https://orcid.org/0000-0001-7397-7048
Figures
Abstract
This study compared marginal and conditional modeling approaches for identifying individual, park and neighborhood park use predictors. Data were derived from the ParkIndex study, which occurred in 128 block groups in Brooklyn (New York), Seattle (Washington), Raleigh (North Carolina), and Greenville (South Carolina). Survey respondents (n = 320) indicated parks within one half-mile of their block group used within the past month. Parks (n = 263) were audited using the Community Park Audit Tool. Measures were collected at the individual (park visitation, physical activity, sociodemographic characteristics), park (distance, quality, size), and block group (park count, population density, age structure, racial composition, walkability) levels. Generalized linear mixed models and generalized estimating equations were used. Ten-fold cross validation compared predictive performance of models. Conditional and marginal models identified common park use predictors: participant race, participant education, distance to parks, park quality, and population >65yrs. Additionally, the conditional mode identified park size as a park use predictor. The conditional model exhibited superior predictive value compared to the marginal model, and they exhibited similar generalizability. Future research should consider conditional and marginal approaches for analyzing health behavior data and employ cross-validation techniques to identify instances where marginal models display superior or comparable performance.
Citation: Wende ME, Hughey SM, McLain AC, Hallum S, Hipp JA, Schipperijn J, et al. (2024) Identifying multilevel predictors of behavioral outcomes like park use: A comparison of conditional and marginal modeling approaches. PLoS ONE 19(4): e0301549. https://doi.org/10.1371/journal.pone.0301549
Editor: Sathishkumar Veerappampalayam Easwaramoorthy, Sunway University, MALAYSIA
Received: November 17, 2023; Accepted: March 18, 2024; Published: April 16, 2024
Copyright: © 2024 Wende et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: This study was supported by the National Cancer Institute (www.cancer.gov) under Award number R21CA202693 (PI: Andrew Kaczynski). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funding agency did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Park use is associated with greater community cohesion [1,2], mental health [3], stress reduction [4,5], and physical activity (PA) [6,7]. Despite this, the percentage of the population that did not use parks increased from 25% to 30% from 1992 to 2015 [2]. To increase park use [8–10], it is important to identify key predictors of park use [11,12].
Methods for collecting information on park use predictors are diverse, including assessments of neighborhood, park, and individual-level characteristics [6,13–16]. At the neighborhood or administrative level, geographic information systems (GIS) tools are often used to map and quantify park availability [17], in conjunction with U.S. Census indicators (e.g., median household income) [18]. At the park level, audits conducted with reliable and valid tools are increasingly common [19], including the Community Park Audit Tool (CPAT) [20]. CPAT measures park access, neighborhood features, activity areas, quality, and safety [21,22]. At the individual level, intercept surveys administered to individuals visiting parks capture information about behaviors or sociodemographic characteristics [23,24]. While few studies sample park users and non-users [21,22,25–27], global positioning systems (GPS) and location-based resident surveys provide opportunities to study predictors of park use and non-use [22,28–36].
With more innovative and detailed methods for data collection, analyses that account for correlated observations are increasingly common. Observations can be clustered if there are multiple observations for the same individual (at several time points), or if data is hierarchically organized in the same geographic location (e.g., city, neighborhood) or organizational unit (e.g., school, workplace). In general, there are two popular types of statistical models used to account for correlated observations: conditional and marginal models [37,38]. For conditional models, clustered observations are accounted for through the inclusion of shared random effects in the linear predictor [39]. For marginal modeling, the mean function is modeled directly and the correlation structure is regarded as a nuisance parameter (i.e., it is not of interest) [37,38]. Conditional models specify a model on observation-level data, requiring assumptions on the random effect distribution and the independence of clustered observations given the random effects, which are often impossible to verify [37–40]. Alternatively, marginal methods model the average response on the population level and are robust to misspecification of the dependence structure [37,38]. Marginal models also make inferences about population (not individuals), which is advantageous for studying the impacts of multi-level factors on neighborhood park use.
In previous studies, analytical techniques that assume independent observations (e.g., ordinary least squares (OLS) regression) have been used to understand the relationship between multilevel factors and park use [8,21,22,41–44]. Loukaitou-Sideris and Sideris used OLS multiple linear regression to identify factors that bring children to parks, including recreation facilities, sport programs, natural features, and maintenance [42]. Additional studies have accounted for correlated observations, and largely used such methods to control for non-independence between people residing within the same geographic area using mixed regression, conditional models with random effects [45–52]. Stewart et al. linked GPS and individual travel diary data with park audit data and GIS measures for park access to examine if park facilities were related to the amount of park-based PA [53]. Authors used conditional models with random effects at the individual, park, and individual-park combination level [53].
Most past research on multi-level park use predictors employs conditional modeling. Limited research has compared conditional and marginal modeling techniques to determine advantages and disadvantages of different approaches for analyzing and interpreting multilevel park use predictors [38]. Therefore, this study will compare marginal and conditional modeling techniques for identifying neighborhood, park, and individual level park use predictors. Methods from our study can be replicated to inform decision making on model selection for research on multilevel predictors of park use and other behavioral outcomes, and results can provide robust input on significant predictors of park use.
Materials and methods
Study design and setting
Data came from a parent study, ParkIndex, which is described fully elsewhere [21,54]. Data collection and recruitment occurred from Spring to Fall 2017 in four major U.S. localities: Brooklyn, NY, Greenville County, SC, Seattle, WA, & Raleigh, NC.
Census block groups (CBG) were classified into quartiles for park availability and income. CBGs in each location were enumerated by identifying all CBGs that fall within city and county limits (using Census shapefiles). Park availability and income quartiles were based on the number of parks intersecting the CBG and American Community Survey (ACS) 2011–2015 5-year estimates [18]. Using methods similar to other studies [45], 32 CBGs were selected within each location (128 total). This study included all public parks designed to facilitate active or passive use with an area size ≥ .25 acres. GIS park files for pre-selected CBGs were obtained from local parks and recreation agencies. A sample of residents in pre-selected CBGs were recruited to use an online, map-based survey (Maptionnaire.com) that included written consent (although consent was waived for this study) and displayed neighborhood parks to report park use and non-use and answer related questions [21]. Study procedures were granted exempt status by the University of South Carolina Institutional Review Board.
Data collection
From June to October 2017, all parks in the study areas were audited in person by trained research staff using the CPAT [55]. CPAT contains four sections related to park information, access and surrounding area, activity areas, and quality/safety [55], and has demonstrated excellent inter-rater reliability and validity [46,56–59]. Overall, 275 parks were audited across the study areas: 94 in Seattle, 64 in Brooklyn, 71 in Raleigh, and 46 in Greenville County.
Adult residents within each CBG were recruited to take part in the study. Addresses in 32 CBGs per city were identified and mailed postcards containing a brief study description and online survey link. Postcards were mailed to verified residential addresses from June to October 2017, three separate times over a three-month period. Maptionnaire was used for administering the online survey [60]. Participants were shown a map of their neighborhood (i.e., area within a half-mile of the resident’s CBG) and were asked to locate and answer questions about parks they had used within the past month, until the participant indicated they had not used any other parks in their neighborhood within the past month.
