Walkability around the worksite and self-reported and accelerometer-measured physical activity among adults

Abstract

This cross-sectional study investigated the relationship between GIS-measured worksite and home neighborhood walkability and several measures of physical activity (PA) in employed adults. Results revealed no significant correlation between worksite walkability and PA outcomes, contradicting the hypothesis of increased PA with improved walkability. However, for women and households without young children, a positive association was observed between worksite walkability and moderate-to-vigorous physical activity (MVPA). Additionally, home neighborhood walkability was linked to self-reported walking. The study highlights the need for further research into social and environmental factors at worksites impacting PA, and examination of PA behaviors in the context of increased remote work due to the COVID-19 pandemic.

Introduction

Physical activity (PA) is a potent, protective health behavior associated with many positive health outcomes, including decreased risk of cardiovascular disease, diabetes, obesity, depression, and several cancers (Moore et al., 2016; Powell et al., 2018). This relationship persists across age, gender, race, independent of genetic risk factors (Kokkinos, 2012; Kokkinos and Myers, 2010). Individuals who meet the federally recommended 150 minutes of aerobic moderate-to-vigorous-intensity PA (MVPA) per week have a 33% lower risk of all-cause mortality (Powell et al., 2018); however, in 2020, only 46.9% of adults in the United States met this recommendation based on self-report (Elgaddal et al., 2022).

Given the prevalence of inactivity and its associated health risks, there is a clear need for research that hones in on the individual and contextual factors that influence PA behaviors. One such contextual factor from ecological models is the built environment (BE): the collection of built urban features, such as buildings, streets, public spaces, and walkways that people utilize for multiple purposes, including recreational and transportation PA (Bauman et al., 2012; Frank et al., 2003; Saelens and Handy, 2008). The built environment around worksites may influence physical activity through activity-supportive features such as access to a variety of destinations, connected street networks, access to public transportation, parks, and recreational facilities, pedestrian and bike-friendly infrastructure, and safety from traffic and crime (Young et al., 2020). There is consistent evidence that one’s neighborhood BE influences PA behaviors: more walkable home neighborhoods are associated with higher levels of PA (McCormack and Shiell, 2011), and this relationship has been found across the world (Sallis et al., 2020).

While many studies have analyzed the effects of BE features around individuals’ homes, less research has been conducted on the influence surrounding the workplace (Lin et al., 2020). According to the Bureau of Labor Statistics’ American Time Use Survey, 82% of employed Americans worked at a location other than their home in 2019, for 7.9 hours per workday (Statistics, 2020). Although the number of people working at home has increased during the COVID-19 pandemic (Statistics, 2023), many people spend long periods at work outside of the home, making the worksite neighborhood a potentially significant environment that may allow for a more comprehensive understanding of the influence of the BE on PA behaviors.

Relatively fewer studies have investigated workplace BE and PA, with most having important methodological limitations. Previous research has relied solely on self-reported measures of both neighborhood walkability and PA (Adlakha et al., 2015; Badland et al., 2010; Prodaniuk et al., 2004). Self-reported data is prone to biases and errors (Barrington et al.,2015), especially single-source errors (Campbell and Fiske, 1959). Lin et al.’s ((Lin et al., 2020) review highlighted that many studies lacked precision in defining worksite neighborhood boundaries and did not measure “where” participants performed physical activities. While not exclusively related to worksite walkability, Ding and Gebel (Ding and Gebel, 2012) also identified similar issues and emphasized the need to investigate potential moderators of the built environment and physical activity relationship. Additional research using objective PA and worksite neighborhood walkability measures is necessary to inform the extent to which the workplace neighborhood could be impacting PA behaviors.

Accordingly, the current study used a combination of self-reported PA engaged in both within and outside of participants’ home neighborhood, objectively-measured PA from accelerometers, and objectively-measured worksite walkability using geographical information systems (GIS) data. Through a cross-sectional study design using data collected at baseline from employed, healthy, insufficiently active adults recruited for a randomized controlled trial focused on improving MVPA, this study examined the relationship between worksite neighborhood walkability and PA. We hypothesized that worksite neighborhood walkability would explain additional variance in and positively covary with accelerometer-measured PA, total self-reported walking and bicycling, and self-reported transportation bicycling, after adjusting for potential confounders, including home neighborhood walkability. Furthermore, we hypothesized that worksite walkability would explain variation in self-reported PA that occurs outside the home neighborhood but not inside the home neighborhood.

