WALKING AND HEALTHY ? ON THE RELATIONSHIP AMONG UTILITARIAN WALKING, HEALTH, AND RESIDENTIAL CHOICE By MIGUEL A LUGO ORTIZ A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 5
Â© 2015 Miguel Angel Lugo Ortiz
3 ACKNOWLEDGMENTS I thank my dissertation committee , Dr. Siva, Dr. Steiner, Dr. Washburn, Dr. Lily, and Dr. Yi n, for their attentive supervision and mentoring in this endeavor . I extend this appreciation to my work colleagues at Weil Hall , past and present, for their support, and my f amily and friends in Gainesville , in Puerto Rico, and abroad for their encouragement through this journey.
4 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 3 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 ABSTRACT ................................ ................................ ................................ ..................... 8 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ ...... 9 2 LITERATURE REVIEW ................................ ................................ .......................... 12 Introduction ................................ ................................ ................................ ............. 12 The Physical Activity, Built Environment, and Health Framework ........................... 12 Active Transportation and Health ................................ ................................ ............ 14 Built Environment, Physical Activity, and Active Transportation ............................. 16 Built Envi ronment and Health ................................ ................................ ................. 20 Data Fusion ................................ ................................ ................................ ............ 23 Summary ................................ ................................ ................................ ................ 25 3 MAKING THE CONNECTION BETWEEN UTILITARIAN WALKING AND HEALTH MEASURES USING THE AM ERICAN TIME USE SURVEYS ................ 2 8 Data ................................ ................................ ................................ ........................ 28 Effect of Walking on Health ................................ ................................ ..................... 32 Effect of Health on Walking ................................ ................................ ..................... 34 Findings ................................ ................................ ................................ .................. 35 4 IMPUTING HEALTH ME ASURES BY DATA FUSION ................................ ........... 44 Introduction ................................ ................................ ................................ ............. 44 Application of the T wo M ethods ................................ ................................ .............. 45 Validation ................................ ................................ ................................ ................ 48 5 USE OF THE FLORI DA NHTS TO LINK LAND USE AND HEALTH ...................... 57 Data A ssembly ................................ ................................ ................................ ........ 57 Sample ................................ ................................ ................................ .................... 58 Data ................................ ................................ ................................ ........................ 58 Land U se D escriptors ................................ ................................ ............................. 60 Model ................................ ................................ ................................ ...................... 60 Findings ................................ ................................ ................................ .................. 61
5 6 CONCLUSION: WHERE DO WE GO FROM HERE? ................................ ............. 68 LIST OF REFERENCES ................................ ................................ ............................... 71 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 78
6 LIST OF TABLES Table page 3 1 Summary c haracteristics of the a nalysis s ample. ................................ ............... 37 3 2 Purpose of weekday and weekend walk trips ................................ ................... 38 3 3 Summary Statistics on health measures ................................ ............................ 38 3 4 Distribution of walking behavior by health measures ................................ .......... 39 3 5 Distribution of health measures by walking behavior ................................ .......... 39 3 6 Effect of walking (10 minute threshold) on health ................................ ............... 40 3 7 Effect o f walking (15 minute threshold) on health ................................ ............... 41 3 8 Effect of health on walking (10 minute threshold) ................................ ............... 42 3 9 Effect of health on walking (15 minute threshold) ................................ ............... 43 4 1 Descriptive statistics of ATUS donor and receiver databases. ........................... 50 4 2 Deterministic model Dataset B employed in validation. ................................ ...... 51 4 3 Blocking variables and method. ................................ ................................ .......... 52 4 4 Matching variables and methods ................................ ................................ ........ 52 4 5 Errors across models. ................................ ................................ ......................... 56 5 1 Comparison of Florida NHTS sample and the ATUS sample. ............................ 63 5 2 BMI and walking participation in Florida sample ................................ ................. 64 5 3 Land use descriptors ................................ ................................ .......................... 65 5 4 Predicted w eekday 10 min walk ................................ ................................ ......... 66 5 5 Predicted w eekday 15 min walk ................................ ................................ ......... 66 5 6 Effect of land use and predicted walking on health BMI ................................ ..... 67
7 LIST OF FIGURES Figure page 2 1 Utilitarian walking, health and built environment relationships within the context of this study ................................ ................................ ............................ 27 4 1 Matched BMI vs. reported BMI for five runs ................................ ....................... 53 4 2 Mean matched BMI (dark) and predicted BMI (light) vs. reported values ........... 54 4 3 Frequency distribution of BMI ................................ ................................ ............. 55 4 4 Error distributions ................................ ................................ ............................... 56 5 1 Histogram of BMIs matched to Florida sample. ................................ .................. 64
8 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy WALKING AND HEALTHY ? ON THE RELATIONSHIP AMONG UTILITARIAN WALKING, HE ALTH, AND RESIDENTIAL CHOICE By Miguel A. Lugo Ortiz August 2015 Chair: Sivaramakrishnan Srinivasan Major: Civil Engineering Walking as a mode of transportation , or utilitarian walking, is available to 92.7% of the adults in the United States and as su ch may be identified as one of the most available modes. While the qualitative impact of walking on health can be inferred from the energy balance model, empirical studies are necessary t o ascertain the magnitude of this impact. Further, such empirical mod els should capture the effects of land use patterns. This study address ed this gap in the literature by using data fusion techniques to merge transportation and health surveys and subsequently develop comprehensive models of land use, walking, and health. T wo recent large cross sectional surveys from the U . S . , the American Time Use Survey and the National Household Travel Survey , are employed to explore these relations by following public health policy recommend ations of continuous walking thresholds of at least 10 minutes . The national models showed a negative impact of poor health on utilitarian walking . As an application, a full set of models that links land use, walking, and health was demonstrated for Florida adults under the age of 70 . The applied mode l associated the presence of pedestrian oriented facilities up to a half mile from household with lower BMIs after controlling for socioeconomic characteristics.
9 CHAPTER 1 INTRODUCTION Walking as a mode of transportation is available (defined by physical capability) to 92.7% of the adults (1 ) in the United States and as such may be identified as one of (2 ) also indicates that destinations are often not accessible by walking , limiting the use of walking as a means of transportation. In this context, walking is estimated to account for 10.4% of all trips in the United States (3 ) and at least 2.8% of all work trips (4 ). As m any as 69% of short trips are undertaken by private motorized vehicles (5 ) , indicating a potential for increased shifts towards non motorized modes. Clearly, there are also several reasons to incentivize walking and these include environmental needs (6, 7 ) , alleviation of urban congestion (8 ), and economic impact (5 ). More recently, there has also been an increasing interest in promoting walking from the public health standpoint. Walking can be a catalyst for physical activity (9 ) and thereby discourag e sed entary lifestyles and associated ills. Sedentary lifestyles are characterized as unhealthy as they lead to higher risks of obesity, disease, health conditions and shorter lives (10, 11 ). Walking for transportation, as a moderate pace activity, can be a str ategy used to facilitate daily physical activity, required for a healthy (active) lifestyle in adults. (12 ) which captures the physiological relationship between walking and health measures such as BMI : w alki ng results in a greater expenditure of energy thereby limiting the accumulation of additional adipose tissue (body fat) . Current U . S . Department of Health and Human Services/Centers for Disease Control (13 ) recommendations include at least 10 minutes at a time of moderate intensity aerobic
10 ( e.g., brisk walking) for a total of 150 minutes every week and two days of muscle strengthening activities. In contrast, the U . S . Surgeon General recommends 30 minutes of moderate physical activities on most days of the week (14 ). While the qualitative impact of walking on health can be inferred from the energy balance model, empirical studies are necessary to ascertain the magnitude of the impact of walking on health. S uch empirical models should capture the effects of l and use patterns so that these models may be used to determine if and how public health can be improved via changes in land use patterns. While studies on built environment, walking and physical health exists in the fields of public health, urban planning, and transportation, comprehen sive models addressing all their aspects are limited. Arguably one of the major impediments to the development of such models is the lack of data. Transportation surveys while providing details travel partners and spatial info rmation at a fine resolution do not include health data. Health surveys often do no collect travel behavior data and/or do not provide detailed spatial resolution to construct land use descriptors. This study addresse d this gap in the literature by using data fusion techniques to first merge transportation and health surveys . The objectives of this research were to develop a methodology for comprehensive disaggregate models of land use , utilitarian walking, and health of U.S. adults . This study employ ed two recent large cross sectional surveys from the U . S . : the American Time Use Survey (ATUS) and the National Household Travel Survey (NHTS ) . Once the health data from the ATUS has been suitably linked to NHTS respondents, the final component was to develop a full set of models that link ed land use, walking, and
1 1 health. Such models can be useful in evaluating the impacts of land use changes on walking patterns, and ultimately, on public health. The survey samples are not limited to any one geographic area or to a specific population segment. At the same time individual level time use and health data are modeled. The rest of this dissertation is organized as follows. Chapter 2 presents a summary of relevant literature. Chapter 3 presents models relating walking to health using the American Time Use Surveys. Chapter 4 presents a data fusi on approach to link health and travel data . Chapter 5 provides an application of data fusion techniques in a Fl o rida sample, along with built environment descriptors . Finally, chapter 6 offers a conclusion and recommendations for further study.
12 CHAPTER 2 LITERATURE REVIEW Introduction The purpose of this chapter is to first present a synthesis of empirical findings on the inter dependencies among physical activity, built environment, and health. Subsequently, literature on data fusion is presented as such methods are required to link health and travel surveys to assemble a comprehensive dataset for modeling the relationships. The Physical Activity, Built Environment, and Health Framework Physical activity (15 ). Physical activ ity is often characterized by context in which occurs: occupational/mandatory (e.g. , work or school), household or maintenance (e.g. , cooking and cleaning), leisure time, and transportation. Leisure time physical activity (LPTA) may include recreational ac tivities (e.g. , walking for recreation), sports, and exercise ( planned physical activity with the objective of improving or maintaining physical fitness). Active transportation can be defined as physical activity for transportation; across fields, it can a lso be found as (16 ). The substantive empirical focus of this study is on utilitarian walking, which is one of the modes of active transp ort, which in turn is a component of physical activity (Figure 2 1). Defining built environment is more permissive: e.g., from the field of anthropology, Lawrence (17 ) environment, from heaths to citie s, through construction by humans . . . spaces that are
13 ( 18 ) and last, ecological models add that there can be both an objective built environment as well as a perceived built environment (1 9 ). The empirical focus of this study is on objective measures of land use and the built environment that may be constructed u sing geographic databases of land use and transportation systems (Figure 2 1). Last, health as defined by the President's Council on Physical Fitness and Sports ( 20 ), includes a positive component (wellness) that is associated with a quality of life and positive well (15 ) quotes Buchard et al. by defining health on page 22 , a human condition with physical, social, and psychological dimensions, each characterized on a continuum with positive and negative poles. Positive health is associated with a capacity to enjoy life and to withstand challenges; it is not merely the absence of disease. Negative health is associated with morbidity and in the e xtreme with premature mortality . of this term, applies to studies that relate physical activity with the prevention of disease ( i.e., obesity, overall mortality, cardiovascular diseases, coronary heart disease, stroke, high blood pressure, cancer, osteoarthritis, osteoporosis, bone fract ures, and mental health) or studies that measure physical fitness (e.g. , submaximal aerobic capacity) and quality of life (e.g. , physiological well being). In this study (Figure 2 1) , we focus on measures of subjective wellbeing as well as objective measur es of health (Body Mass Index or BMI ) . T he next several sections present a summary of literature linking each pair of the three items (physical activity, health, and land use) identified in the framework. To limit
14 the scope of the discussions, it focus ses on studies that have addressed the specific measures of interest (such as utilitarian walking in the case of physical activity and BMI and subjective well being in the case of health). However, as the studies are often not consistent in their choice of me asures, additional studies if deemed relevant are included . Finally, it is also useful to note that there are several synthesis papers on the topics of interest. Therefore, the review first draw from these synthesis studies before adding results from recen t individual empirical studies. Active Transportation and Health As a desirable public health policy, the study of active transportation and health often occurs from the epidemiology/ecological and transportation/planning fields. In both fields, research stems from two general methodologies: the objective measurement of levels of active transportation and the incidental effect of increase in walking or biking rates. Frank et al. (2 1 ) looked at the Atlanta metropolitan region and provide the latest review o f this relationship. The framework employed in this study follows a linear sequence with lifestyles influencing residential location (land use), travel behavior and subsequently health. The study found users that preferred walking and lived on walkable nei ghborhoods were less likely to drive and more likely to walk. However walking or walking preference was not significant factor for obesity. Further, their study on 2,056 individuals found a significant fraction of the population as living outside of their preferred neighborhood type. Within the discipline of transportation engineering, most of the literature has focused on walkability and work trips, often treating walking and biking as one category and generally finding limited evidence of lower body weigh ts for adults [ see systematic review (2 2 ) ] . The empirical research focused on commute has found significant
15 relationships at aggregate scales. For instance, higher state level and city level walk and bik e commut ing shares ha ve been associated with lower di abetes and obesity rates (2 3 ). Higher country level share of biking and walking (all purpose trips) has also been correlated with lower obesity rates (2 3 ). In Belgium, a seven day cross sectional study found a negative association between BMI and minutes of walking for transportation (2 4 ). This relationship can also be positive, as one survey in Australia found women that participate d in active transportatio n and liv ed in socioeconomically disadvantaged neighborhoods as more likely to be obese (2 5 ) . Similarly, another survey among 30 low income mothers in England, found those who walked more than their peers had lower Self Assessed Physical Health Score ( SAPH S ) value s (2 6 ). Clinical research, mainly focused on European cities, has found a positive relationship between active transportation (walking or biking) and improved fitness, and reduction in all cause mortality, type 2 diabetes, and hypertension risks [ s ee (2 7 ) for systematic review ] . However, the definitions of meaningful walking vary across such studies with values starting in 10 min/day up to 30 min/day or 120 min/week. Some studies also employ Metabolic Equivalent Task values (MET) instead of actual w alk durations. A smaller number of projects has focused on walking for commute : associating commute times and frequency with reduced health risks and reduced BMI in Denmark and Osaka (2 8 , 2 9 ). Commuting to work, however, only accounts for approximately 16% of all person trips. The only known prospective cohort study of walking for transportation in general, could not find an association between walking >90 min/wk and reduced all cause mortality risk ( 30 ).