Outcome measure
Park visitation. Participants were matched to parks within a ½-mile of their CBG and each of the person-park pairs was assigned yes/no based on reported park use in the past month.
Explanatory measures–Neighborhood level
Count of neighborhood parks. The count of parks within a ½-mile (calculated by measuring the Network distance on ArcGIS 10.2.2) of each participant’s CBG of residence.
Median household income. CBG-level data for the park addresses were downloaded from U.S. Census Bureau’s ACS (5-year estimates, 2011–2015) [18]. Median household income was categorized into site-specific tertiles (low, intermediate, high).
Population density. CBG-level data for park addresses were downloaded from U.S. Census Bureau’s ACS (5-year estimates, 2011–2015) [18]. Population density (population/area) was calculated and categorized into site-specific tertiles (low, intermediate, high).
Age structure. CBG-level data on the percentage of the population over the age of 65 years were downloaded from the U.S. Census Bureau’s ACS (5-year estimates, 2011–2015) [18].
Racial and ethnic composition. CBG-level data for the park addresses were downloaded from the U.S. Census Bureau’s ACS (5-year estimates, 2011–2015) [18], to calculate the percentage of the population identifying as non-Hispanic White.
Walkability. CBG-level National Walkability Index (1–20), based on the Environmental Protection Agency’s Smart Location Mapping Database [61].
Explanatory measures–Park level
Distance to neighborhood parks. Distance between participant addresses to each verified park entrance was calculated using Network distances on ArcGIS 10.2.2. Distances from residences to parks ranged from 0.01 to 12.7 miles (mean = 0.99, median = 0.82), since participants were asked to report park use within a half-mile buffer of their CBG of residence.
Park quality. Park quality (0–100) was calculated using CPAT average park access, facilities, amenities, quality concerns, aesthetics, and facility quality. For each component, scores were determined using yes/no responses to presence, usability and condition [21,22].
Park size. We calculated total acreage of each park, using GIS files for all selected parks.
Explanatory measures–Participant level
Physical activity. Participants indicated the number of days and minutes per day spent engaged in moderate and vigorous PA in the past 7 days using the short form International PA Questionnaire [62]. Total number of weekly minutes of moderate and vigorous PA (MVPA) were summed for each participant [63], and categorized into low, intermediate, and high tertiles.
Socio-demographics. Respondents reported their race (White, American Indian, Alaskan native, Asian, Black or African American, Native Hawaiian, Pacific Islander, other), and gender (male, female, other). Age (in years) was calculated using the date of survey completion and reported date of birth. Race (White/Non-White) and gender (male/female) were dichotomized.
Data management
A total of 360 residents completed the survey, and their data was merged with audit data for all parks within a ½-mile of their CBG of residence. The resultant long-form dataset included 3247 observations, representing all available neighborhood parks for each person. A total of 40 participants (relating to 157 observations) were removed from the analytic sample because they did not have any parks within a ½-mile of their CBG of residence or they only reported using parks outside the ½-mile CBG buffer, resulting in a final sample of 3090 observations and 320 participants. Each observation (n = 3090) included individual-level data (which was the same for all observations from the same participant), park-level data (i.e., audit data unique to each observation/park, reported park use), and CBG-level data (i.e., specific to each park address). On average, participants had 15 observations (i.e., 15 parks within the half-mile CBG buffer).
Statistical analyses
Descriptive statistics.
Descriptive information for participants included in this study were calculated and included ANOVA and chi-square tests to determine crude relationships between participant characteristics and park use. To account for participant non-response of survey data related to age (missing = 74), gender (missing = 45), PA (missing = 105), race and ethnicity (missing = 19), and education (missing = 48), hot deck single imputation was used (with PROC SURVEYIMPUTE, SAS 9.4). Hot deck imputation is a method for handling missing data in which each missing value is replaced with an observed response from a "similar" unit [64].
Conditional model.
To evaluate park use predictors identified using varying statistical techniques, a two-level hierarchical conditional model was used with the imputed dataset. Level 1 was the participant-level and level 2 were the clusters of the units/participants nested within the same CBGs. The number of observations per participant related to the number of parks they could potentially report using in their CBG. Generalized Linear Mixed Models (GLMM) were used to examine the level 1 outcome (i.e., participant park use) as a function of predictors on level 1 and 2 using PROC GLIMMIX (SAS 9.4 software; see S1 File) with a quadrature estimation technique. Park-level clustering was not specifically accounted for, given that participants in the same CBGs were assigned to the same neighborhood parks. Participant’s use of each park was treated as a binary outcome with a logit link function. Variance components were estimated for participant and neighborhood (i.e., CBG) level random effects. Models were presented after adjustment and included participant and CBG level random intercepts, fixed effects for different park visits, and individual level exposures. The equation below illustrates the conditional model predicting park use (Yij, binary outcome) of participant i in park j:
where Xij and Zj are the exposure variables, and Rij and Uj are participant and CBG level random effects. Random effects are assumed to be normally distributed with Var(Rij) = σ2 and Var(Uj) = τ2. Model assumptions include: 1. distribution of the data conditional on the random effects is known, 2. probability of participant park use takes the form of a logistic mixed model, 3. random effects are normally distributed, 4. objective function for the optimization is a function of either the actual log likelihood, an approximation to the log likelihood, or the log likelihood of an approximated model [65]. Median Odds Ratio (MOR) quantified the amount of variance in the outcome for each level of the hierarchical model [66], using
, where Va is the area level variance and 0.6745 is the 75th percentile of the cumulative distribution function of the normal distribution with mean 0 and variance 1 [66–68]. Before adjustment, CBG level MOR = 1.5 indicates that if a person were to move to a CBG with higher park use, median increase in the odds of park use (over all possible CBG) would increase by 50% [66]. Person level MOR = 1.5 indicate that if we compare all pairs of people in the same CBG, median OR, arranged such that the person with higher odds of park use is in the numerator, is 1.5 [66].
Marginal model.
For analyzing marginal models for this study, a hierarchical structure with individuals nested within similar CBGs was used to estimate population level averages with the imputed dataset. As stated previously, the data for this study included multiple observations per participant, with different observations relating to unique neighborhood parks and their characteristics. Relationships between park use and participant and park-level exposure variables were assessed using Generalized Estimating Equations (GEE) with a specified binary outcome distribution and a logit link function (i.e., PROC GEE, SAS 9.4 software; see S1 File). Repeated statements for CBGs and individuals nested within the same CBGs were included, and non-independence of the same parks used by different participants was accounted for on the CBG level. Quasi-likelihood-based criteria (QIC) values for the models with an independence correlation structure and with an exchangeable working correlation structure were compared to ensure independent correlation structure was best for this data. Model assumptions include: 1. the error terms for the same participant/geographic unit are correlated, 2. there is a linear relationship between the covariates and the transformation of the response (using logit link function, in this case), and 3. the covariance structure is correctly specified (GEE are robust to this last assumption). In addition, it is necessary that the sample size is sufficiently large (overall and within CBGs) for robust estimation of standard errors [38]. Models were presented after adjustment for different park visits and independent exposures. The equation below illustrates the marginal model predicting park use (Yij, binary outcome) of participant i in park j:
where X are explanatory variables, and t refers to unique park visits for each person.