A second aim was to examine whether individual-level characteristics such as sex, age, and the presence of children moderated the relationship between worksite walkability and PA. We hypothesized that these individual-level characteristics would not moderate the relationship between worksite walkability and self-reported PA inside the home neighborhood. Instead, they would moderate the relationship between worksite neighborhood walkability and self-reported PA outside the home neighborhood, and total self-reported and accelerometer-measured PA outcomes. We first report on associations between worksite walkability and accelerometer-measured and reported physical activity and then on tests for moderation of these relationships.

Methods

Study Design:

The current study was a secondary analysis of the baseline data gathered during the Walking Intervention Through Texting (WalkIT) Arizona Study; a randomized controlled trial focused on increasing PA behaviors in healthy, insufficiently active adults, ages 18-60, in Maricopa County, Arizona (Author et al, 2019). The analytic sample for this paper comprised participants who provided baseline data, including those not randomized due to being too active (≥150 minutes per week). One of the WalkIT AZ Study aims was to investigate the influence of neighborhood walkability and socioeconomic status (SES) on PA behaviors.

Participants were sampled from census block groups (BGs) that varied in SES and walkability. BGs were rank-ordered separately for SES and walkability and categorized into deciles. Only BGs in the 1st-4th and 7th-10th deciles for each feature were included to maximize the variation of these neighborhood features and avoid misclassification. Participants were sampled from eligible neighborhoods to study SES based on median household income collected from the 2010 Census American Community Survey and categorized as either “Low SES” (1st-5th deciles) or “High SES” (7th-10th deciles). Home neighborhood walkability was measured using BG-level geographical information systems (GIS) data of neighborhood features, categorized as “Low walkable” (lst-4th deciles) or “High walkable” (7th-10th deciles) neighborhoods (Author et al, 2019).

Participants were recruited from 2016 to 2018, with a balanced number of participants recruited across calendar months to control for the effect of variability in temperatures and extreme summer heat in Maricopa County, AZ. The study started with a baseline phase that included a survey to collect self-reported PA behaviors from the last week and a prospective 2-week baseline period of accelerometer-measured PA data collection. Accelerometer data were collected to determine PA behaviors prior to the start of the WalkIT AZ intervention and were used in the current analysis. The detailed study design and methods of the entire WalkIT AZ Study have been previously described (Author et al, 2019).

Objectively-measured Walkability:

For this secondary analysis, individual-level home and worksite neighborhood walkability was determined for a 500-meter street-network buffer around the geocoded home and work addresses or the nearest cross street for those who did not provide a precise worksite address. For both the home neighborhood and workplace neighborhood, an adapted Walkability Index value was computed based on three original components used by Frank et al. (2010): net residential density (the ratio of residential units to the overall area allocated for residential land in the BG), intersection density (the ratio of intersections with at least 3 legs to the total land area in the BG), and land-use mix (degree of diversity in land-use types, including residential, retail, entertainment, office, and institutional--scored as 0 to represent single use up to 1 to indicate an even distribution of use), plus an additional component—transit density (the ratio of transit stops to the total land area in the BG) (Frank et al., 2010). Following Frank et al., z-scores were used to standardize component scores and an individual-level walkability Index was calculated using the following formula: Walkability Index= [(z-net residential density) + (z-intersection density) + (z-land use mix) + (z-transit density)].

Baseline Questionnaires: Overview:

After providing informed consent, participants answered health screening questions to ensure they did not have any health conditions that would prevent them from being physically active. They also completed a baseline survey, which included questions about demographics, PA behaviors, work type and locations, and home neighborhood environment.

Self-reported PA:

Two PA questionnaires were used in WalkIT Arizona because this study aimed to leverage data collected by harmonizing measures with another study of the built environment (Patch et al., 2019). Participants answered questions from the International Physical Activity Questionnaire (IPAQ-Long Form) about the frequency, duration, and purpose (transportation vs. recreation) of walking and bicycling performed in the past week (Craig et al., 2003). The number of minutes per week (min·wk⁻¹) of total walking activity (transportation and recreation combined) and transportation bicycling, respectively, were calculated from reported days per week multiplied by the reported duration per day. Truncation rules were used with a maximum of 180 minutes per day for each activity type and purpose to avoid PA duration overreporting (IPAQ, 2005). Additionally, participants answered questions from the Neighborhood Physical Activity Questionnaire (NPAQ) about the frequency, duration, and purpose (transportation vs. recreation) of walking and bicycling behaviors inside and outside the home neighborhood during a typical week (Giles-Corti et al., 2006). From these questions, we computed the reported number of minutes per week (i.e., number of days per week multiplied by duration per day) of (a) total walking activity (combined transportation and recreation for both inside and outside the home neighborhood), (b) total bicycling activity (same combination as walking activity), (c) walking inside the home neighborhood, and (d) walking outside the home neighborhood. The reported activity was truncated at 840 minutes per week for each activity type, as done in the Active Australia Survey ((AIHW), 2003).