16 In addition to understanding the effect of walking o n health, it is also important to understand the effect of health on walking. Research on the effect of health on utilitarian walking has been scarce but greater emphasis has been placed on the impact of health on overall Leisure Time Physical Activities ( LTPA). A synthesis reviews on physical activity (which may include, among other activities, leisure walking but not utilitarian) by Sallis and Owen (3 1 ) found lack of association between overweight individuals and a reduction of activities. However, an upd ated review by Trost et al. (3 2 ) supports a negative association. While the transportation literature is abundant with studies on as an explanatory factor in these stu dies. Built Environment , Physical Activity , and Active Transportation Though many agree on or at least have studied the relationship between built environment and physical activity in adults , a review of current literature shows diverse outcome variables and an apparent disparity on its measurement . The latest study by Harris et al . (33) , aimed at understanding the current structure of the physical activity and built environment field, maintains the association between active lifestyles and built environme nt, both for objective and perceived environment characteristics. In their paper, the authors set the stage by mapping relations and gaps and classifying studies as reviews, delive ry, policy, theory/methodology , or discovery. A common issue within the lite rature and to those studying the effect of transportation systems , t he classification scheme makes no distinction between recreational or leisure time physical activities, work related, and transportation related, drawing their recommendations from the bulk number of studies focused on the first.
17 A nother review of reviews on built environment, physical activity and obesity by Ding and Gebel (3 4 ) also notes: almost half of the reviews either combined adults with youth or did not specify target age groups reviews failing to select conceptually matched associations , or to examine behavior specific correlates review ers/studies should provide more information on environmental and physical activity measures, and stratify the summary of results based on measurement modes. Besides the need for more rigorous studies, a review by Ferdinand et al. (3 5 ) identifies considerable gaps : minority populations and with their use of parks, school playgrounds, and active transportation rural populations facilitating PA for the elderly . Heath et al. (3 6 ) raise additional issues on measurements, among these: street ) neighborhood characteristics and does this relationship vary by perceived preference? In terms of an organizational structure, how should the built environment be conceptualized and what is the best way to measure or quantify components of the built environment (e.g., accessibility, aesthetics, safety, walkability)? What is the geographic scale(s) at which the neighborhood environment is most strongly correlated with physical activity? Does multivariate adjustment for potential confounding factors (e. g., age, income, gender) change the relationship between the built environment, policies, and physical activity? If so, what potential confounders are most important? Is it possible to use existing data to assess the impact of selection bias (e.g., stratif ying data sets by income group)?
18 F urthermore, Ferdinand et al. (3 5 ) found that studies whose main outcome variable was active transport were less likely to obs erve a beneficial relationship. Surprisingly, the use of objectively measured PA data ( as opposed to self reported data ) was associated with a reduced likelihood in finding a beneficial relationship between the built environment and PA or obesity rates. A systematic review of physical activity and built environment as measured by smart growth principles (3 7 ) and by the Guide to Community Preventive Services (3 6 ) also found significant associations for walking or physical activity. The first review focused on 37 cross sectional and longitudinal studies of adults and children in the U . S . (25), Au stralia (9) , Europe (5) , and Canada (3). The authors found a relationship between physical activity and principle seven (open space preservation), and walking (studies could include recreational, transport, or total) and principles one (range of housing choices), six (mixed land use), nine (development toward existing communities) and ten (compact building design). The second review grouped policies in three strategies (community scale urban design and land use policies and practices to increase p hysical activity; street scale urban design and land use policies to increase physical activity; and transportation and travel policies and practices), finding strong evidence in all but transport policies , as the authors could only found one time series s tudy focused on school mode share. A recent review specific to green space and physical activity by Lachowycz and Jones (3 8 ) found mixed evidence for an association between access to green space and physical activity. A majority of studies (66%) found some evidence of a positive association, although only 40% found an association that appeared unambiguous.
19 A 2010 review by McCormack and Shiell (3 9 ) found preliminary results towards causality between built environment and physical activity for some of the effects reviewed above. The discussion thus far has focused on the relationship between physical activities in general and built environment. Studies that narrow focus on th e relationship between built environment and active transportation are mainly drawn from the transportation/planning and the public health fields. Pulleyblank Patrick et al . ( 40 ) provide insight on how both fields have approached this . The report, commissi oned by USBGBC and partially funded by U.S. E nvironmental Protection Agency (E PA ) and the Center s for Disease Control and Prevention ( CDC ) , generally conclude s a positive association between active transportation and the increase or presence of the followi ng features: regional accessibility, population and employment density, land use mix, access to transit, neighborhood streetscape design that do not inhibit individuals from walking/cycling, higher street connectivity, and on street parking . However, there is no consensus on what degree or threshold values these characteristics are effective , with some expectation of a collective rather than isolated effect . Additional features, such as access to recreational facilities, appear to influence recreational tri ps only and not utilitarian trips . In t he most recent review on walking and built environment, Saelens and Handy (4 1 ) find that built environment association found on general physical activity carries to specifics of the association are less The studies reviewed, almost all cross sectional, were conducted in a small number of cities or neighborhoods , where the issue of self selection remains .
20 While the issue of self selection remains , Handy, Cao, and Mokhtarian (4 2 ) found built environment influence in eight neighborhoods in Northern California even aft er accounting for self selection. The study found walking accessibility, physical activity options, safety concerns, socializing opportunities, and neighborhood attractiveness as characteristics that influenced walking trip behavior for 688 neighborhood movers. A nother self selection study , led by Frank et al. (4 3 ) in 2002 on the Atlanta, Georgia region , found greater walkability associated with great er share of people walking for non discretionary purposes (work, school and shopping) and discretionary trips (e.g. , entertainment, exercise), but that the absolute amount of walking remained extremely low for those who do not prefer a walkable environment . Walkability was defined as an index composed of commercial floor area ratio to the total land area devoted to commercial ; l and use mix of residential, commercial, and office use ; number of residential units per residential acre ; and the number of interse ctions with more than three legs. The most recent i ndividual study by Knuiman (4 4 ) concludes that neighborhood connectivity, land use mix, number of public transit stops, and the diversity of destination types accessible by walking were significantly related to neighborhood transport walking for the city of Perth, Australia. The 1,703 adults long itudinal study used weekly frequency of 15 min . utilitarian walk. Built Environment and Health This section will focus on studies that link built environment characteristics and more direct measures of individual health, mainly physical fitness outcomes. T he latest state of research regarding the built environment health relationship can be found in two epidemiologic synthes e s (4 5 , 4 6 ).
21 The first epidemiologic review , by Feng et al. (4 6 ) , looked at a total of 67 studies (61 cross sectional), the majority coming from the United States (52 papers) and focused on adults (45 studies). In total, 48 studies found statistically significant associations (expected or unexpected direction). The ir methodology considered studies related to one of three bu ilt environment subdomains: land use/transportation environment, food environment, and physical activity environment . The most common built environment characteristics among studies were density, diversity, design, connectivity, spatial access to facilitie s, walkability, and sprawl. Among outcome measures, a categorical measure ( i.e., normal BMI vs . obese) was predominant, followed by continuous BMI values or Z score. The authors took a closer look at 22 contextual based and 15 individually defined geograph ic buffer papers and conclude d, S tudies focused on walkability and health outcomes f ound relations between walkability (be it defined as land use, as an index, or distance to amenities) and obesity in older Americans, by ethnicity, or gender at a neighborhood scale (2 1 , 4 7 50 ) and county scale (5 1 ). Meanwhile, a comparison of walkability on Black and White neighborhoods by Doyle et al. ( 5 2 ) finds that BMI, socioeconomic status, and safety perception were stronger indicators than land use measures. The second , independent epidemiology review by Mackenbach et al. (5 3 ), focuses in adults, and updates the literature to 2013, while assessing the methodological quality of each paper. The authors reviewed the findings of a total of 92 papers, 62 coming from the US, exploring health through use of BMI, overweight or
22 obesity o utcomes. Their finding remai ns consistent w discussion: strong associations could only be found for urban sprawl and land use mix in the US. At the same time, Gebel reveals a mismatch between perceived walkability and health benefits (5 4 ). A revi ew by Ferdinand et al (3 5 ) found t he use of direct measures of body weight (e.g., BMI) as a measure of physical activity and the main outcome variable in a study was associated with a reduced likelihood in finding a beneficial relationship. A similar syst ematic review of built environment smart growth principles and health , among other factors, found no strong associations between BMI and smart growth principles (3 7 ), though not all studies had matching principles. The review focused on 44 cross sectional studies of adults and children, mostly in North America. The authors suggest ed the lack of longitudinal studies as a reason; or that smart growth's impact on body mass operates through mediation mechanisms not factored in these studies . A 2001/2002 study on the Atlanta, Georgia region by Frank et al. (4 3 ) found g reater walkability (by means of a walkability index) associated with lower odds of obesity for 2066 individuals . However, the absolute amount of walking remained low for those who d id not prefer a walkable environment. Recent studies published after the latest review s are consistent with the impact of measures of BMI and the role of perceptions: Ullman et al. (5 5 ) found w omen with better neighborhood perception (safety, social) we re associated with lower BMIs and in a longitudinal study of 975 adults in Los Angeles. Similarly, Barry (5 6 ) found that perceptions and level of physical activity had stronger obesity and diabetes effects on
23 women than men in a cross sectional study of 4, 273 adults in Allegheny County, PA. Last ly , Glazier et al. (5 7 ) develop ed multiple land use indices for a Toronto survey of 9,757 residents age 30 and older, and found that residential density and walkable destinations had strong association s with over weight status and diabetes . Data Fusion While there are several datasets available to study each pair of items from the physical activity built environment health framework, these do not support the development of a comprehensive model of the entire fr amework. As an alternative to conducting a new survey to collect additional data, we examine the applicability of data fusion methods to link travel and health surveys to generate the comprehensive dataset needed for the analysis. Specifically, the propose d approach add s health data from the ATUS dataset to the NHTS surveys at the individual level. An overview of the literature on data fusion is presented in this section. Data linkage of two or more datasets can be conducted through manual inspection, dete rministic procedures, or probabilistic matching. Manual inspection methods are not feasible with large datasets. Deterministic procedures require building a statistical model correlating the outcome variable with explanatory factors observed in both datase ts and subsequently using the estimated model to make predictions of the outcome variable in the target dataset. In th e context of this study , a regression model of BMI as a function of socio economic location variables can be built using the ATUS data and in turn applied to the NHTS data. Such methods are also fairly straight forward. The rest of this section focuses on probabilistic matching, also known as record linkage within the health/safety literature and sever al other names outside it. For a list o f
24 corresponding terms within various specialties see Gu et al. (5 8 ) . The process of linking records spans from the assumption that there is a potential pair match within two sources. As Gomat am et al. (5 9 ) explain, suppose file A has n a records and file B has n b records, then the file AÃ—B contains n a Ã—n b record pairs. Each of the n b records in file B is a potential match for each of the n a records in file A . Thus there are n a Ã—n b record pairs whose match=non match status is to be determined. Record pair comparison will lead two three outcomes: (a) match, (b) possible match, and (c) non match. The degree of separation between datasets can be an indication of the level of difficulty of the linkage and amount of Type I and II errors (5 8 ) . The sta ndard algorithm requires a method that minimizes the probability of possible matches, as opposed to true matches and non matches. This is done by estimating the ratio between conditioned probabilities of observed known identifying fields given that the rec ord pair is a true match and true not match. In many instances, the expectation maximization algorithm (EM), as described by Jaro (5 9 ), is used to estimate the weights or threshold values of this association. An EM linkage requires initial values for match and unmatched probabilities thresholds, selection of blocking attributes, the threshold of the weights for possible matches, and the type of comparison function. The inclusion of a blocking attribute (e.g. , sex) will yield comparisons of pairs within each matching attribute (e.g. , for sex as blocking attribute, the comparison of female records exclusively with other female records, and male records with other male records). The decision component, often score based, relies either on freq uency based distrib ution ( 60 ) of attributes (e.g. , there are fewer individuals living in rural areas, therefore a match of urban carries les s weight than a match of rural) or uniform, average of all matching weights.