Model comparisons.
We compared the direction of beta estimates for the same explanatory variables in full, adjusted GLMM and GEE models. We also compared which explanatory variables were significant predictors of park use in final (or reduced) GLMM and GEE models. Final, reduced models were chosen using backwards stepwise selection (based on α = 0.05 cutoff) and model fit. To assess model fit, penalized-likelihood information criteria were calculated for the GLMM model (i.e., Akaike & Bayesian Information Criterion), and quasi-likelihood-based criteria (i.e., QIC) were calculated for the GEE model. Marginal model odds ratios were interpreted as population averages, and the conditional model odds ratios were interpreted for the participant/CBG (i.e., holding all values fixed, including random effects).
The out of sample prediction error for each modeling approach were compared using tenfold cross validation methods [69]. Cross validation can assess the performance of predictive models to determine the generalizability of results [69], by randomly partitioning the original sample into 10 equal size subsamples. A single subsample is used as the validation data for testing the model, and the remaining 9 subsamples are used as training data. This process is repeated 10 times until each of the 10 subsamples are used once as validation data, and then the 10 results are averaged to produce a single estimation (S1 File) [69]. To compare predictive value using cross validation, we compared GLMM and GEE predictions based on fixed and random effects. To assess generalizability, we compared predictions based on only fixed effects. Data for this study can be found in S2 File.
Results
Sample characteristics
A total of 320 participants were included in the analytic sample, with 68.1% (n = 218) who reported using any neighborhood park(s) in the past month and 31.9% (n = 102) who did not (Table 1). Participants mostly reported intermediate PA levels (37.5%, n = 120), identified as White (69.1%, n = 221), were female (58.4%, n = 187), and had 2–4 years of college education (46.2%, n = 148). Average age of the sample was 47.1 years (SD = 15.3) (Table 1). Participants also had, on average, 57.4 park acres and 18.5 parks in their neighborhoods. These parks were 1.2 miles away and had a park quality score of 48.5, on average. Table 1 also shows the crude relationships between participant-level characteristics and park use. Results show that only education level (χ2 = 12.4, p = 0.0021) and park distance (F = 17.4, p < .0001) had a significant relationship with park use.
Model specifications
MOR estimates showed that if a person moves to another CBG with a higher probability of park use, their risk of park use will (in median) increase 3.1 times (MOR = 3.07). Block group level effects explained a large proportion of the variance in the outcome. This median increase in the OR is larger than the estimated OR for any predictor included in our model (see Table 2).
Comparison of predictors of park use
Marginal and conditional models identified the same significant predictors of park use (and similar effect estimates) in the full model, aside from park size (Table 2 and S1 Fig). Significant predictors of park use in the reduced, conditional model included participant race, participant education, distance to parks, park size, park quality, and percent of the neighborhood population age 65 and older (Table 2). For the participant level, there were 43% decreased odds of park use for Non-White compared to White participants, and 90% increased odds of park use for those with 2–4 year college and 149% increased odds of park use for those with an advanced degree compared to less than college (holding other predictors fixed). For the park level, there were 78% increased odds of park use for every mile distance to neighborhood parks, 0.07% increased odds of park use for every one unit increase in park acreage, and 2% increased odds of park use for every one unit increase in park quality for each given participant (holding other predictors fixed). For the neighborhood level, there were 2% increased odds of park use for every percent increase in residents over 65 years (holding other predictors fixed, including random effects).
Significant predictors of park use in the reduced, marginal model included participant race, participant education, distance to parks, park quality, and percent of the neighborhood population age 65 and older (Table 2). For the participant level, on average, there were 34% decreased odds of park use for Non-White compared to White participants, 104% increased odds of park use for those with 2–4 year college, and 158% increased odds of park use for those with an advanced degree compared to less than college. For the park level, on average, there was 95% increased odds of park use for every mile distance to parks and there were 2% increased odds of park use for every one unit increase in park quality. For the neighborhood level, on average, there were 2% greater odds of park use for every one percent increase in residents over 65 years.
Conducting tenfold cross validation with both fixed and random effects for the conditional model, Root Mean Squared Error (RMSE) for the conditional model was 0.32 compared to 0.36 for the marginal model (Table 2). Mean Absolute Percentage Error (MAPE) for the conditional model was 0.22, compared to 0.52 for the marginal model (Table 2). RMSE and MAPE were lower for the conditional model, meaning conditional models had superior predictive value of results. With only fixed effects for the conditional model, RMSE for the conditional model was 0.36 compared to 0.36 for the marginal model. MAPE for the conditional model was 0.52, compared to 0.52 for the marginal model (Table 2). RMSE and MAPE were similar between conditional and marginal models, so generalizability was comparable.
Discussion
Conditional and marginal models identified similar park use predictors on the individual- and park-levels. Both models identified participant race, participant education, distance to each park from residence, park quality, and percentage of the neighborhood population over 65 years as park use predictors. Additionally, the conditional model identified park acreage as a predictor, although the effects of park acreage on park use in the conditional model were modest. Results from ten-fold cross validation showed that the conditional model exhibited superior predictive value compared to the marginal model, but they exhibited similar generalizability.
Previous research that used conditional models have reported that number of nearby parks [46], neighborhood park space [46], park proximity [46,50], neighborhood walkability [48,49], park quality [46,50], and violent crime were park use predictors [52]. Few studies indicated that White and higher income or educated groups were more likely to report park use compared to racial minority populations and lower income or educated populations [46,48]. Our results showed that increased distance to neighborhood parks related to higher odds of park use. This contradicts past research showing that perceived park proximity positively impacts park use [48], and other research establishing a null association between objectively-measured proximity and park use [70]. This finding could in part be explained by the fact that participants were only able to report parks used within a ½-mile of their CBG of residence (ranging 0.01 to 12.7 miles). Even the furthest parks were still somewhat proximal to participant residences. Though logic may indicate that a close park would result in greater visitation, growing evidence also shows that park use and park-based PA is influenced by a variety of complex factors, including facilities present, quality, social interaction, safety, and demographic characteristics [21,22,41–43,45–49,71–74]. Further, our sample was predominately White and educated, which may indicate greater access to resources that influence travel patterns to parks, such as car transportation [75–77].