Of the 728 consented adults who completed the baseline survey, 103 were excluded because they did not report full- or part-time employment. A further 113 participants were excluded because they worked from home (n=61), their work address or nearest cross-street was missing (n=36), they resided at their workplace (n=9), they worked outside of Maricopa County (n=6), or they did not report their household income (n=1). The remaining 512 participants contributed survey data for the self-reported analyses in this study (coincidently independent of the n=512 randomized sample from the WalkIT AZ Study). Of the 512 participants who contributed self-reported data, 40 had missing accelerometer data, leaving 472 participants for analyses of accelerometer-measured PA.

Objectively-Measured PA:

After taking the Baseline Survey, participants wore a wrist-worn ActiGraph GT9X Link accelerometer for at least 10 hours each day for 10 days to two weeks. They were instructed to continue their routine behaviors to measure their baseline PA behaviors and verify WalkIT AZ intervention eligibility. Using the ActiGraph GT9X Link, we measured vector magnitude (VM) counts and moderate-to-vigorous PA (MVPA) in 1-minute epochs. VM counts were derived through the ActiGraph’s measurements of acceleration and deceleration of movements in the vertical, antero-posterio, and medio-lateral axes. VM counts per minute were calculated using the formula: $VM=\sqrt{{(Axis1)}^2+{(Axis2)}^2+{(Axis3)}^2}$ (Development, 2012).

A treadmill walking test was performed during the study’s Baseline to determine the ActiGraph GT9x counts identifying each participant’s walking cutpoint at a moderate-to-vigorous intensity. The treadmill protocol started with a 5-minute standing rest period, followed by three stages of 6-minute bouts of walking at 2.0 mph, 3.0 mph, and 4.0 mph, all at a 0% grade (Author et al, 2019). Participants wore an ActiGraph GT9X, and their breath-by-breath oxygen uptake (i.e., VO2 in milliliters per kilogram per minute (ml·kg⁻¹·min⁻¹)) was analyzed by the CareFusion System’s Oxycon Mobile portable metabolic system. This process enabled the research team to identify the participant-specific threshold of ActiGraph GT9X VM counts per minute for walking at a moderate-intensity level (energy cost >3.0 METs, where 3.5 ml·kg⁻¹·min⁻¹ is a standard MET). Following Barnett et al., a quadratic regression model established the VM cut-point for each participant at MVPA (Barnett et al., 2015). For MVPA minutes to be counted, participants had to walk for at least 3 consecutive minutes at or above the VM cut-point. This approach allowed for the remote intervention to send daily information back to the researchers via mobile phone and was decided upon after consultation with ActiGraph and measurement limitations at the time.

Statistical Methods:

Analyses were conducted using SPSS 25. The PA outcome variables had non-normal distributions with abundant zeros. Accordingly, generalized linear models (GZLMs) with a negative binomial distribution, a log-link, and a maximum likelihood-estimated dispersion parameter (α) were used to address study aims for both self-reported and objective measures of PA. For the first aim of the study, the relationships between worksite walkability and eight PA outcomes were examined through GZLMs controlled for background covariates and home neighborhood walkability. The 8 PA outcomes investigated were: IPAQ Total Walking min·wk⁻¹, IPAQ Transportation Bicycling min·wk⁻¹, NPAQ Total Walking min·wk⁻¹, NPAQ Walking Inside Home Neighborhood min·wk⁻¹, NPAQ Walking Outside Home Neighborhood min·wk⁻¹, MVPA min·wk⁻¹ (averaged over Baseline days), and summed VM counts/week (averaged over Baseline days). Based on bivariate analyses of potentially relevant covariates to use in the GZLMs, six (i.e., sex, race, cohabitating status, age, distance from home to worksite, and self-reported annual household income) were found to be associated with PA (ps< 0.20) and added to the models.