25 Common performance measures are the number of record pairs linked correctly and incorrectly, and the number of pairs unlinked correctly and incorrectly. But, as Baker (6 1 ) notes, the accuracy of the process is a function of the size of the donor dataset and, counterintuitively the use of too many variables may not lead to better linkage. The transportation literature offers examples of record linkage, such as a study by Kukasabe and Asakura (6 2) linking smart card transit data stops and personal trip surveys , police and hospital road crash records [ Amorim et al. (6 3 ), Rosman (6 4 ) ] , work fatigue and crashes [ Williamson and Boufous (6 5 ) ] , and state and federal motor carrier safety databases (6 6 ). Pawlak et al. (6 7 ) u se d separate datasets on digital lifestyle (ICT use) and physical mobility to match records as an al ternative where suitable data cannot be obtained in a single dataset . Similarly, Kressner and Garrow (6 8 ) also found third party data from targeted marketing to be interchangeable and useful to incorporating lifestyles variables to transportation survey, in particular with age, gender, income, and presence of children . Summary ve shown how the previous relationships have been explored among walking, health, and built environment, with a focus on utilitarian walking. Evidence supports t he causal link of walking and bet ter weight or fewer diseases, however, multiple studies rely on aggregate measures of total leisure time physical activity or on commute only behavior the former too broad and the later too limited to employed adults and ma ndatory activities. The built environment and walking correlates are better known. While there is an expected confounding effect between desirable built environment features, these studies often rely on samples geographically limited to a
26 neighborhood or a particular city. For built environment and health, there is again an agreement, especially when measuring county sprawl index and land use mix , of correlations with obesity (or walking), as some review construct walking behavior as an implicit health bene shown how there are few studies that attempt to incorporate the three inter dependencies, much less using a large national sample. Often, the scarcity of rich datasets will involve extens ive data fusion techniques. The use of probabilistic record linkage has been proved to be effective with datasets of the same population for various purposes, including crash records and health and Information and Communication Technologies (ICT).
27 Figure 2 1. Utilitarian walking, health and built environment relationships within the context of this study . Ph ysical activity Built environment Health Mandatory Maintenance Leisure Transportation Observed Perceived Quality of life Disease prevention Fitness Walking Biking Other
28 CHAPTER 3 MAKING THE CONNECTION BETWEEN UTILITARIAN WALKING AND HEALTH MEASURES USING THE AMERICAN TIME USE SURVEYS The object of this chapte r is to use a large cross sectional data from the U . S . to explore the relationship between healthy individuals and their propensity to walk as well perceived) through rec ommended guidelines. The analysis was focused on adults. Data This analysis uses data from the 2006, 2007, and 2008 American Time Use Surveys (ATUS) and the corresponding Eating & Health (EH) Modules. ATUS collected detailed socio economic, demographic, and one day activity travel information for a large sample of persons. The survey is limited to one randomly selected household member (individuals age 15 or older living on non institutionalized facilities) per household and the survey samples were drawn from the Current Population Surveys (CPS) of the same years. The EH Module of the ATUS, administered between 2006 08 collected additional eating, meal preparation , and health information for a random selection of household members participating in the ATUS . To our knowledge, only one other study (6 9 ) has looked at one year of ATUS/EH data to examine the association between physical activity and BMI, noting lesser BMIs with the presence of active transportation, defined as one minute or more of walk or bike mode. Overall, both time use and health data are available for about 36,000 individuals. The socio economic characteristics of the sample are presented in Table 3 1 . Since the ATUS surveys collected weekend time use data for 50% of the samples (and weekday data for the rest), the results are shown separately for the two cases (i.e., weekdays and weekend days). Both the weekday and weekend sample have similar attributes.
29 The sample is largely Caucasian and there are more females than males (55% versus 45%). The survey respondents had some secondary degree education and were significantly likely to be employed full time. The most common household comprises two adults and their children, with a combined income between $35,000 $100,000, and living in an owned ho usehold within a Metropolitan Statistical Area according to 2000 Census (have at least one urbanized area of 50,000 or more population, plus adjacent territory that has a high degree of social and economic integration with the core). Another motivation for separating out weekday and weekend samples is that the purpose of the walk trips are significantly different across these two cases ( Table 3 2). Work related trips lead the walk trip purpose on weekdays (18.4%), while social/leisure/recreational activitie s (27.2%) characterize weekend walk trips. In order to correlate health measures to walking patterns, it is first required to Among all individuals surveyed, about 12% walked at some point during the survey day ( for any amount of duration). However, many instances of such walking were for rather short durations (such as less than 10 minutes) and these may be (1) incidental trips not reflective of general behavior and/or (2) not be significant enough from the stand point of health. To be sure, as already indicated CDC recommends at least 10 minutes at a time of moderate intensity aerobic activity such as brisk walking for a total of 150 minutes every week and the U . S . Surgeon General recommends 30 minutes of moderate physical activities on most days of the week. The ATUS is a one day survey and therefore longer term (such as weekly) walking patterns of the respondents are unknown. The number of respondents in the sample who walked continuously for 30 minutes on the su rvey day were too few.
30 instances of continuous travel related walking for 10 or more minutes. With this definition, 5.7% of the persons were observed to have walked on a wee kday and 4.7% on a weekend day. As a second definition, the threshold for continuous walking was increased to 15 minutes , the average time it takes a person to walk a mile, yielding 3.4% of the sample to be walkers on weekdays and 3% to be walkers on weeke nds. As opposed to leisure walking, most utilitarian walking was undertaken on weekdays. There are two measures of health available from the well being module of the ATUS. These are the Body Mass Index (BMI) and a Self Assessed Physical Health Score (SAPHS ). BMI is calculated from the self health was exc Table 3 3 presents summary descriptive statistics on these two heath measures. Although BMI is intrinsically a continuous variable (and is treated as such in the models later on), it has been converted into a categorical variable for cross tabulation purposes. Following the National Institutes of Health classification scheme (10 ), a person is classified into one of eight categories based on their BMI values: Severely Underweight (SU, BMI <14.9), Very Underweight (VU, BMI =15 15.9), Underweight (U, BMI =16 18.4), Normal (N, BMI =18.5 24.9), Overweight (OW, BMI =25 29.9), Obese I (Ob I, BMI =30 34.9), Obese II or Very Obese (Ob II, BMI = 35 39.9), or Obese III or Severely Overweight (Ob III, BMI >40). Due to the very small shares of underweight individuals, categories SU, VU and U are combined into an aggregate category with an
31 overall share of 1.7% of the sample. Note that 37% of the sample have a normal weight and 35% are overweight (see last column of Table 3 3). The rest of the sample is obese with 3.5% of the sample falling under the Obese III category. Based on the SAPHS, 19% indi cated their physical health to be excellent and 34.4% to be very good (see last row of Table 3 3). 4.4% of the respondents indicated poor health and 12.3% indicated fair health. Table 3 3 also presents a cross tabulation of these two health measures (the t op half of the table presents the BMI distribution within each SAPHS level and the bottom half presents the SAPHS distribution within each BMI level). These cross tabulations show a significant correlation between normal BMI and high ratings on the SAPHS s core. Specifically, 54% individuals with excellent SAPHS rating are normal weight (40% for individuals with very good SAPHS rating) while 65% of the normal weight persons reported very good or excellent on the SAPHS. Next, cross tabulation of health measures (BMI and SAPHS) are presented against walking patterns (defined using 10 and 15 minute walk time thresholds). The intent of this exercise was to present levels of correlations between walking and health at an aggre gate level. Table 3 4 presents the distribution of walking patterns by health measures. In general, individuals with a better (healthier) BMI are more likely to be days) . For example 7.1% of the normal weight persons walked 10 minutes or more on weekdays compared to 2.9% of persons who were obese III. Those with a better SAPHS rating were also found to be more likely to be walkers, although this result is more pronounced in the case of the 10 minute definition of walking. Table 3 5 presents the distribution of health by walking patterns. In general, walkers are more likely (about
32 walkers (about 35%) and this holds irrespective of the def inition of walkers and for both weekdays and weekend days. However, the distributions of SAPHS ratings across the walkers and non walkers do not appear to be pronouncedly different.hese cross tabulations are consistent with the general expectation of a pos Effect of W alking on H ealth This section presents an analysis of the effects of walking on health. As already indicated , two health measures are available. BMI is modeled using a linea r regression model and SAPHS is modeled using ordered measure, two models are developed capturing the effect of weekdays a nd weekend ve been adopted (based on minimum duration thresholds of 10 and 15 minutes), all the models (for the two health measures and the two days of the week) are estimated using each of these definitions of walking leading to a total of eight models. The results for the effect of 10 minute walking on health are presented in Table 3 6 and those for the effect of 15 minute walking are presented in Table 3 7. In each of these tables, the first two major columns present the regression model results for BMI and the last two major columns present the results of the ordered response model for SAPHS. The ATUS provides information on whether the respondent walked on the survey day. It would not be entir ely appropriate to use this directly as an explanatory variable in the models for health. This is because the short term (daily) decision of walking on a are generally desc riptors of longer term conditions. Further, walk episodes are perhaps
33 not undertaken daily. Therefore, an instrumental variables approach is adopted with a Recognizing that leisure w alking/exercise could play a substitute or complementary role, MNL model was developed to determine the probability of utilitarian walking only, exercising only, and walking and exercising, as a function of the socio economic characteristics of the individual and his/her household, residential location characteristics, and other characteristics of time use on that day. These models were developed separately for weekdays and weekend days and using each of the two time the predicted probability of walking for each person in the analysis sample. The predicted probabilities of walking are used as the explanato stage instrumental variable approach is well recognized in the field of statistics as a means to address endogenous explanatory factors and has been employed in transportation research [ see ( 40) for an application of such an approach in the context of modeling interdependencies among virtual and real activity participation choices ] . Tables 3 6 and 3 7 . All effects reported are statistically significant at 95% confidence or higher. The predicted probability of weekday/weekend walking (10 minutes or more) has a negative impact on BMI and a p ositive impact on SAPHS ( Table 3 6 ). This indicates that individuals who are m ore likely to walk (on either weekday or weekend day) have O n examining the effect of predicted probability of walking 15 minutes or more, the effect on BMI remains statistically
34 significant and of larger magn itude; positive and significant on SAPHS ( Table 3 7 ). and the effect of walking on BMI is significant even after controlling for exercise . It is useful to highlight that the models capture the marginal ef fect of walking after controlling for the effect of a variety of other factors on health. Specifically, health is affected by age, gender, ethnicity, nutrition, exercise, education, and employment status of the person. Further low income persons are likely to have higher BMI and report lower SAPHS. Similarly, those in metropolitan areas are likely to have lower BMI and report higher SAPHS. Variations in health are also found across the different census regions of the country. Effect of H ealth on W alking Th is section presents an analysis of the effects of health on walking. Binary logit models are estimating considering the dichotomous outcome variable (walked or not). Separate models are estimated to capture the differential impacts of health on weekday and weekend walking. Further, separate models are also estimated to capture the effects of each of the health outcome variables (BMI and SAPHS). Finally, given that of 10 and 15 minutes), all the models (for the effect of each of the two health measures and the two days of the week) are estimated using each of these definitions of walking leading to a total of eight models. The results for the effects of health on 10 minute w alking are presented in Table 3 8 and those for the effect of health on 15 minute walking are presented in Table 3 9. In each of these tables, the first two major columns present the regression model results for BMI and the last two major columns present the results of the ordered
35 response model for SAPHS. All effects reported are statistically significant at 95% confidence or higher. The Nagelkerke R 2 , a goodness of fit value reported on multinomial logistic regression as the normalized likelihood ration, show model performance. The models indicate a negative relationship between BMI and walking for both weekdays and weekend days; i.e., persons with higher BMI are less likely to walk. This result holds irrespective of the definition of walking used. Howev er, the relationship between SAPHS and walking holds only for weekend days. Specifically, those who feel healthier (higher SAPHS) are more likely to walk on weekends but the effect on weekday walking is statistically insignificant. This difference could be because weekday walking is more likely to be for mandatory purposes whereas weekend walking is leisure oriented. Once again, the models capture the marginal effect of health on walking after controlling for the effect of several other factors. The likeli hood of walking is affected by age, gender, ethnicity, education, and employment status of the person and the magnitude of these impacts vary by day of the week. The daily time use pattern of the individual also affects (negatively) the likelihood of under taking utilitarian walking. This is reasonable as the need to invest greater amounts of time in activity participation can motivate people to choose faster modes to get to their destinations given the overall daily time budget constraint. Variations in wal king behavior are also found across the different census regions of the country and across the different seasons of the year. Findings The relationship between utilitarian walking and health was investigated using a cross sectional analysis of a large sam ple US time use and health dataset. In particular, a series of models was developed to analyze (a) the effect of walking on health, and (b)
36 the effect of health on walking. Two thresholds (10 minute bouts and 15 minute bouts) were used in defining walking patterns as part of two health metrics (reported BMI and Self Assessed Physical Health Score). The impacts of/on weekday and weekend walking were examined separately. The study indicates that individuals who are more likely to walk (on either weekday or w However, on examining the effect of predicted probability of walking 15 minutes or more, the effect is statistically insignificant for BMI but positive and significant for SAPHS. The study ind icates a negative relationship between BMI and walking for both weekdays and weekend s ; i.e., persons with higher BMI are less likely to walk. This result holds irrespective of the definition of walking used. However, those who feel healthier (higher SAPHS) are more likely to walk on weekends , but the effect on weekday walking is statistically insignificant. This difference could be because weekday walking is more likely to be for mandatory purposes , whereas weekend walking is leisure oriented. In general , the results for the positive effects of walking on health are encouraging , particularly considering the large and diverse sample and the inclusion of a large number of control variables. At the same time, the results on the negative effects of BMI on wal king highlight the need for encouraging people who are currently not overweight to walk , as the onset of obesity can have a detrimental impact on walking. Overall, this chapter demonstrate d the benefits of utilitarian walking on health ( decreas ing BMI valu es and fostering better health perception s) and highlights the need to control obesity, as an increase in this population can reduce expected walking rates.