Findings showing conditional models had superior predictive value compared to marginal may partially validate the use of conditional modeling in past research [45–52]. Cross validation results also demonstrated that generalizability is similar between marginal and conditional models when only comparing predicted values using fixed effects. Past research has compared conditional and marginal approaches [78–80], and stated that marginal (GEE) models may be favorable since they are robust to misspecification of the covariance structure [78,79]. Conditional (GLMM) models only provide unbiased interpretation when the correct random effects has been selected [78], so the significance of the factors that contribute to the different behavioral outcomes are more reliable from marginal models [78,79]. Our research introduced ten-fold cross validation as a method for assessing predictive value and generalizability of these different statistical approaches that can be replicated [69]. For interpretations, the conditional model slope coefficients reflect the average within neighborhood parks log(OR) of park use and the marginal model log(OR) reflects probabilities of park use averaged across all neighborhood parks, both as a result of significant neighborhood, park, and participant level predictors [38]. The goal of this research was to estimate park use predictors across all neighborhood parks, so the marginal model may be preferable for interpretation purposes.
This study has important research implications. First, we carefully outlined modeling approaches for handling multi-level predictors for park use, which can be applied to a wide range of health behavior outcomes that are influenced by environmental/structural factors. Researchers can use similar procedures as those outlined in this paper to compare analytical approaches for complex datasets and conduct more appropriate statistical analyses. As methods for collecting data on multiple levels of influence become more advanced and multi-level interventions gain popularity, commentary on related analytic approaches is imperative [12,81–84]. Second, our study provides input on methods for multi-level data collection. Researchers should consider available tools for multi-level data collection as they plan their research and incorporate more distal influences of health behaviors (e.g., policies, neighborhoods), such as those outlined in social ecological models [11,12]. Third, our study used imputation methods to account for missing data on the individual level and minimize bias [85]. Future research can employ more rigorous techniques by incorporating follow up phone calls or engaging communities in data collection to establish commitment/trust [85–88]. Lastly, researchers can build on our findings by collecting multi-level data longitudinally to understand if there is a temporal relationship between identified predictors and park use. Longitudinal research can employ GEE weighting methods to account for missing data over time (i.e., drop outs, skipped time points) [89,90].
Findings from this research can also be used to ameliorate low park use and promote outdoor PA [91,92]. Specifically, determining neighborhood-, park-, and individual-level park use predictors in diverse geographic locations was a significant practice-based contribution of this study. In addition, methods used to collect multi-level data on park use predictors in this study can be employed by local public health practitioners to better understand community-level resource access. For instance, Maptionnaire.com surveys for individual-level, CPAT for park-level, and Census Bureau’s ACS data for neighborhood-level data were carefully chosen by our study team as accessible, high-quality options for multi-level measurement [15,18,60].
This research had several limitations. While this study offered granular-level data on park use patterns for adults in four geographically diverse U.S. cities, the survey sample was more affluent and had higher percent female than the target population and park users may be more interested and willing to answer a survey regarding local parks, which may have skewed our sample/estimates. Conversely, ~68% of participants in our study reported being park users, which closely matches past estimates of park use [93]. Next, some variables may be subject to over and under-estimation, such as self-reported PA. While this research employed a valid and reliable questionnaire [94], findings may reflect this potential bias. Finally, despite repeated recruitment efforts, the survey sample was smaller than anticipated with only about 300 participants and 3,000 potential park visits. Since our study used mail-based recruitment and had low response rates, future research should adopt novel approaches for community recruitment, such as social media-based recruitment which has showed promise for location-based sampling and recruiting during COVID-19 [95,96].
Conclusions
This study takes an empirical approach to addressing low levels of park use by comparing methods for analyzing multi-level predictors of park visitation, including both well-known conditional and more underutilized marginal approaches. Conditional and marginal modeling techniques identified similar park use predictors and showed similar generalizability, and conditional models showed superior predictive value. This is the first study to compare these popular analytical approaches to determine multi-level predictors of park use, and calls attention to underused marginal modeling approaches and its strengths when handling multi-level data. Future research can intentionally compare key features of conditional and marginal approaches for analyzing park use and other health behavior data, including their assumptions, interpretations, and specific applications outlined in this paper. Adding to this, researchers can employ cross validation techniques to identify instances where marginal models may display superior or comparable predictive performance.
Supporting information
S1 Checklist. Human participants research checklist.
https://doi.org/10.1371/journal.pone.0301549.s001
(DOCX)
S1 File. Sample code for conditional models, marginal models, and ten fold cross validation.
https://doi.org/10.1371/journal.pone.0301549.s002
(DOCX)
S2 File. ParkIndex dataset used for analyses.
https://doi.org/10.1371/journal.pone.0301549.s003
(XLSX)
S1 Fig. Odds ratios with 95% confidence limits for conditional and marginal models predicting park use.
https://doi.org/10.1371/journal.pone.0301549.s004
(TIF)
References
- 1. Moyle BD, Weiler B. Revisiting the importance of visitation: Public perceptions of park benefits. Tour Hosp Res. 2017 Jan 1;17(1):91–105.
- 2. Pitas NAD, Mowen AJ, Graefe AR, Godbey GC. Whose Park is it Anyway? Americans’ Access to, Use of, and Perceived Benefits from Local Park and Recreation Services in 1992 and 2015. J Park Recreat Adm [Internet]. 2019 [cited 2022 Mar 17]; Available from: https://js.sagamorepub.com/jpra/article/view/9681
- 3. Buckley R. Nature tourism and mental health: parks, happiness, and causation. J Sustain Tour. 2020 Sep 1;28(9):1409–24.
- 4. Orsega-Smith E, Mowen AJ, Payne LL, Godbey G. The Interaction of Stress and Park Use on Psycho-physiological Health in Older Adults. J Leis Res. 2004 Jun 1;36(2):232–56.
- 5. Morita E, Fukuda S, Nagano J, Hamajima N, Yamamoto H, Iwai Y, et al. Psychological effects of forest environments on healthy adults: Shinrin-yoku (forest-air bathing, walking) as a possible method of stress reduction. Public Health. 2007 Jan 1;121(1):54–63. pmid:17055544
- 6. Bedimo-Rung AL, Mowen AJ, Cohen DA. The significance of parks to physical activity and public health. Am J Prev Med [Internet]. 2005 Feb [cited 2019 Apr 5];28(2):159–68. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0749379704003046
- 7. Kaczynski AT, Henderson KA. Environmental Correlates of Physical Activity: A Review of Evidence about Parks and Recreation. Leis Sci [Internet]. 2007 Jul [cited 2019 Apr 5];29(4):315–54. Available from: http://www.tandfonline.com/doi/abs/10.1080/01490400701394865
- 8. Baran PK, Smith WR, Moore RC, Floyd MF, Bocarro JN, Cosco NG, et al. Park Use Among Youth and Adults: Examination of Individual, Social, and Urban Form Factors. Environ Behav [Internet]. 2014 Aug 1 [cited 2021 Jul 12];46(6):768–800. Available from: https://doi.org/10.1177/0013916512470134
- 9. Cohen DA, Marsh T, Williamson S, Derose KP, Martinez H, Setodji C, et al. Parks and physical activity: Why are some parks used more than others? Prev Med [Internet]. 2010 Jan [cited 2019 Apr 5];50:S9–12. Available from: https://linkinghub.elsevier.com/retrieve/pii/S009174350900485X pmid:19850067
- 10. McCormack GR, Rock M, Toohey AM, Hignell D. Characteristics of urban parks associated with park use and physical activity: A review of qualitative research. Health Place [Internet]. 2010 Jul [cited 2021 Jul 12];16(4):712–26. Available from: https://linkinghub.elsevier.com/retrieve/pii/S1353829210000316 pmid:20356780
- 11. Sallis JF, Cervero RB, Ascher W, Henderson KA, Kraft MK, Kerr J. An ecological approach to creating active living communities. Annu Rev Public Health [Internet]. 2006 Apr [cited 2019 Mar 13];27(1):297–322. Available from: http://www.annualreviews.org/doi/10.1146/annurev.publhealth.27.021405.102100 pmid:16533119
- 12.
