicycle and pedestrian data: Sources, needs, and gaps. Report No. BTS00 02. Washington, DC: U.S. Department of Transportation, Bureau of Transportation Stati stics. Wan, B., Cucinella, J., & Stein, T. (2009). Florida National Scenic Trail visitor assessment. Unpublished report to be presented to: National Forests in Florida, U.S.D.A. Forest Service, Florida Trail Association. Wiggins, J. S. (1973). Personality and prediction: Principles of personality assessment. Reading, MA: Addison Wesley Pub. Co.
120 Wilson, J., Lindsey, G., & Liu, G. (2008). Viewshed characteristics of urban pedestrian trails, Indianapolis, Indiana, USA. Journal of Maps, 2008 108 118. Woodsid e, A. G., Caldwell, M., & Spurr, R. (2006). Advancing Ecological Systems Theory in lifestyle, leisure, and travel research. Journal of Travel Research, 44, 259272.
121 BIOGRAPHICAL SKETCH Joshua Cucinella was born in St. Augustine, Florida. He attended the University of Central Florida and graduated with a Bachelor of Science in Liberal Studies (Environmental Studies Track) in 2007. He then went on to the University of Florida where he received a Master of Science degree in Forest Resources and Conservation in 2010. He enjoys hiking, canoeing, and brewing beer.