37 Table 3 1 . Summary c haracteristics of the a nalysis s ample. Classification Description Weekday Weekend Sample Size Randomly selected member per household 17,607 17,992 Sex Male 45.3% 44.6% Female 54.7% 55.4% Age Person is under 20 7.4% 7.6% Person is between 20 and 39 30.9% 31.7% Person is between 40 and 59 37.6% 37.4% Person 60 or older 24.2% 23.3% Race Identified as Caucasian 69.8% 68.9% Identified as Black/African America 13.0% 12.8% Identified as Hispanic 12.3% 13.4% Identified as Asian 3.0% 2.8% Other ethnicities 1.9% 2.1% Exercise Some exercise/active leisure (non transportation) 1 4.3 % 13.6 % Nutrition Food stamps recipient household 6.8 % 7.0 % Education Did not finish high school 16.6% 16.9% High school/GED degree 26.5% 26.6% Completed or some advanced degree education 56.9% 56.5% Employment Person enrolled in school (part or full time) 5.5% 6.0% Person identified as unemployed 17.4% 17.5% Person identified as part time worker 11.8% 11.1% Person identified as full time worker 53.8% 54.8% Person identified as retired from work 17.1% 16.8% Household Lives alone 25.6% 25.3% Couple without children or relatives household 20.9% 20.0% Household with minor(s) and two parents 35.4% 36.0% Household with minor(s) and one adult 5.5% 5.6% Other household types: roommates, siblings, etc. 12.5% 13.2% Income Combined household income less than $35k 31.0% 31.8% Combined household income between $35k $100k 40.3% 39.9% Combined household income greater than $100k 14.7% 15.5% HH Tenure Residence owned by a household member 75.2% 74.6% Region Household located in Northeast census region 16.7% 17.2% Household located in Midwest census region 24.9% 24.7% Household located in South census region 36.5% 36.2% Household located in West census region 21.8% 21.9% Metro Household located within a Metropolitan Area 81.5% 82.0% Household located outside a Metropolitan Area 18.5% 18.0% Missing household location 0.7% 0.8% Season Survey day between Nov 01 Feb 28 33.1% 33.8% Survey day between Mar 01 April 30 17.3% 17.9% Survey day between May 01 Jun 14 12.7% 11.5% Survey day between Jun 15 Aug 14 16.5% 16.4% Survey day between Aug 15 Sep 14 7.7% 7.1% Survey day between Sep 15 Oct 31 12.7% 13.4%
38 Table 3 2 . Purpose of weekday and weekend walk trips. Walk trip purpose Weekday Walk trip purpose Weekend Work 18.4% Socializing, relaxing and leisure 27.2% Socializing, relaxing and leisure 18.2% Consumer purchases 22.5% Consumer purchases 14.5% Eating and drinking 10.9% Caring for household members 11.7% Work 6.1% Eating and drinking 10.2% Sports, exercise & recreation 5.9% School 5.7% Household activities 5.7% Household activities 4.1% Caring for household members 4.6% Sports, exercise & recreation 4.1% Caring for non household members 3.9% Caring for non household members 3.3% Religious and spiritual activities 3.4% Professional & personal care services 1.9% Volunteering activities 1.5% Volunteering activities 1.4% Professional & personal care services 1.4% Other discretionary activitie s 1.0% Other discretionary activitie s 1.2% Religious and spiritual activities 0.9% Personal care 1.0% Personal care 0.8% Household services 0.6% Household services 0.3% School 0.5% Table 3 3 . Summary Statistics on health measures. Category Poor Fair Good Very good Excellent Total Within SAPHS (%) SU U 2.2 1.6 1.5 1.6 2.2 1.7 N 25.8 24.0 29.3 39.8 54.7 37.0 OW 29.8 31.8 35.4 37.7 32.6 35.0 Ob I 20.3 22.3 21.9 15.1 7.9 16.9 Ob II 11.0 11.2 7.7 4.2 1.8 5.9 Ob III 10.9 9.0 4.2 1.6 .7 3.5 Within BMI (%) SU U 5.7 11.7 26.0 31.8 24.8 N 3.1 8.0 23.5 37.0 28.3 OW 3.8 11.2 30.0 37.1 17.8 Ob I 5.4 16.3 38.6 30.8 9.0 Ob II 8.2 23.3 38.3 24.5 5.7 Ob III 13.7 31.4 35.3 15.9 3.7 Total 4.4 12.3 29.7 34.4 19.1
39 Table 3 4 . Distribution of walking behavior by health measures. Weekday Weekend Type Category Nonwalkers (%) Walkers (%) Nonwalkers (%) Walkers (%) Threshold 10 min 15 min 10 min 15 min 10 min 15 min 10 min 15 min BMI SU U 91.2 95.6 8.8 4.4 95.1 95.9 4.9 4.1 N 92.9 95.8 7.1 4.2 94.2 96.4 5.8 3.6 OW 94.8 96.9 5.2 3.1 95.5 97.0 4.5 3.0 Ob I 95.9 97.3 4.1 2.7 96.2 97.8 3.8 2.2 Ob II 95.6 97.4 4.4 2.6 97.0 98.4 3.0 1.6 Ob III 97.1 98.5 2.9 1.5 96.5 97.7 3.5 2.3 SAPHS Poor 95.5 96.9 4.5 3.1 97.3 98.2 2.7 1.8 Fair 94.7 96.5 5.3 3.5 95.3 96.6 4.7 3.4 Good 94.0 96.3 6.0 3.7 95.2 97.0 4.8 3.0 Very good 94.6 96.9 5.4 3.2 95.4 97.1 4.6 2.9 Excellent 93.8 96.7 6.2 3.3 94.7 96.7 5.3 3.3 Table 3 5 . Distribution of health measures by walking behavior. Weekday Weekend Type Category Nonwalkers (%) Walkers (%) Nonwalkers (%) Walkers (%) Threshold 10 min 15 min 10 min 15 min 10 min 15 min 10 min 15 min BMI SU U 1.5 1.5 2.4 2.0 1.9 1.9 2. 2.6 N 36.3 36.6 46.4 45.6 36.6 36.8 45.1 44.6 OW 35.3 35.2 32.4 32.7 34.9 34.8 33.1 34.5 Ob I 17.3 17.1 12.4 13.6 16.9 16.8 13.4 12.4 Ob II 6.1 6.0 4.6 4.5 6.0 6.0 3.8 3.1 Ob III 3.5 3.5 1.7 1.5 3.8 3.7 2.7 2.8 SAPHS Poor 4.4 4.4 3.5 4.0 4.6 4.6 2.6 2.8 Fair 12.4 12.3 11.6 13.0 12.3 12.2 12.3 13.8 Good 29.6 29.6 31.3 32.3 29.6 29.7 30.3 29.3 Very good 34.9 34.8 33.0 32.5 34.2 34.2 33.1 32.8 Excellent 18.7 18.9 20.5 18.2 19.3 19.3 21.7 21.2
40 Table 3 6 . Effect of walking (10 minute threshold) on health. BMI [Weekday] BMI [Weekend] SAPHS [Weekday] SAPHS [Weekend] B S.E. B S.E. B S.E. B S.E. Walk probability 7.42 1.63 8.16 2.13 1.56 0.302 1.00 0.365 Walk + Exercise probability 8.97 2.90 2.05 0.877 Exercise probability 0.895 0.130 0.559 0.146 Female 0.883 0.084 1.09 0.088 0.125 0.018 0.077 0.018 Age 20 39 3.10 0.220 3.06 0.210 0.431 0.047 0.271 0.047 Age 40 59 3.84 0.224 4.20 0.214 0.641 0.046 0.506 0.048 Age 60+ 3.76 0.254 3.86 0.254 0.743 0.053 0.660 0.056 Black 2.01 0.150 2.01 0.153 0.214 0.027 0.210 0.027 Asian 2.12 0.217 2.54 0.198 0.286 0.049 0.106 0.049 Hispanic 0.768 0.152 0.922 0.152 0.237 0.029 0.253 0.027 Income <$35k 0.561 0.109 0.621 0.109 0.187 0.021 0.210 0.021 HS/GED education 0.218 0.030 College 0.542 0.098 0.528 0.099 0.479 0.030 0.346 0.019 Student 0.513 0.224 0.611 0.210 0.149 0.044 Retired 0.699 0.169 0.788 0.180 0.167 0.040 Part time job 0.746 0.131 0.502 0.142 0.303 0.038 0.252 0.039 Full time job 0.360 0.035 0.314 0.035 Unemployed 0.158 0.041 0.190 0.028 Own house 0.461 0.152 0.138 0.027 0.052 0.021 Nuclear household 0.107 0.019 0.087 0.027 Other multi adult household 0.606 0.140 0.385 0.134 Food stamps recipient 1.58 0.229 1.35 0.219 0.380 0.037 0.327 0.036 Constant 24.6 0.227 25.0 0.281 1.72 0.072 1.82 0.073 tau1 n/a n/a 0.878 0.070 0.863 0.072 tau2 n/a n/a 1.88 0.070 1.86 0.072 tau3 n/a n/a 2.93 0.071 2.90 0.072 Goodness of fit R 2 0.0743 0.0837 0.0655 0.0671
41 Table 3 7 . Effect of walking (15 minute threshold) on health. BMI [weekday] BMI [weekend] SAPHS [weekday] SAPHS [weekend] Parameter B S.E. B S.E. B S.E. B S.E. Walk probability 11.979 2.724 7.33 3.29 2.44 0.494 2.10 0.555 Walk + exercise 13.27 5.55 2.89 1.031 Exercise 0.840 0.127 0.558 0.125 Female 0.891 0.084 1.10 0.09 0.127 0.018 0.078 0.018 Age 20 39 3.543 0.173 3.31 0.20 0.444 0.047 0.398 0.053 Age 40 59 4.333 0.168 4.43 0.20 0.656 0.046 0.628 0.053 Age 60+ 4.301 0.204 4.10 0.24 0.765 0.053 0.771 0.060 Black 1.961 0.148 1.99 0.15 0.216 0.027 0.204 0.027 Asian 2.246 0.208 2.52 0.20 0.267 0.048 0.112 0.049 Hispanic 0.743 0.149 0.916 0.16 0.239 0.029 0.242 0.028 Income <$35k 0.476 0.112 0.698 0.11 0.196 0.021 0.204 0.021 HS/GED education 0.231 0.030 0.235 0.030 College 0.618 0.097 0.567 0.10 0.495 0.031 0.524 0.030 Student 0.473 0.21 0.163 0.046 0.193 0.026 Retired 0.734 0.170 0.811 0.18 Part time job 0.712 0.130 0.479 0.14 0.283 0.039 0.235 0.039 Full time job 0.351 0.035 0.306 0.035 Unemployed 0.178 0.042 0.169 0.040 Own house 0.442 0.149 0.143 0.028 Nuclear household 0.109 0.019 0.060 0.021 Other multi adult household 0.594 0.139 0.264 0.13 0.093 0.027 Food stamps recipient 1.524 0.228 1.33 0.22 0.388 0.037 0.317 0.036 Constant 24.575 0.235 24.30 0.20 1.737 0.070 1.762 0.072 tau1 n/a n/a 0.878 0.069 0.869 0.071 tau2 n/a n/a 1.876 0.069 1.872 0.070 tau3 n/a n/a 2.93 0.070 2.91 0.071 Goodness of fit R 2 0.0742 0.083 0.0656 0.0685
42 Table 3 8 . Effect of health on walking (10 minute threshold). Weekday Weekend Weekday Weekend B S.E. B S.E. B S.E. B S.E. BMI 0.039 0.007 0.040 0.007 n/a n/a SAPHS n/a n/a n.s. 0.137 0.037 Female 0.248 0.078 0.212 0.078 Age 20 39 0.654 0.132 0.770 0.130 Age 40 59 0.590 0.137 0.743 0.134 Age >=60 0.576 0.173 0.723 0.171 Asian 0.608 0.159 0.697 0.158 Black 0.669 0.098 0.610 0.105 0.603 0.097 0.561 0.104 Hispanic 0.672 0.095 0.565 0.099 0.639 0.094 0.549 0.099 HS/GED 0.541 0.110 0.583 0.110 College education 0.466 0.100 0.523 0.100 Part time work 0.569 0.122 0.549 0.122 Full time work 0.627 0.105 0.640 0.105 0.241 0.083 Couple household 0.409 0.108 0.425 0.108 Single parent household 0.391 0.154 0.347 0.161 0.409 0.154 0.363 0.161 Nuclear household 0.554 0.105 0.282 0.085 0.570 0.105 0.293 0.085 Other multi adult hh 0.427 0.121 0.460 0.120 Retired 0.455 0.156 0.452 0.155 At home maintenance 0.170 0.022 0.136 0.021 0.167 0.022 0.134 0.021 Non home maintenance 0.117 0.013 0.083 0.017 0.118 0.013 0.082 0.017 At home discretionary 0.143 0.058 0.141 0.058 At home discretionary 2 0.005 <0.001 0.009 0.002 0.005 <0.001 0.009 0.002 Non home discretionary 2 0.007 0.002 0.005 0.002 0.007 0.002 0.005 0.002 Northeast US Census 0.527 0.091 0.527 0.091 0.611 0.098 Midwest US Census 0.299 0.104 0.311 0.104 0.276 0.113 Southern US Census 0.681 0.098 0.695 0.098 0.626 0.106 HH tenure 0.850 0.078 1.035 0.079 0.835 0.077 1.051 0.079 Metropolitan area 0.640 0.123 0.658 0.133 0.667 0.123 0.666 0.133 Jun Aug Season 0.294 0.091 0.279 0.091 Constant 1.29 0.294 0.848 0.541 0.408 0.251 2.206 0.516 Goodness of Fit Nagelkerke R 2 0.143 0.137 0 .137 0 . 0 134 Note: n/a (not applicable)