Sallis JF, Owen N, Fisher EB. Ecological models of health behavior. In: Health behavior and health education: Theory, research, and practice, 4th ed. San Francisco, CA, US: Jossey-Bass; 2008. p. 465–85.
- 13. Padial-Ruz R, Puga-González ME, Céspedes-Jiménez Á, Cabello-Manrique D. Determining Factors in the Use of Urban Parks That Influence the Practice of Physical Activity in Children: A Systematic Review. Int J Environ Res Public Health [Internet]. 2021 Mar 31 [cited 2022 Mar 18];18(7):3648. Available from: https://www.mdpi.com/1660-4601/18/7/3648 pmid:33807466
- 14. Zhang R, Wulff H, Duan Y, Wagner P. Associations between the physical environment and park-based physical activity: A systematic review. J Sport Health Sci [Internet]. 2018 Nov [cited 2019 Mar 14]; Available from: https://linkinghub.elsevier.com/retrieve/pii/S209525461830098X pmid:31534816
- 15. Besenyi GM, Schooley B, Turner-McGrievy GM, Wilcox S, Wilhelm Stanis SA, Kaczynski AT. The Electronic Community Park Audit Tool (eCPAT): Exploring the Use of Mobile Technology for Youth Empowerment and Advocacy for Healthy Community Policy, Systems, and Environmental Change. Front Public Health [Internet]. 2018 Nov 20 [cited 2020 Jan 16];6:332. Available from: https://www.frontiersin.org/article/10.3389/fpubh.2018.00332/full
- 16. Evenson KR, Jones SA, Holliday KM, Cohen DA, McKenzie TL. Park characteristics, use, and physical activity: A review of studies using SOPARC (System for Observing Play and Recreation in Communities). Prev Med [Internet]. 2016 May [cited 2019 Apr 5];86:153–66. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743516300081 pmid:26946365
- 17. Glennon R. Measuring access to parks [Internet]. ArcGIS Blog. 2011 [cited 2022 Mar 18]. Available from: https://www.esri.com/arcgis-blog/products/arcgis-desktop/analytics/measuring-access-to-parks/
- 18.
United States Census Bureau. The United States Census Bureau. [cited 2020 Nov 5]. American Community Survey Data. Available from: https://www.census.gov/programs-surveys/acs/data.html
- 19. Joseph RP, Maddock JE. Comparative Analysis of Five Observational Audit Tools to Assess the Physical Environment of Parks for Physical Activity, 2016. Prev Chronic Dis. 2016 Dec 15;13:E166. pmid:27978411
- 20. Besenyi GM, Diehl P, Schooley B, Turner-McGrievy BM, Wilcox S, Stanis SAW, et al. Development and testing of mobile technology for community park improvements: validity and reliability of the eCPAT application with youth. Transl Behav Med [Internet]. 2016 Dec [cited 2020 Jan 15];6(4):519–32. Available from: https://academic.oup.com/tbm/article/6/4/519-532/4563203 pmid:27146276
- 21. Kaczynski AT, Hughey SM, Stowe EW, Wende ME, Hipp JA, Oliphant EL, et al. ParkIndex: Validation and application of a pragmatic measure of park access and use. Preventive Medicine Reports. 2020;20:101218. pmid:33354490
- 22. Kaczynski AT, Schipperijn J, Hipp JA, Besenyi GM, Wilhelm Stanis SA, Hughey SM, et al. ParkIndex: Development of a standardized metric of park access for research and planning. Prev Med [Internet]. 2016 Jun [cited 2020 Jan 16];87:110–4. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743516000517 pmid:26868092
- 23. Hughey SM, Stowe EW, Trello S, Bornstein D, Brown K, Davis WJ, et al. A Multimethod Study of Patterns and Motivations of Greenway-Based Physical Activity. Transl J Am Coll Sports Med [Internet]. 2021;6(1). Available from: https://journals.lww.com/acsm-tj/Fulltext/2021/01150/A_Multimethod_Study_of_Patterns_and_Motivations_of.1.aspx
- 24. Schneider RJ. Measuring transportation at a human scale: An intercept survey approach to capture pedestrian activity. J Transp Land Use [Internet]. 2013 Nov 14 [cited 2023 Feb 23];6(3):43. Available from: https://www.jtlu.org/index.php/jtlu/article/view/311
- 25. Liu H, Li F, Li J, Zhang Y. The relationships between urban parks, residents’ physical activity, and mental health benefits: A case study from Beijing, China. J Environ Manage [Internet]. 2017 Apr 1 [cited 2021 Sep 10];190:223–30. Available from: https://www.sciencedirect.com/science/article/pii/S0301479716310416 pmid:28056355
- 26. Evenson KR, Herring AH, Huston SL. Evaluating change in physical activity with the building of a multi-use trail. Am J Prev Med [Internet]. 2005 Feb 1 [cited 2021 Sep 10];28(2, Supplement 2):177–85. Available from: https://www.sciencedirect.com/science/article/pii/S0749379704003009 pmid:15694526
- 27. Fitzhugh EC, Bassett DR, Evans MF. Urban Trails and Physical Activity. Am J Prev Med [Internet]. 2010 Sep [cited 2020 Dec 28];39(3):259–62. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0749379710003612
- 28. Dunton GF, Almanza E, Jerrett M, Wolch J, Pentz MA. Neighborhood Park Use by Children. Am J Prev Med [Internet]. 2014 Feb [cited 2020 Oct 20];46(2):136–42. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0749379713005825
- 29. Prince SA, Butler GP, Rao DP, Thompson W. Evidence synthesis Where are children and adults physically active and sedentary?–a rapid review of location-based studies. Health Promot Chronic Dis Prev Can Res Policy Pract [Internet]. 2019 Mar [cited 2021 Jul 22];39(3):67–103. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6478053/
- 30. Hallo JC, Beeco JA, Goetcheus C, McGee J, McGehee NG, Norman WC. GPS as a Method for Assessing Spatial and Temporal Use Distributions of Nature-Based Tourists. J Travel Res [Internet]. 2012 Sep [cited 2021 Jul 22];51(5):591–606. Available from: http://journals.sagepub.com/doi/
- 31. Evenson KR, Wen F, Hillier A, Cohen DA. Assessing the Contribution of Parks to Physical Activity Using Global Positioning System and Accelerometry. Med Sci Sports Exerc [Internet]. 2013 Oct [cited 2020 Oct 20];45(10):1981–7. Available from: http://journals.lww.com/00005768-201310000-00018 pmid:23531716
- 32. Zhai Y, Baran PK, Wu C. Spatial distributions and use patterns of user groups in urban forest parks: An examination utilizing GPS tracker. Urban For Urban Green [Internet]. 2018 Oct 1 [cited 2021 Jul 22];35:32–44. Available from: https://www.sciencedirect.com/science/article/pii/S1618866717307501
- 33. Zhai Y, Li D, Wu C, Wu H. Urban park facility use and intensity of seniors’ physical activity–An examination combining accelerometer and GPS tracking. Landsc Urban Plan [Internet]. 2021 Jan 1 [cited 2021 Jul 22];205:103950. Available from: https://www.sciencedirect.com/science/article/pii/S0169204619315385
- 34.
Xu W. Urban Explorations: Analysis of Public Park Usage using Mobile GPS Data. ArXiv180101921 Cs [Internet]. 2018 Jan 5 [cited 2021 Jul 22]; Available from: http://arxiv.org/abs/1801.01921
- 35. McCrorie PR, Fenton C, Ellaway A. Combining GPS, GIS, and accelerometry to explore the physical activity and environment relationship in children and young people—a review. Int J Behav Nutr Phys Act [Internet]. 2014 Dec [cited 2019 Apr 21];11(1):93. Available from: http://ijbnpa.biomedcentral.com/articles/10.1186/s12966-014-0093-0
- 36. Zandieh R, Martinez J, Flacke J. Older Adults’ Outdoor Walking and Inequalities in Neighbourhood Green Spaces Characteristics. Int J Environ Res Public Health [Internet]. 2019 Nov [cited 2021 Apr 14];16(22). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6888485/ pmid:31717514
- 37. Lee Y, Nelder JA. Conditional and Marginal Models: Another View. Stat Sci [Internet]. 2004 May 1 [cited 2021 Jul 12];19(2). Available from: https://projecteuclid.org/journals/statistical-science/volume-19/issue-2/Conditional-and-Marginal-Models-Another-View/10.1214/088342304000000305.full
- 38. Hubbard AE, Ahern J, Fleischer NL, Van der Laan M, Lippman SA, Jewell N, et al. To GEE or not to GEE: comparing population average and mixed models for estimating the associations between neighborhood risk factors and health. Epidemiol Camb Mass. 2010 Jul;21(4):467–74. pmid:20220526
- 39. Heagerty PJ, Zeger SL. Marginalized Multilevel Models and Likelihood Inference. Stat Sci [Internet]. 2000 [cited 2021 Oct 7];15(1):1–19. Available from: https://www.jstor.org/stable/2676670
- 40. Cox DR. Role of Models in Statistical Analysis. Stat Sci [Internet]. 1990 [cited 2021 Oct 7];5(2):169–74. Available from: https://www.jstor.org/stable/2245676
- 41. Sato M, Inoue Y, Du J, Funk DC. Access to parks and recreational facilities, physical activity, and health care costs for older adults: Evidence from U.S. counties. J Leis Res [Internet]. 2019 May 27 [cited 2020 Dec 7];50(3):220–38. Available from: https://www.tandfonline.com/doi/full/
- 42. Loukaitou-Sideris A, Sideris A. What Brings Children to the Park? Analysis and Measurement of the Variables Affecting Children’s Use of Parks. J Am Plann Assoc [Internet]. 2009 Dec 31 [cited 2020 Dec 8];76(1):89–107. Available from: http://www.tandfonline.com/doi/abs/10.1080/01944360903418338
- 43. Mowen A, Orsega-Smith E, Payne L, Ainsworth B, Godbey G. The role of park proximity and social support in shaping park visitation, physical activity, and perceived health among older adults. J Phys Act Health. 2007 Apr;4(2):167–79. pmid:17570886
- 44. Knapp M, Gustat J, Darensbourg R, Myers L, Johnson C. The Relationships between Park Quality, Park Usage, and Levels of Physical Activity in Low-Income, African American Neighborhoods. Int J Environ Res Public Health [Internet]. 2019 Jan [cited 2021 Jul 12];16(1):85. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6338998/
- 45. Schipperijn J, Cerin E, Adams MA, Reis R, Smith G, Cain K, et al. Access to parks and physical activity: An eight country comparison. Urban For Urban Green [Internet]. 2017 Oct [cited 2019 Jun 13];27:253–63. Available from: https://linkinghub.elsevier.com/retrieve/pii/S1618866717302571 pmid:29805351
- 46. Kaczynski AT, Besenyi GM, Stanis SAW, Koohsari MJ, Oestman KB, Bergstrom R, et al. Are park proximity and park features related to park use and park-based physical activity among adults? Variations by multiple socio-demographic characteristics. Int J Behav Nutr Phys Act [Internet]. 2014 Dec [cited 2019 Apr 5];11(1). Available from: http://ijbnpa.biomedcentral.com/articles/10.1186/s12966-014-0146-4 pmid:25480157
- 47. Huang JH, Hipp JA, Marquet O, Alberico C, Fry D, Mazak E, et al. Neighborhood characteristics associated with park use and park-based physical activity among children in low-income diverse neighborhoods in New York City. Prev Med [Internet]. 2020 Feb [cited 2020 Nov 28];131:105948. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743519304311 pmid:31836479
- 48. Moran MR, Rodríguez DA, Cotinez-O’Ryan A, Miranda JJ. Park use, perceived park proximity, and neighborhood characteristics: Evidence from 11 cities in Latin America. Cities [Internet]. 2020 Oct [cited 2020 Dec 7];105:102817. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0264275119314064 pmid:33012941
- 49. Kaczynski AT, Koohsari MJ, Stanis SAW, Bergstrom R, Sugiyama T. Association of Street Connectivity and Road Traffic Speed with Park Usage and Park-Based Physical Activity. Am J Health Promot [Internet]. 2014 Jan [cited 2019 Apr 5];28(3):197–203. Available from: http://journals.sagepub.com/doi/10.4278/ajhp.120711-QUAN-339 pmid:23875985
- 50. Ries AV, Voorhees CC, Roche KM, Gittelsohn J, Yan AF, Astone NM. A Quantitative Examination of Park Characteristics Related to Park Use and Physical Activity Among Urban Youth. J Adolesc Health [Internet]. 2009 Sep [cited 2021 Jul 12];45(3):S64–70. Available from: https://linkinghub.elsevier.com/retrieve/pii/S1054139X09001864 pmid:19699439
- 51. Rigolon A, Browning M, Jennings V. Inequities in the quality of urban park systems: An environmental justice investigation of cities in the United States. Landsc Urban Plan [Internet]. 2018 Oct [cited 2020 Sep 8];178:156–69. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0169204618304316
- 52. Han B, Cohen DA, Derose KP, Li J, Williamson S. Violent Crime and Park Use in Low-Income Urban Neighborhoods. Am J Prev Med [Internet]. 2018 Mar 1 [cited 2021 Jul 12];54(3):352–8. Available from: https://www.sciencedirect.com/science/article/pii/S0749379717306487 pmid:29338953