43 Table 3 9 . Effect of health on walking (15 minute threshold). Weekday Weekend Weekday Weekend B S.E. B S.E. B S.E. B S.E. BMI 0.04 0.008 0.045 0.009 n/a n/a SAPHS n/a n/a n.s. 0.099 0.044 Female 0.275 0.093 0.245 0.093 Age 15 19 0.518 0.15 0.672 0.146 Age >=60 0.503 0.215 Asian 0.631 0.12 Black 0.674 0.12 0.552 0.131 0.661 0.117 0.469 0.132 Hispanic 0.661 0.116 0.727 0.115 0.655 0.117 Dropped school 0.399 0.114 0.468 0.114 Student 0.622 0.111 Unemployed 0.631 0.111 0.361 0.112 0.423 0.113 Part time work 0.412 0.103 Single person household 0.392 0.103 Nuclear household 0.206 0.105 At home mandatory 0.209 0.032 0.172 0.036 0.156 0.036 Non home mandatory 0.218 0.019 0.172 0.022 0.203 0.032 0.168 0.022 At home maintenance 0.227 0.029 0.189 0.026 0.217 0.019 0.187 0.026 Non home discretionary 0.148 0.025 0.221 0.029 At home discretionary 2 0.008 0.001 0.006 0.001 0.146 0.025 0.006 0.001 Non home discretionary 2 0.012 0.002 0.008 0.001 0.012 0.002 Northeast US Census 0.608 0.116 0.745 0.103 0.615 0.118 Midwest US Census 0.29 0.137 0.617 0.116 0.284 0.142 Southern US Census 0.549 0.125 0.51 0.118 0.287 0.137 0.64 0.131 HH tenure 0.989 0.093 0.982 0.095 0.549 0.126 0.934 0.098 Metropolitan area 0.598 0.157 0.733 0.171 0.975 0.093 0.726 0.172 Nov Mar Season 0.239 0.095 0.61 0.157 Mar May Season 0.239 0.095 Constant 0.64 0.401 0.235 0.363 0.505 0.329 1.576 0.316 Goodness of Fit Nagelkerke R 2 0.149 0.134 0.144 0.129 Note: n/a (not applicable)
44 CHAPTER 4 IMPUTING HEALTH MEASURES BY DATA FUSION Introduction The previous chapter presented models that capture the relationships between utilitarian walking and health measures. However, these models did not include any controls for land use variables location) . This was primarily because the ATUS data used in the analysis do not provide detailed spatial information about the residential location of the survey respondents. Other surveys of daily tr avel such as the National Household Travel Surveys (NHTS) do provide detailed (latitude and longitude) residential location information. However, these surveys do not collect information on the health of the survey respondents. Therefore, developing models that examine walking, health, and land use either require new surveys that collect all these data or the use of data fusion approaches that systematically link individual records from ATUS and NHTS surveys to generate a comprehensive dataset . In this stud y , we explore the latter approach of data fusion. Insights from this study can help provide insights into the need for additional surveys in the future to comprehensively study the land use transportation health relationships. In this chapter, two met hods for performing disaggregate data fusion were examined. The first method (regression based matching) involves first developing a regression model of BMI using socio dataset and, subsequently, using this mode l to predict the BMI for the respondents in a identify the -
45 defined socio In order to evaluate these methods, it is necessary to know the BMI of the based and the probabilistic matching methods. The probabilistic matching method is shown to be better and is subsequently used (next chapter) to link the ATUS and NHTS records. Application of the T wo M ethods A comparison of Donor and Receiver datasets and socioec onomic indicators is shown on Table 4 1. Both ATUS samples follow similar shares, including a female majority (55%), as well as some higher education (57%), and full time employed (55%). The datasets provide a fair representation of non civilian U.S. adult s, including participation of all family types and income levels. The regression based matching method is conceptually straightforward and requires a regression model developed on the Donor dataset to be applied to the Receiver dataset . Table 4 2 presents the results of such OLS regression as a function of matching attributes significant at 95% confidence or higher . This probabilistic data fusion employed Link Plus, a probabilistic record linkage program developed at the U.S. CDC, based on the theoretical f ramework of Fellegi and Sunter and Demptser et al. (57) . A review of key elements for common record linkage and previous applications can be found in Chapter 2, under Data Fusion. D isaggregate (individual level) li n king is performed on the basis of several variables. The software allows these variables of interest to be specified as either
46 blocking (Table 4 3) or matching (Table 4 4) variables. In this study, gender and the census region of the household (Northeast, Midwest, South, and West) were used as si the Northeast region (in the receiver dataset) will necessarily be obtained from males from the Northeast in the donor dataset. As such the matching on these attributes is d eterministic (the software will report that a matching record could not be found is such a deterministic match is not possible). Variables that are not blocked are matched probabilistically. Three matching methods are employed: (a) exact matching (binary all or nothing comparison), (b) generic string (comparison of characters and the needed number of operations needed to transform one string into the other), and value specific (sets weights matching values based on the frequencies of values in the file bei ng compared). Matching variables and methods are included in Table s 4 3 and 4 4 . In preparation for matching, certain variables were recoded from their usual binary or continuous values. The use of generic string in age, race, employment and income accounts for the ordinal associations or expected associations by coding these as a series of similar strings ( i.e., match young adults with other young adults (20 39), similarly for middle adults (40 59) and late adults (60+); increase likelihood of matching minorities with other minorities (Hispanic, Asian, Black, other), and Hispanics with C aucasian and Black; decrease likelihood of full time or part time employment to be matched with unemployed individuals). The age string was constructed as a letter describing age range (A, <20; 20 < B < 40; 40 < C < 60; D, >60). The ethnicity/race string u ses and XYZ format, where X assigns minority status for Hispanics, Asians, and
47 Blacks (yes/no), Y differentiates between Caucasian, Hispanic, Black, Asian, and Other, and Z assigns a similar value to Caucasian and Blacks due to the possible misrepresentati on of Hispanic populations as Caucasian or Black. The race match hierarchy follows: perfect match (e.g. , Caucasian with Caucasian), Hispanic related minorities (e.g. , Hispanic with Black), minorities (e.g. , Asian with Black) or Hispanic related (e.g. , Hisp anic with Caucasian), Caucasian and non Hispanic related (e.g. , Caucasian with Asian), and last any catch , Other with Hispanic). Employment match follows the next hierarchy: perfect match (e.g. , unemployed with unemployed), any employed (e.g. , Part time with Full time), unemployed/part time, and last unemployed/full time. Income hierarchy follows: perfect match (e.g. , low income with low income), one level difference (e.g. , high income with medium income), two level difference (e .g. , high income with low income), presence of missing ( e.g., missing with high income).The employment of value specific type matching for household size lies in the expected resolution to prioritize match of unusual cases, usually higher household sizes, with each other ( e.g., household of six with another household of six). The M probability is the probability that a matching variable agrees given that the comparison pair being examined is a match ; t he same M probability applies for all records . Likewise one to many matches are possible and allowed. The u probabilities are calculated based on the donor dataset, and represents the probability that a matching variable agrees given that comparison pair being examined as a non match . The u probabilities are e stimated based on the frequencies of reported values in the donor dataset.
48 For each record pair, the weights across all fields are calculated from M probabilities and u probabilities to obtain the total weight. The highest weights corresponds to pairs whi ch agree across all fields prior probability that a randomly selected record from file A matches a randomly selected record from file B. Threshold score values assign t rue matches, matches for review, and non matches. For this exercise, matches for review a re assumed to be true matches and the cut off value for non matches is a weight of zero . Multiple donor dataset size iterations revealed the largest available dataset (full donor sample) as the best performing. However, the duplication of cases within the sample that share matching attributes but differ on reported BMI (matches for review) led to the adoption of the multi run mean as a way to recognize these differences. The mean BMI was determined by resorting and rematching the donor dataset five times and finding the average of the five matched BMI values. Validation A comparison of mat ched BMI , predicted BMI by regression and reported BMI shows the record linkage being able to provide a more varied range of health measures than the regression values (BMI range of 31.6 versus 13.3). Likewise , the loose linkage criteria allowed the matchi ng of Obese II category individuals, values that were not predicted by regression models. Figures 4 1 thru 4 4 show a comparison of predicted BMI , the matched ATUS BMI , frequency and their error distribution . Inspection of errors (Table 4 5) shows similar performance of the mean matched model in comparison to the deterministic model . A comparison of matched and reported BMI, as shown on Figure 4 1 (each color represents a different matching run) allowe d for some assignment of extreme values
49 interchangeably, undesirable but arguably an example of similarities across unhealthy BMIs in both spectrums (under and overweight). Conversely, the deterministic model offered a narrower match, bounded by normal and overweight predictions. Th is contrast between matched and deterministic approach is also evident when considering the freque ncy distributions shown in Figure 4 3 (B ) and Figure 4 3 (H ) against the known reported values as shown on Figure 4 3 (A ). This probabilistic linkage method allowed for matching of ATUS BMI and SAPHS information to NHTS pairs . The validation of record linkage within the ATUS cases allowed for all but 5 cases to receive a record pair. While a deterministic approach is dominated by t he sample mean, the probabilistic linkage may reflect BMI ranges (i.e. , normal, overweight, obese) more closely to expected values.