- 53. Stewart OT, Moudon AV, Littman AJ, Seto E, Saelens BE. The Association Between Park Facilities and Duration of Physical Activity During Active Park Visits. J Urban Health [Internet]. 2018 Dec [cited 2021 Jul 12];95(6):869–80. Available from: http://link.springer.com/10.1007/s11524-018-0311-1 pmid:30232689
- 54. Oliphant EL, Hughey M, Stowe EW, Kaczynski AT, Schipperijn J, Hipp JA. ParkIndex: Using Key Informant Interviews to Inform the Development of a New Park Access Evaluation Tool. J Park Recreat Adm [Internet]. 2019 [cited 2020 Jan 16];37(1). Available from: https://js.sagamorepub.com/jpra/article/view/8926 pmid:31692932
- 55. Kaczynski AT, Wilhelm Stanis SA, Besenyi GM. Development and Testing of a Community Stakeholder Park Audit Tool. Am J Prev Med [Internet]. 2012 Mar [cited 2019 Apr 5];42(3):242–9. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0749379711009032 pmid:22341161
- 56. Vaughan KB, Kaczynski AT, Wilhelm Stanis SA, Besenyi GM, Bergstrom R, Heinrich KM. Exploring the Distribution of Park Availability, Features, and Quality Across Kansas City, Missouri by Income and Race/Ethnicity: an Environmental Justice Investigation. Ann Behav Med [Internet]. 2013 Feb [cited 2019 Jun 26];45(S1):28–38. Available from: https://academic.oup.com/abm/article/45/suppl_1/S28-S38/4563960
- 57. Hughey S, Kaczynski AT, Child S, Moore JB, Porter D, Hibbert J. Green and lean: Is neighborhood park and playground availability associated with youth obesity? Variations by gender, socioeconomic status, and race/ethnicity. Prev Med [Internet]. 2017 Feb [cited 2019 Jun 26];95:S101–8. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743516303802 pmid:27932053
- 58. Parsons AA, Besenyi GM, Kaczynski AT, Wilhelm Stanis SA, Blake CE, Barr-Anderson DJ. Investigating Issues of Environmental Injustice in Neighborhoods Surrounding Parks. J Leis Res [Internet]. 2015 Apr [cited 2019 Jun 26];47(2):285–303. Available from: https://www.tandfonline.com/doi/full/10.1080/00222216.2015.11950361
- 59. Hughey SM, Walsemann KM, Child S, Powers A, Reed JA, Kaczynski AT. Using an environmental justice approach to examine the relationships between park availability and quality indicators, neighborhood disadvantage, and racial/ethnic composition. Landsc Urban Plan [Internet]. 2016 Apr [cited 2019 Mar 31];148:159–69. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0169204615002583
- 60.
Maptionnaire [Internet]. [cited 2022 Jan 4]. Maptionnaire | Community Engagement Platform. Available from: https://maptionnaire.com
- 61.
United States Environmental Protection Agency. Smart Growth. [cited 2021 Aug 15]. Smart Location Mapping: National Walkability Index. Available from: https://www.epa.gov/smartgrowth/smart-location-mapping#walkability
- 62. Craig CL, Marshall AL, Sjöström M, Bauman AE, Booth ML, Ainsworth BE, et al. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc. 2003 Aug;35(8):1381–95. pmid:12900694
- 63.
2018 Physical Activity Guidelines Advisory Committee. 2018 Physical Activity Guidelines Advisory Committee Scientific Report. Washington, DC: U.S. Department of Health and Human Services; 2018.
- 64. Andridge RR, Little RJA. A Review of Hot Deck Imputation for Survey Non-response. Int Stat Rev Rev Int Stat. 2010 Apr;78(1):40–64.
- 65. SAS. SAS. Overview: Proc Glimmix Procedure Assumptions. Available from: https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_glimmix_a0000001396.htm
- 66. Merlo J, Chaix B, Ohlsson H, Beckman A, Johnell K, Hjerpe P, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health. 2006 Apr;60(4):290–7. pmid:16537344
- 67. Larsen K, Merlo J. Appropriate assessment of neighborhood effects on individual health: integrating random and fixed effects in multilevel logistic regression. Am J Epidemiol. 2005 Jan 1;161(1):81–8. pmid:15615918
- 68. Larsen K, Petersen JH, Budtz-Jørgensen E, Endahl L. Interpreting Parameters in the Logistic Regression Model with Random Effects. Biometrics [Internet]. 2000 [cited 2023 Apr 20];56(3):909–14. Available from: https://www.jstor.org/stable/2676942 pmid:10985236
- 69. Jung Y, Hu J. A K-fold Averaging Cross-validation Procedure. J Nonparametric Stat [Internet]. 2015 [cited 2021 Nov 5];27(2):167–79. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5019184/ pmid:27630515
- 70. Richardson AS, Ghosh-Dastidar M, Collins RL, Hunter GP, Troxel WM, Colabianchi N, et al. Improved Street Walkability, Incivilities, and Esthetics Are Associated with Greater Park Use in Two Low-Income Neighborhoods. J Urban Health [Internet]. 2020 Apr 1 [cited 2022 Aug 17];97(2):204–12. Available from: pmid:31989419
- 71. Schultz CL, Wilhelm Stanis SA, Sayers SP, Thombs LA, Thomas IM. A longitudinal examination of improved access on park use and physical activity in a low-income and majority African American neighborhood park. Prev Med [Internet]. 2017 Feb [cited 2019 Apr 5];95:S95–100. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743516302420 pmid:27568236
- 72. Parks Rigolon A. and young people: An environmental justice study of park proximity, acreage, and quality in Denver, Colorado. Landsc Urban Plan [Internet]. 2017 Sep [cited 2021 Jul 12];165:73–83. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0169204617301111
- 73. Cohen DA, Han B, Derose KP, Williamson S, Marsh T, Raaen L, et al. The Paradox of Parks in Low-Income Areas: Park Use and Perceived Threats. Environ Behav [Internet]. 2016 Jan [cited 2020 Jan 15];48(1):230–45. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4821183/ pmid:27065480
- 74. Floyd MF, Spengler JO, Maddock JE, Gobster PH, Suau LJ. Park-Based Physical Activity in Diverse Communities of Two U.S. Cities. Am J Prev Med [Internet]. 2008 Apr [cited 2019 Nov 20];34(4):299–305. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0749379708000615
- 75.
Stacy CP, Su Y, Noble E, Stern A, Blagg K, Rainer M, et al. Access to Opportunity through Equitable Transportation [Internet]. Urban Institute; 2020 Oct [cited 2022 Jan 12]. Available from: https://www.urban.org/research/publication/access-opportunity-through-equitable-transportation
- 76. McKenzie BS. Neighborhood Access to Transit by Race, Ethnicity, and Poverty in Portland, OR. City Community [Internet]. 2013 Jun 1 [cited 2022 Jan 12];12(2):134–55. Available from: https://doi.org/10.1111/cico.12022
- 77. Sanchez TW. Equity Analysis of Personal Transportation System Benefits. J Urban Aff [Internet]. 1998 Mar 1 [cited 2022 Jan 12];20(1):69–86. Available from: https://doi.org/10.1111/j.1467-9906.1998.tb00411.x
- 78. Cools M, Moons E. Handling Intrahousehold Correlations in Modeling Travel: Comparison of Hierarchical Models and Marginal Models. Transp Res Rec J Transp Res Board [Internet]. 2016 Jan [cited 2022 Jan 4];2565(1):8–17. Available from: http://journals.sagepub.com/doi/10.3141/2565-02
- 79. Carrière I, Bouyer J. Choosing marginal or random-effects models for longitudinal binary responses: application to self-reported disability among older persons. BMC Med Res Methodol [Internet]. 2002 Dec 5 [cited 2022 Jan 4];2(1):15. Available from: pmid:12466027
- 80. Koper N, Manseau M. Generalized estimating equations and generalized linear mixed-effects models for modelling resource selection. J Appl Ecol [Internet]. 2009 [cited 2022 Jan 4];46(3):590–9. Available from: https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1365-2664.2009.01642.x
- 81. Diez-Roux AV. Multilevel Analysis in Public Health Research. Annu Rev Public Health [Internet]. 2000 [cited 2021 May 17];21(1):171–92. Available from: pmid:10884951
- 82. Agurs-Collins T, Persky S, Paskett ED, Barkin SL, Meissner HI, Nansel TR, et al. Designing and Assessing Multilevel Interventions to Improve Minority Health and Reduce Health Disparities. Am J Public Health [Internet]. 2019 Jan 1 [cited 2022 Jan 4];109(S1):S86–93. Available from: https://ajph.aphapublications.org/doi/10.2105/AJPH.2018.304730 pmid:30699029
- 83. Von Korff M, Koepsell T, Curry S, Diehr P. Multi-level analysis in epidemiologic research on health behaviors and outcomes. Am J Epidemiol. 1992 May 15;135(10):1077–82. pmid:1632420
- 84. Compernolle S, De Cocker K, Lakerveld J, Mackenbach JD, Nijpels G, Oppert JM, et al. A RE-AIM evaluation of evidence-based multi-level interventions to improve obesity-related behaviours in adults: a systematic review (the SPOTLIGHT project). Int J Behav Nutr Phys Act [Internet]. 2014 Dec 6 [cited 2022 Jan 4];11(1):147. Available from: pmid:25480391
- 85. Kang H. The prevention and handling of the missing data. Korean J Anesthesiol [Internet]. 2013 May [cited 2022 Jan 4];64(5):402–6. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3668100/ pmid:23741561
- 86. Claassen C, Kurian B, Trivedi MH, Grannemann BD, Tuli E, Pipes R, et al. Telephone-based Assessments to Minimize Missing Data in Longitudinal Depression Trials: A Project IMPACTS Study Report. Contemp Clin Trials [Internet]. 2009 Jan [cited 2022 Jan 4];30(1):13–9. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3268227/ pmid:18761427
- 87. Botchwey N, Conway TL, Floyd M, Hipp A, Kim A, Pollack Porter KM, et al. Challenges recruiting diverse youth for physical activity research. Prev Med [Internet]. 2020 Feb [cited 2022 Jan 10];131:105888. Available from: https://linkinghub.elsevier.com/retrieve/pii/S0091743519303688 pmid:31770541
- 88.
Sanders Thompson V, Hood S. Community-based Participatory Research. In: Public Health Research Methods for Partnerships and Practice. New York, NY: Routledge; 2018. p. 1–22.
- 89. Salazar A, Ojeda B, Dueñas M, Fernández F, Failde I. Simple generalized estimating equations (GEEs) and weighted generalized estimating equations (WGEEs) in longitudinal studies with dropouts: guidelines and implementation in R. Stat Med [Internet]. 2016 [cited 2022 Jan 4];35(19):3424–48. Available from: https://onlinelibrary.wiley.com/doi/abs/10.1002/sim.6947 pmid:27059703
- 90. Lin G, Rodriguez RN. Weighted Methods for Analyzing Missing Data with the GEE Procedure.: 8.
- 91. Volenec ZM, Abraham JO, Becker AD, Dobson AP. Public parks and the pandemic: How park usage has been affected by COVID-19 policies. PLOS ONE [Internet]. 2021 May 19 [cited 2021 Aug 13];16(5):e0251799. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0251799 pmid:34010353
- 92. Spencer LH, Lynch M, Lawrence CL, Edwards RT. A Scoping Review of How Income Affects Accessing Local Green Space to Engage in Outdoor Physical Activity to Improve Well-Being: Implications for Post-COVID-19. Int J Environ Res Public Health [Internet]. 2020 Dec 12 [cited 2021 Jul 12];17(24):9313. Available from: https://www.mdpi.com/1660-4601/17/24/9313 pmid:33322829
- 93.
National Recreation and Park Association. 2018 Americans’ engagement with parks report [Internet]. 2018. Available from: https://www.nrpa.org/globalassets/engagement-survey-report-2018.pdf
- 94. Lee PH, Macfarlane DJ, Lam T, Stewart SM. Validity of the international physical activity questionnaire short form (IPAQ-SF): A systematic review. Int J Behav Nutr Phys Act [Internet]. 2011 [cited 2020 Nov 5];8(1):115. Available from: http://ijbnpa.biomedcentral.com/articles/10.1186/1479-5868-8-115 pmid:22018588
- 95. Archer-Kuhn B, Beltrano NR, Hughes J, Saini M, Tam D. Recruitment in response to a pandemic: pivoting a community-based recruitment strategy to facebook for hard-to-reach populations during COVID-19. Int J Soc Res Methodol [Internet]. 2021 Jun 16 [cited 2022 Jan 4];0(0):1–12. Available from: https://doi.org/10.1080/13645579.2021.1941647
- 96. Pechmann C, Phillips C, Calder D, Prochaska JJ. Facebook Recruitment Using Zip Codes to Improve Diversity in Health Research: Longitudinal Observational Study. J Med Internet Res [Internet]. 2020 Jun 5 [cited 2022 Jan 4];22(6):e17554. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7305557/ pmid:32501274