50 Table 4 1. Descriptive statistics of ATUS donor and receiver databases. Classification Description Donor Receiver Sample Size No. randomly selected member per household 25,599 10,000 Sex Male 45.0% 44.9% Female 55.0% 55.0% Age Average age of respondent 46.1 45.9 Race Identifies as Caucasian 69.0% 69.0% Identifies as Black/African American 13.0% 13.0% Identifies as Hispanic 13.0% 13.0% Identifies as Asian 3.0% 3.0% Education Did not finish high school 16.7% 16.9% High school/GED degree 28.5% 26.2% Completed or some advanced degree education 57.0% 57.0% Employment Person enrolled in school (any level, part or full time) 6.0% 5.9% Person identified as unemployed 18.0% 17.0% Person identified as part time worker 12.0% 11.0% Person identified as full time worker 54.0% 55.0% Household Household members 2.75 2.80 Income Combined household income up to $35k 31.4% 31.5% Combined household income between $35k $75k 29.6% 28.7% Combined household income greater than $75k 25.6% 26.3% HH Tenure Residence owned by a household member 75.0% 75.0% Region Household located in Northeast census region 17.0% 16.0% Household located in Midwest census region 25.0% 24.0% Household located in South census region 36.0% 37.0% Household located in West census region 22.0% 23.0% Metro Household located within a Metropolitan Area 82.0% 82.0%
51 Table 4 2 . Deterministic model Dataset B employed in validation. Parameter B S.E. p value Constant 24.8 0.30 <0.01 Age 0.04 0.00 <0.01 Female 0.84 0.09 <0.01 Black 2.02 0.13 < 0.01 Hispanic 0.70 0.14 <0.01 Asian 2.54 0.25 <0.01 Student 2.19 0.21 <0.01 Unemployed 0.59 0.16 <0.01 Full time work 0.92 0.14 <0.01 Retired (work) 0.71 0.19 <0.01 College education 0.21 0.09 0.02 Low income HH 0.70 0.10 <0.01 Other multi adult HH 0.38 0.13 <0.01 Household size 0.10 0.03 <0.01 HH tenure 0.30 0.11 0.01 Metropolitan area 0.47 0.11 <0.01 Midwest US 0.64 0.13 <0.01 Southern US 0.40 0.12 <0.01 Northeast US 0.35 0.14 0.01
52 Table 4 3 . Blocking variables and method. Data item Type Matching method Gender Binary Exact* Midwest census region Binary Exact* Northeast census region Binary Exact* Southern census region Binary Exact* West census region Binary Exact* *Note: Blocking all variables as AND and not OR Table 4 4 . Matching variables and methods. Data item Type Matching method M prob U prob College education binary exact 0.95 0.5186 Metropolitan area binary exact 0.95 0.6593 Household tenure binary exact 0.95 0.6953 Student binary exact 0.95 0.9087 Age string generic string 0.95 0.0168 Race string generic string 0.95 0.6360 Employment string generic string 0.95 0.3788 Income string generic string 0.95 0.3309 Household size continuous value specific 0.95 0.2191
53 Figure 4 1. Matched BMI vs. reported BMI for five runs.
54 Figure 4 2. Mean matched BMI (dark) and predicted BMI (light) vs. reported values.
55 Figure 4 3. Frequency distribution of BMI: (A) reported, (B) deterministic model, (C G ) m atched runs 1 5, and (H) mean matched.
56 Figure 4 4. Error distributions: (A) deterministic model, ( B F ) matched runs 1 5 , (G) mean matched. Table 4 5. Errors across models. Model MAE Average error Variance Deterministic 0.16 0.038 0.037 Match 1 0.22 0.043 0.083 Match 2 0.22 0.041 0.082 Match 3 0.22 0.039 0.085 Match 4 0.22 0.041 0.084 Match 5 0.21 0.033 0.082 Mean match 0.19 0.039 0.060
57 CHAPTER 5 USE OF THE FLORIDA NHTS TO LINK LAND USE AND HEALTH The object of this chapter is to explore the relationship between land use, healthy individua ls and their propensity to walk. This chapter employs matched health information from ATUS with of Florida households that participated in the NHTS. Data Assembly The assembly uses data from the 2006 2008 American Time Use Surveys (ATUS) and the 2009 National Household Travel Survey (NHT S). Both ATUS and NHTS collected detailed socio economic, demographic, and one day activity travel information for a large sample of persons. Initial sample includes one household member from each household that (a) has no missing trip purpose, gaps or ove rlap in their individual trip schedule, (b) is not reported as being out of the country during the travel survey day, (c) is 15 years old or older. Individuals who stay at home (no out of home activities) are included in the sample. Household selection: Fo r the NHTS, all Florida households with known location and with weekday travel surveys were selected (close to 9,000 ). For the ATUS, all available cases were selected, or about 35,000 households. Person selection: To meet ATUS limitations, all individuals under age 15 or with unknown age were removed from NHTS sample and only one person per household was selected. Selection of that individual included all one person household members and random selection of household member for multiple member households. F or households with more than one individual, uniform random values were assigned to all and the person with smallest value was selected.
58 The NHTS trip files reports one day out of home travel from 4:00 am to 4:00 am (of the following day) on a specified tr avel day. Trip duration and reported mode are used to determine out of home utilitarian walking. Walking trips that did not end at home by the end of the day, would be assigned the last known trip purpose and duration calculated to 4 a . m . of the following day. Sample For this study, both socio economic and walking data are available for 5,168 ( Florida NHTS) and 35,599 (ATUS) individuals. A comparison of shared sample characteristics is presented in Table 5 1, a dditional description of ATUS dataset activities can be found in C hapter 3 Tables 3 1 to 3 2. T he ATUS and NHTS sample s have similar attributes, w ith age being a discerning factor for Florida . To address the Florida age gap, the final NHTS sample includes adults under the age of 70 . The final sample also excludes individuals surveyed on weekend (about 20% of the cases), as the model application focused on weekday travel behavior . The represented samples are largely Caucasian and there are more females than males (55% versus 45%). The survey re spondents had some secondary degree education and were significantly likely to be employed full time. The most common household liv es in an owned property within a Metropolitan Statistical Area according to 2000 Census (have at least one urbanized area of 50,000 or more population, plus adjacent territory that has a high degree of social and economic integration with the core). Data This analysis uses a representative subset of weekday behavior for 5,168 people that participated in the 2009 Florida add on of the NHTS . The NHTS collected detailed socio economic, de mographic, and one day travel information for a large sample of
59 persons. A comparison of walking pattern yields higher shares of utilitarian walking reported in the NHTS (1 6.6 % vs 5%) . Employing th e 10 min and 15 min bout recommendations described in Chapter 2, 12.1% and 9.4% of the entire weekday sample employed walking mode for at least one leg of their trips , respectively . The add on component enhances walking information by identifying the spati al location of the household as a set of coordinates , allowing for supporting information on the surrounding neighborhood , a common trip producer and attractor . The ATUS Florida NHTS match followed the steps recreated in Chapter 4. A successful record lin kage allowed for 100 % of the NHTS households to be matched and follow similar demographics of ATUS (Table 5 1) . As Figure 5 1 demonstrates, matching occurred over a wide range of BMI values. Table 5 2 shows a BMI and observed utilitarian walking breakdown by walking guidelines. Built environment infor mation is derived from the U.S. Environmental Protection Agency Smart Location Database (EPA SLD) ( 70 ) , a nationwide geographic dataset at Census block group level of detail. The dataset included more than 90 attributes summarizing characteristics such as housing density, diversity of land use, neighborhood design, destination accessibility, transit service, employment , and demographics from the 2010 decennial Census, the 2010 American Community Survey, 2011 In foUSA, the Census Longitudinal Employer Household Dynamics (LEHD) survey, 2010 Census TIGER/Line, Navteq Navstreets, and the 2011 General Transit Feed Specification (GTFS) .
60 Land U se D escriptors within the state of Florida using the provided Florida add on latitude and longitude in ArcGIS . Weekday trips and activities were reported for a period of 24 hours by all household members. Census block level information, taken from the EPA SLD dataset was superimposed to aggregate and apportion population, employment, and transportation information at Â½ mile and 1 mile buffer radii from household location (see Table 5 3 ). Employment was measured through the employment mix variable with the entropy denomin ator set to account observ ed existing employment types (retail, office, service, industrial, entertainment, education, healthcare, and public administration) within each block group. The SLD classified all street links as either auto oriented (pedestrian r estricted or burdensome), multi modal (urban, arterials ), or pedestrian oriented ( low speed arterials, locals , trails ) and calculated the facility miles for each. Counts of intersections for each classification accounted for three and four link intersecti ons within the group. Transit stop information was not available for the entire state and therefore not directly accounted. Model At first glance, BMI and reported walking behavior may not offer much differentiation, therefore BMI is modeled using a linea r regression model as part of a two stage approach . G have been adopted (based on minimum duration thresholds of 10 and 15 minutes), all the models are estimated using each of these definitions of walking. B ecause the short term (daily) decision of walking on a single weekday outcome measures that are generally descriptors of longer term condition, an
61 instrumental ity of walking on a week Recognizing that leisure walking/exercise could play a substitute or complementary role, OLS model was developed to determine the probability of utilitarian walking as a function of the socio economic characteristics of the in dividual and his/her household and other characteristics of time use on that day ( Tables 5 4 and 5 5 ) . These models use each of the two time definitions are then used to calculate the predicted probability of walking for each person in the analysis sample. T he Nagelkerke R 2 goodness of fit value is reported for comparison, as the calculated value is normalized to be between 0 and 1 . The predicted models for health. These models employ the mean BMI matching methodology developed from ATUS sample in the previous chapter. The 3 5,599 cases available for the ATUS sample were matched in five instances to 5,168 available Florida add on person households. The mean BMI value was then assigned to each person for subsequent analysis. Findings As Figure 5 1 indicates, disaggregate matching of ATUS health values to N HTS respondents occurred ove r a wide range of BMI s, particularly those in the overweight and normal range . The comprehensive models for health outcomes are presented in Table 5 6 . All effects reported are statistically significant at 95% confidence or higher. The predicted probability of weekday walking (10 minutes or more) showed no impact on BMI after controlling for respondent and household characteristics as well as a land use attribute . A similar finding occurs with t he predicted probability of weekday walking 15 minute recommendation . This is not
62 consistent with the national sample trend. The availability of land use variables and household location showed discernible impact using the half mi le radius. Miles of pede strian oriented facilities were associated with lower BMIs. No other population or entropy measure remained significant after accounting for personal demographics and household characteristics . Among these characteristics, increased education and higher ho usehold income s were associated with lower BMIs in the Florid a sample . Overall , this analysis was able to show a n application of a utilitarian walking , land use and health measurement model by linking the NHTS (travel), ATUS (health), and SLD (land use) datasets . The use of a statewide Florida travel survey and a land use database reflected the impact of pedestrian facilities up to a half mile from the household location on walking for transportation, even after controlling for socioeconomic indicators . In particular, greater pedestrian facilities were associated with desirable BMIs, but the strength of this association remains uncertain. Practitioners acquainted with the NHTS or other travel surveys with discrete household location could benefit from t his methodology when enhancing their mod els. Outcomes from the Florida models continues to support the positive impact of a strong pedestrian oriented network around residential locations (low speed roads, sidewalks, trails and paths).
63 Table 5 1. Compa rison of Florida NHTS sample and the ATUS sample. Matched ATUS donor attributes NHTS Run 1 Run 2 Run 3 Run 4 Run 5 Receiver Sample size 35,599 35,599 35,599 35,599 35,599 5,1 68 % matched 100% 100% 100% 100% 100% n/a Gender Male 46.5% 46.5% 46.5% 46.5% 46.5% 46.5% Female 53.5% 53.5% 53.5% 53.5% 53% 53.5% Age <20 2.2% 2.2% 2.2% 2.2% 2.2% 2.1% 20 39 16.2% 16.2% 16.2% 16.2% 16.2% 16.4% 40 59 44.1% 44.2% 44.2% 44.1% 44.1% 44.1% 60 69 37.5% 37.4% 37.4% 37.5% 37.5% 37.5% Race Hispanic 7.4% 7.5% 7.5% 7.4% 7.5% 9.2% Black 7.2% 7.2% 7.5% 7.4% 7.5% 6.0% Asian 0.4% 0.4% 0.4% 0.5% 0.4% 0.9% Caucasian 84.2% 84.1% 83.9% 83.9% 83.8% 81.3% Education College (some/completed) 23.1% 23.1% 23.1% 23.1% 23.1% 21.2% Employment Partial/multiple jobs 26.4% 26.3% 28.8% 29.6% 28.5% 12.5% Full time job 34.5% 34.5% 32.1% 31.3% 32.3% 48.4% Household Members 2.09 2.09 2.09 2.09 2.09 2.10 Owned HH 90.8% 90.8% 90.8% 90.8% 90.8% 90% HH Income <$35k 22.2% 23.9% 23.0% 21.1% 26.9% 26.9% HH Income $35k $75k 40.0% 39.4% 39.7% 39.4% 38.0% 38.0% HH Income $75k+ 28.6% 28.5% 28.7% 28.7% 28.0% 28.0% Location Within Metropolitan region 81.2% 81.2% 81.2% 81.2% 81.2% 79% Western US Region 0.2% 0.3% 0.2% 0.3% 0.1% n/a Northeast US Region 0.2% 0.2% 0.2% 0.1% 0.25% n/a Midwest US Region 0.1% 0.1% 0.1% 0.1% 0.12% n/a Southern US Region 99.5% 99.5% 99.5% 99.5% 99.50% 100% Note: n/a ( not applicable )
64 Figure 5 1. Histogram of BMIs matched to Florida sample. Table 5 2. BMI and walking participation in Florida sample. Duration BMI range 10 min 15 min Underweight (all) 15.0% 5.0% Normal 11.9% 9.2% Overweight 12.3% 9.5% Obese (I,II,III) 11.8% 9.4%
65 Table 5 3 . Land use descriptors 1/2 mi buffer 1 mi buffer Min Max Mean Min Max Mean Total population within nearby CBSA no CBSA 5,564,635 2,155,637 no CBSA 5,564,635 2,151,154 Total workers within nearby CBSA no CBSA 2,118,833 837,350 no CBSA 2,118,833 835,791 Total employment within nearby CBSA no CBSA 2,088,064 822,581 no CBSA 2,088,064 820,990 Population 2.3 2895 406 7.40 4855 441 Households 1.6 1247 188 3.48 2081 207 Workers 0.9 1060 157 2.72 1694 166 Total jobs 0.0 2787 125 0.00 5540 144 Retail jobs 0.0 554 20 0.00 1186 21 Office jobs 0.0 967 12 0.00 388 13 Industrial jobs 0.0 1068 20 0.00 2345 27 Service related jobs 0.0 863 22 0.00 867 25 Entertainment jobs 0.0 690 18 0.00 739 19 Education jobs 0.0 1197 8 0.00 1948 12 Health affiliated jobs 0.0 845 18 0.00 911 21 Public service jobs 0.0 743 6 0.00 458 6 8 tier employment entropy [0 1] 0.0 1.0 0.6 0.00 0.6 0.1 Miles of auto oriented links 0.0 10.0 1.2 0.00 33665.9 35.7 Miles of multimodal oriented links 0.0 7.3 1.3 0.00 25.3 5.3 Miles of pedestrian oriented links 0.3 28.9 11.3 1.30 93.4 42.7 Auto oriented intersections 0.0 12.0 0.9 0.00 40.7 3.5 Pedestrian oriented intersections having four or more legs 0.0 25.0 10.4 0.00 405.9 49.8
66 Table 5 4 . Predicted w eekday 10 min walk Parameter B S.E. Sig. Past week walk trips 0 .057 0 .005 < 0 .00 1 Part time job 0.350 0.145 0 .016 Full time job 0.428 0.097 < 0 .00 1 HH income >75k 0.289 0.099 0 .004 Shared HH vehicle 0 .995 0 .221 < 0 .00 1 Non shared vehicle HH 1.065 0 .200 < 0 .00 1 Total Educational jobs [1 mi] 0 .001 0 .000 0 .028 Total pedestrian links [1 mi] 0 .008 0 .002 < 0 .00 1 Constant 1.526 0.224 < 0 .00 1 Goodness of fit Nagelkerke R 2 0.080 Table 5 5 . Predicted w eekday 15 min walk Parameter B S.E. Sig. Past week walk trips 0 .050 0 .005 <0 .00 1 Part time job 0 .380 0 .162 0 .019 Full time job 0 .562 0 .109 <0 .00 1 HH income >75k 0 .327 0 .114 0 .004 Shared HH vehicle 0 .929 0 .243 <0 .00 1 Non shared vehicle HH 1.129 0 .209 <0 .00 1 Number of HH adults 0 .188 0 .082 0 .023 Total pedestrian links [1 mi] 0 .006 0 .002 0 .012 Constant 1.22 0 . 264 <0 .00 1 Goodness of fit Nagelkerke R 2 0 .074
67 Table 5 6 . Effect of land use and predicted walking on health BMI Parameter B S.E. Sig. Predicted 10 min walking probability n/s Predicted 1 5 min walking probability n/s Total pedestrian links (mi) [ 1/2 mi] 0.025 0.009 0 .00 6 Age 0.027 0.004 <0.001 Black 0.780 0.221 <0.001 Hispanic 0.823 0.186 <0.001 College education 0.690 0.128 <0.001 Part time job 0.401 0.170 0 .018 Full time job 0.481 0.117 <0.001 HH children (total) 0.192 0.097 0.0 48 Medium income HH 0.312 0.125 0.013 High income HH 1.296 0.139 < 0 .001 Constant 27.273 0.279 < 0 .001 Goodness of fit Adjusted R 2 0 .0 54 Note: n/s (not significant at p=0.05)
68 CHAPTER 6 CONCLUSION : WHERE DO WE GO FROM HERE? Utilitarian w alking can be part of a set of strategies employed to facilitate recommended physical activity levels for healthy adults. Despite the fact the qualitative impact of walking on health can be inferred from the energy balance model, empirical studies that capture the effects of land use patterns are necessary to ascertain the magnitud e of the impact of walking on health . While studies on built environment, walking and physical health exists in the fields of public health, urban planning, and transportation, comprehe nsive models addressing all their aspects are limited by lack of health data, in particular transportation surveys . Health surveys often do no collect travel behavior data and/or do not provide detailed spatial resolution to construct land use descriptors. This research supports the incorporation of health measures such as Bo dy Mass Index and Self Assessed Physical Activity Score (SAPHS) in future transportation studies and household travel surveys. The availability of one reported ( BMI ) and one perceived measure ( SAPHS ), allowed for their contrast as well. While BMI exemplifi es a long er term variable, SAPHS can represent shorter durations. H igher BMI does constrict utilitarian walking ; a lower SAPHS (poor health perception) did not seem to carry this pattern in a national sample after controlling for exercise, nutrition, and socioeconomic status. Utilitarian walking has often been measured in inconsistent ways. The review and adoption of two threshold values already part of health policy guidelines (continuous bouts of 10 and 15 minutes) reinforce the expected relationship, wh ile supporting a diminishing magnitude after controlling for multiple indicators and separate weekend effects . It is evident that exercise and nutrition continue to play a substantial role.
69 This research also addressed th e travel diary and health gap in the literature by merg ing transportation and health surveys using data fusion techniques to develop comprehensive models of land use, utilitarian walking, and health. The use of probabilistic matching to populate a statewide Florida travel survey w ith BMI values proved a practical application . R ecord linkage allowed for the production of extreme cases and an aggregate distribution similar t o the general population. T he national sample showed some impact of utilitarian walking on health, while the Fl orida sample , with an older population , did not show such correlation . The presence of pedestrian facilities within a half mile radius of the household and dense locations continue to be associated with lower BMIs for the Florida statewide sample . The work conducted continues to support much of the knowledge gained from local, scoped studies through a large, national survey. It also offers an alternative to unknown health information through the use of probabilistic matching of common parame ters to link NHTS person files with ATUS records . Last ly , the work highlight ed the need to continuously support pedestrian facilities and cautions how increasing U.S. overweight and obesity rates could be detrimental to future mode shifts to active transpo rtation. Future research should aim at examining the relationships between long term walking and health as well as incorporating perception of nearby infrastructure . The availability o f individual nutrition patterns could also enhance current resea rch. Moving forward, researchers and policy makers can re examine the impact and projections on active transportation and relevant transportation facilities as the U.S. population experiences changes in its health profile. Similarly, practitioners can
70 inco rporate available health measures at earlier stages of their planning and modeling processes as opposed to final stage outputs of health and air quality . From a Florida perspective, the unique composition of the state population ma y warrant a closer inspec tion of transportation facilities among the elderly and young adults .
71 LIST OF REFERENCES 1. Blackwell, D. L., J. W. Lucas, and T. C. Clarke. Summary Health Statistics for U. S . Adults: National Health Interview Survey, 2012. Vital and Health Statistics. Series 10, Data from the National Health Survey , Vol. 260, No. 260, 2014, pp. 1 171. 2. Iacono, M., K. J. Krizek, and A. El Geneidy. Measuring Non Motorized Accessibility: Issues, Alternatives, and Execution. Journal of Transport Geography , Vol. 18, No. 1, 20 10, pp. 133 140. 3. U.S. Department of Transportation, Federal Highway Administration. 2009 National Household Travel Survey , 2014. http://nhts.ornl.gov . Accessed Jul. 31, 2014. 4. U.S. Census Bureau. 2011 American Community Survey , 2014. http://www.census.gov/a cs/www/data_documentation/2011_release/ . Accessed Jul. 30, 2014. 5. Milne, A., and M. Melin. Bicycling and Walking in the United States: 2014 Benchmarking Report. 2014, pp. 268. 6. U.S. Environmental Protection Agency. Transportation Conformity Rule Restructurin g Amendments. 77 FR 14979 , 2012 . 7. Elsom, D. Smog Alert: Managing Urban Air Quality. Routledge , 1996 . 8. Bernardo, C., and C. Bhat. Non Motorized Travel as a Sustainable Travel Option. Handbook of Sustainable Travel , Springer Netherlands, 2014, pp. 277 291. 9. Sallis, J., A. Bauman, and M. Pratt. Environmental and Policy Interventions to Promote Physical Activity. American Journal of Preventive Medicine , Vol. 15, No. 4, 1998, pp. 379 397. 10. NHLBI Obesity Education Initiative Expert Panel. Clinical Guidelines on th e Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. 1998. 11. Tremblay, M. S., R. C. Colley, T. J. Saunders, G. N. Healy, and N. Owen. Physiological and Health Implications of a Sedentary Lifestyle. Applied Physiology, Nutrition, a nd Metabolism , Vol. 35, No. 6, 2010, pp. 725 740. 12. Slentz, C. A., B. D. Duscha, J. L. Johnson, K. Ketchum, L. B. Aiken, G. P. Samsa, J. A. Houmard, C. W. Bales, and W. E. Kraus. Effects of the Amount of Exercise on Body Weight, Body Composition, and Measure s of Central Obesity: STRRIDE a Randomized Controlled Study. Archives of Internal Medicine , Vol. 164, No. 1, 2004, pp. 31 39.
72 13. Physical Activity Guidelines Advisory Committee. Physical Activity Guidelines for Americans. Washington, DC: US Department of Heal th and Human Services , 2008, pp. 15 34. 14. United States. Department of Health. Physical Activity and Health: A Report of the Surgeon General. DIANE Publishing, 1996. 15. United States. Public Health Service. Office of the Surgeon General, National Center for Chronic Disease Prevention, Health Promotion (US), President's Council on Physical Fitness, and Sports (US). Physical Activity and Health: A Report of the Surgeon Gene ral. Jones & Bartlett Learning, 1996. 16. Sallis, J. F., L. D. Frank, B. E. Saelens, and M. K. Kraft. Active Transportation and Physical Activity: Opportunities for Collaboration on Transportation and Public Health Research. Transportation Research Part A: Pol icy and Practice , Vol. 38, No. 4, 2004, pp. 249 268. 17. Lawrence, D. L., and S. M. Low. The Built Environment and Spatial Form. Annual Review of Anthropology , 1990, pp. 453 505. 18. Handy , S . L . , M.G. Boarnet, R. Ewing , and R.E. Killingsworth. How the built enviro nment affects physical activity: Views from urban planning . Am erican J ournal of Prev entive Med icine , Vol 23, No. 2, Supplement 1, 2002 , pp. 64 73. 19. Sallis, J. F., N. Owen, and E. B. Fisher. Ecological Models of Health Behavior. Health Behavior and Health Education: Theory, Research, and Practice , Vol. 4, 2008, pp. 465 485. 20. Corbin, C. B., R. P. Pangrazi, and B. D. Franks. Definitions: Health, Fitness, and Physical Activity. President's Council on Physical Fitness and Sports Research Digest , 2000. 21. Frank, L., D, J. Kerr, J. Sallis F, R. Miles, and J. Chapman. A Hierarchy of Sociodemographic and Environmental Correlates of Walking and Obesity. Preventive Medicine , Vol. 47, No. 2, 2008, pp. 172 178. 22. Wanner, M., T. GÃ¶tschi, E. Martin Diener, S. Kahlmeier, and B. W . Martin. Active Transport, Physical Activity, and Body Weight in Adults: A Systematic Review. American Journal of Preventive Medicine , Vol. 42, No. 5, 2012, pp. 493 502. 23. Pucher, J., R. Buehler, D. R. Bassett, and A. L. Dannenberg. Walking and Cycling to H ealth: A Comparative Analysis of City, State, and International Data. American Journal of Public Health , Vol. 100, No. 10, 2010, pp. 1986 1992. 24. Van Dyck, D., E. Cerin, G. Cardon, B. Deforche, J. F. Sallis, N. Owen, and I. de Bourdeaudhuij. Physical Activit y as a Mediator of the Associations between Neighborhood Walkability and Adiposity in Belgian Adults. Health & Place , Vol. 16, No. 5, 2010, pp. 952 960.
73 25. Cleland, V., K. Ball, C. Hume, A. Timperio, A. C. King, and D. Crawford. Individual, Social and Environ mental Correlates of Physical Activity among Women Living in Socioeconomically Disadvantaged Neighbourhoods. Social Science & Medicine , Vol. 70, No. 12, 2010, pp. 2011 2018. 26. Bostock, L. Pathways of Disadvantage? Walking as a Mode of Transport among Low inc ome Mothers. Health & Social Care in the Community , Vol. 9, No. 1, 2001, pp. 11 18. 27. Saunders, L. E., J. M. Green, M. P. Petticrew, R. Steinbach, and H. Roberts. What are the Health Benefits of Active Travel? A Systematic Review of Trials and Cohort Studies . PLoS ONE , Vol. 8, No. 8, 2013, pp. e69912. 28. Andersen, L. B., P. Schnohr, M. Schroll, and H. O. Hein. All Cause Mortality Associated with Physical Activity during Leisure Time, Work, Sports, and Cycling to Work. Archives of Internal Medicine , Vol. 160, No. 11, 2000, pp. 1621 1628. 29. Hayashi, T., K. Tsumura, C. Suematsu, K. Okada, S. Fujii, and G. Endo. Walking to Work and the Risk for Hypertension in Men: The Osaka Health Survey. Annals of Internal Medicine , Vol. 131, No. 1, 1999, pp. 21 26. 30. Besson, H., U. Ek elund, S. Brage, R. Luben, S. Bingham, K. Khaw, and N. J. Wareham. Relationship between Subdomains of Total Physical Activity and Mortality. Medicine and Science in Sports and Exercise , Vol. 40, No. 11, 2008, pp. 1909 1909 1915. 31. Sallis, J. F., and N. G. Ow en. Physical Activity and Behavioral Medicine. Sage Publications, 1998. 32. Trost, S. G., N. Owen, A. E. Bauman, J. F. Sallis, and W. Brown. Correlates of Adults' Participation in Physical Activity: Review and Update. Medicine & Science in Sports & Exercise , V ol. 34, No. 12, 2002, pp. 1996 2001. 33. Harris, J. K., J. Lecy, J. A. Hipp, R. C. Brownson, and D. C. Parra. Mapping the Development of Research on Physical Activity and the Built Environment. Preventive Medicine , Vol. 57, No. 5, 2013, pp. 533 540. 34. Ding, D., and K. Gebel. Built Environment, Physical Activity, and Obesity: What have we Learned from Reviewing the Literature? Health & Place , Vol. 18, No. 1, 2012, pp. 100 105. 35. Ferdinand, A. O., B. Sen, S. Rahurkar, S. Engler, and N. Menachemi. The Relationship bet ween Built Environments and Physical Activity: A Systematic Review. American Journal of Public Health , Vol. 102, No. 10, 2012, pp. e7 e13.
74 36. Heath, G. W., R. C. Brownson, J. Kruger, R. Miles, K. E. Powell, L. T. Ramsey, and Task Force on Community Preventive Services. The Effectiveness of Urban Design and Land use and Transport Policies and Practices to Increase Physical Activity: A Systematic Review. Journal of Physical Activity & Health , Vol. 3, 2006, pp. S55. 37. Durand, C. P., M. Andalib, G. F. Dunton, J. Wolch, and M. A. Pentz. A Systematic Review of Built Environment Factors Related to Physical Activity and Obesity Risk: Implications for Smart Growth Urban Planning. Obesity Reviews , Vol. 12, No. 5, 2011, pp. e173 e182. 38. Lachowycz, K., and A. Jones. Greenspace and Obesity: A Systematic Review of the Evidence. Obesity Reviews , Vol. 12, No. 5, 2011, pp. 183 189. 39. McCormack, G. R., and A. Shiell. In Search of Causality: A Systematic Review of the Relationship between the Built Environment and Physical Activity among Adults. Int J Behav Nutr Phys Act , Vol. 8, No. 125, 2011, pp. 101 186. 40. Pulleyblank Patrick, S., M. Raimi, R. Ewing, L. D. Frank, J. Chapm an, and R. Kreutzer. Understanding the Relationship between Public Health and the Built Environment: A Report Prepared for the LEED ND Core Committee. U.S. Green Building Council, 2006. 41. Saelens, B. E., and S. L. Handy. Built Environment Correlates of Walking: A Review. Medicine and Science in Sports and Exercise , Vol. 40, No. 7 Suppl, 2008, pp. 550 566 . 42. Handy, S., X. Cao, and P. Mokhtarian. Correlation o r Causality between the Built Envir onment and Travel Behavior? Evidence from Northern California. Transportation Research Part D: Transport and Environment , Vol. 10, No. 6, 2005, pp. 427 444. 43. Frank, L. D., B. E. Saelens, K. E. Powell, and J. E. Chapman. Stepping Towards Causation: Do Built Environments Or Neighborhood and Travel Preferences Explain Physical Activity, Driving, and Obesity? Social Science & Medicine , Vol. 65, No. 9, 2007, pp. 1898 1914. 44. Knuiman, M. W., H. E. Christian, M. L. Divitini, S. A. Foster, F. C. Bull, H. M. Badland, a nd B. Giles Corti. A Longitudinal Analysis of the Influence of the Neighborhood Built Environment on Walking for Transportation: The RESIDE Study. American Journal of Epidemiology , Vol. 180, No. 5, 2014, pp. 453 461. 45. Mackenbach, J. D., H. Rutter, S. Comper nolle, K. Glonti, J. Oppert, H. Charreire, I. De Bourdeaudhuij, J. Brug, G. Nijpels, and J. Lakerveld. Obesogenic Environments: A Systematic Review of the Association between the Physical Environment and Adult Weight Status, the SPOTLIGHT Project. BMC Publ ic Health , Vol. 14, No. 1, 2014, pp. 233.
75 46. Feng, J., T. A. Glass, F. C. Curriero, W. F. Stewart, and B. S. Schwartz. The Built Environment and Obesity: A Systematic Review of the Epidemiologic Evidence. Health & Place , Vol. 16, No. 2, 2010, pp. 175 190. 47. Hes s, D. B., and J. K. Russell. Influence of Built Environment and Transportation Access on Body Mass Index of Older Adults: Survey Results from Erie County, New York. Transport Policy , Vol. 20, March , 2012, pp. 128 137. 48. Wen, M., and L. Kowaleski Jones. The B uilt Environment and Risk of Obesity in the United States: Racial ethnic Disparities. Health & Place , Vol. 18, No. 6, 2012, pp. 1314 1322. 49. Rundle, A., A. V. Diez Roux, L. M. Free, D. Miller, K. M. Neckerman, and C. C. Weiss. The Urban Built Environment and Obesity in New York City: A Multilevel Analysis. American Journal of Health Promotion , Vol. 21, No. 4 Suppl, 2007, pp. 326 334. 50. Bauman, A., and F. Bull. Environmental Correlates of Physical Activity and Walking in Adults and Children: A Review of Reviews. London: National Institute of Health and Clinical Excellence , 2007. 51. Frank, L. D., J. F. Sallis, T. L. Conway, J. E. Chapman, B. E. Saelens, and W. Bachman. Many Pathways from Land use to Health: Associations between Neighborhood Walkability and Active Tra nsportation, Body Mass Index, and Air Quality. Journal of the American Planning Association , Vol. 72, No. 1, 2006, pp. 75 87. 52. Doyle, S., A. Kelly Schwartz, M. Schlossberg, and J. Stockard. Active Community Environments and Health: The Relationship of Walka ble and Safe Communities to Individual Health. Journal of the American Planning Association , Vol. 72, No. 1, 2006, pp. 19 31. 53. Mackenbach, J., H. Rutter, S. Compernolle, K. Glonti, J. Oppert, H. Charreire, I. De Bourdeaudhuij, J. Brug, G. Nijpels, and J. La kerveld. Obesogenic Environments: A Systematic Review of the Association between the Physical Environment and Adult Weight Status, the SPOTLIGHT Project. BMC Public Health , Vol. 14, No. 1, 2014, pp. 233. 54. Gebel, K., A. E. Bauman, T. Sugiyama, and N. Owen. M ismatch between Perceived and Objectively Assessed Neighborhood Walkability Attributes: Prospective Relationships with Walking and Weight Gain. Health & Place , Vol. 17, No. 2, 2011, pp. 519 524. 55. Ullmann, S. H., N. Goldman, and A. R. Pebley. Contextual Fact ors and Weight Change Over Time: A Comparison between U.S. Hispanics and Other Population Sub Groups. Social Science & Medicine , Vol. 90, No. 0, 2013, pp. 40 48.
76 56. Barry, A. An Analysis of the Perception of the Built Environment, Physical Activity, and Body Mass Index in Allegheny County, Pennsylvania . Doctoral Dissertation, University of Pittsburgh, 201 3. 57. Glazier, R. H., M. I. Creatore, J. T. Weyman, G. Fazli, F. I. Matheson, P. Gozdyra, R. Moineddin, V. K. Shriqui, and G. L. Booth. Density, Destinations o r B oth? A Comparison of Measures of Walkability in Relation to Transportation Behaviors, Obesity and Diabetes in Toronto, Canada. PloS One , Vol. 9, No. 1, 2014, pp. e 85 2 95. 58. Gu, L., R. Baxter, D. Vickers, and C. Rainsford. Record Linkage: Current Practice and Future Directions. CSIRO Mathematical and Information Sciences Technical Report , Vol. 3, No. 83, 2003. 59. Jaro, M. A. Advances in Record Linkage Methodology as Applied to Ma tching the 1985 Census of Tampa, Florida. Journal of the American Statistical Association , Vol. 84, No. 406, 1989, pp. 414 420. 60. Winkler, W. E. Frequency Based Matching in Fellegi Sunter Model of Record Linkage. Bureau of the Census Statistical Research Div ision , Vol. 14, 2000. 61. Baker, K., P. Harris, and J. O'Brien. Data Fusion: An Appraisal and Experimental Evaluation. Journal of the Market Research Society , 1989 , pp. 153 212 . 62. Kusakabe, T., and Y. Asakura. Behavioural Data Mining of Transit Smart Card Data: A Data Fusion Approach. Transportation Research Part C: Emerging Technologies , Vol. 46, 2014, pp. 179 191. 63. Amorim, M., S. Ferreira, and A. Couto. Linking Police and Hospital Road Accident Records. Transportation Research Record: Journal of the Transportati on Research Board , Vol. 2432, No. 1, 2014, pp. 10 16. 64. Rosman, D. L. The Western Australian Road Injury Database (1987 1996) : Ten Years of Linked Police, Hospital and Death Records of Road Crashes and Injuries. Accident Analysis & Prevention , Vol. 33, No. 1 , 2001, pp. 81 88. 65. Williamson, A., and S. Boufous. A Data Matching Study of the Role of Fatigue in Work Related Crashes. Transportation Research Part F: Traffic Psychology and Behaviour , Vol. 10, No. 3, 2007, pp. 242 253. 66. Thakuriah, V., J. Lee, and S. Nium pradit. Motor Carrier Safety Databases: Strategies f or Developing Linkages, Assessing Completeness a nd Making Imputations. In International Truck and Bus Safety Research and Policy Symposium , 2002.
77 67. Pawlak, J., J. W. Polak, and A. Sivakumar. An Imputation Approach to the Fusion of Travel Diary and Lifestyle Data: Application to the Analysis of the Interaction of ICT and Physical Mobility. In New Techniques and Technologies for Statistics Conference. , Brussels, Belgium, 2013. 68. Kressner, J. D., and L. A. Garrow. Using Third Party Data for Travel Demand Modeling: Comparison of Targeted Marketing, Census, and Household Travel Survey Data. In Transportation Research Board 93rd Annual Meeting , 2014. 69. Dunton, G . F., D. Berrigan, R. Ballard Barbash, F. M. Perna, B. I. Graubard, and A. A. Atienza. Adolescents' Sports and Exercise Environments in a U.S. Time use Survey. American Journal of Preventive Medicine , Vol. 39, No. 2, 2010, pp. 122 129. 70. Rams ey, Kevin, and Alexander Bell. The smart location database: A nationwide data resource characterizing the built environment and destination accessibility at the neighborhood fiscalement and destination accessibi lity at the neighborhood scale. Cityscape: A Journal of Policy Development and Research , Vol 16, No. 2 , 2014 .
78 BIOGRAPHICAL SKETCH Miguel Lugo graduated from the transportation e ngineering doctoral program within the Department of Civil and Coastal Engineering at the University of Florida Engineerin g School of Sustainable Infrastructure & Environment. He received a Master of Engineering in Civil Engineering with a M inor in Urban and Regional Planning from the University of Florida and a B achelor of Science in Civil Engineering from the University of Puerto Rico at MayagÃ¼ ez. During his time at Florida, he has was recognized with a pre doctoral fellowship from the Louis Stokes Alliance for Minority Participation Program, as well as service and leadership awards through various professional and community leadership roles. His research interests include public infrastructure policy, active transportation , and travel behavior.