A sequence of four negative binomial regression models was estimated for each outcome: Model 1, a random intercepts model (with no explanatory variables); Model 2, adding the six background covariates; Model 3, adding worksite walkability (the focal predictor); and finally, Model 4, with home neighborhood walkability added to consider its confounding effect, and the ability of worksite to explain additional variability in PA outcomes. We compared relative model fit at each step in the modeling sequence using likelihood ratio tests (LRTs) with α=.05 and changes in Bayesian Information Criterion (BIC) values. While LRTs allow for formal evaluations of differences for α, they can be highly sensitive to sample size, such that relatively trivial changes in predictive power can produce statistically significant differences. Unlike information criterion indices, such as the BIC, LRTs do not penalize overall model complexity. We used Raftery’s model fit improvement criteria (Weak: ΔBIC=0-2, Positive: ΔBIC=2-6, Strong: ΔBIC=6-10, Very Strong: ΔBIC > 10) to characterize support for one model over another term (Raftery, 1995).

The study’s second aim was to examine if individual-level characteristics moderated the relationship between worksite walkability and each of the eight PA outcomes. The potential moderators, age, sex, race, annual household income, number of children under 18 in the household, the ratio of cars to drivers living in the household, and individual-level home neighborhood walkability (with a 500-meter street-network buffer), were considered in separate models.

Results

Participant Descriptive Data:

Participants in the analytic sample (n=512) were, on average, 44 years old and primarily female (59%) and White, non-Hispanic (70%), with an average self-reported BMI of 33 kg/m² (Table 1). All reported either full-time (89%) or part-time employment (11%). About half (52%) of participants reported having no children in the household, while 16.6% had 1 child, 20.5% had 2 children, and 11% had 3 or more children. The median household income was $60,000-$79,000, and the median education level was college graduate. The sample used in the analyses of self-reported outcomes (n=512) and analysis of objective accelerometer-measured analyses (n=472) did not differ substantially on demographic characteristics (Table 1), and neither sample differed markedly from the total sample (n=728) who completed the Baseline Survey.

Table 1.

Demographic and Descriptive Characteristics by Analytical Sample.

Analytical Sample Type:			Aims 1 and 2
	Self-Report (n = 512)	Accelerometer (n = 472)
Age, Mean (SD)	44.3 (9.3)		44.2 (9.3)
Female, %	59.4		58.5
BMI self-reported, Mean (SD)	33.1 (7.0)		33.0 (6.9)
Race and Ethnicity a
White (non-Hispanic), %	70.1		69.9
Hispanic, %	18.9		19.1
Black, %	6.6		6.6
Asian, %	2.3		2.3
American Indian or Native American, %	2.9		3.0
Hawaiian, %	1.2		1.3
Prefer not to answer, %	4.7		5.1
Married or living with partner, %	66.2		67.6
Employment Status
Full-time, %	88.7		89.0
Part-time, %	11.3		11.0
Current Tobacco or E-cig Smoker, %	5.9		5.9
Children in household under 18 years old
Median	0		0
Mean (SD)	1.0 (1.2)		1.0 (1.2)
Zero children under 18, %	52.0		51.3
One child under 18, %	16.6		16.7
Two children under 18, %	20.5		20.3
Three or more under 18, %	10.9		11.6
Household Income
Median, for all samples		$60,000 – 79,999	3.0
Less than $20,000	2.9		3.0
$20,000 - $39,999	10.9		10.4
$40,000 - $59,999	20.7		19.9
$60,000 - $79,999	18.9		19.1
$80,000 - $99,999	13.9		13.6
$100,000 - $119,999	12.9		13.6
Greater than $120,000	19.7		20.6
Education
Median, for all samples		College graduate
8th grade or less, %	.2		.2
Some high school, %	.2		.2
High school graduate or GED, %	5.5		5.5
Trade or technical school, %	3.5		3.6
Some college, %	25.4		24.2
College graduate, %	31.4		32.2
Post-graduate training, %	7.0		7.4
Graduate degree (MS, PhD, MD, etc.), %	26.8		26.7
Ratio of vehicles to drivers, Mean (SD)	1.1 (.5)		1.1 (.5)
Any active commuting to work, %	8.0		8.1
Reason moved to neighborhood, Mean (SD)	2.9 (1.0)		2.9 (1.0)
Years at current resident, Mean (SD)	6.8 (7.1)		6.8 (7.1)
Accelerometer Wear Time hours/day, Mean (SD)	--		15.5 (3.1)
Sufficiently Activeb at baseline, % (Measure)c	28.1 (NPAQ) 24.0 (IPAQ)		45.8 (MVPA bout)

^aRace/Ethnicity cumulative >100% as response allowed ‘select all that apply’.

^bSufficiently active based on NPAQ total time or MVPA bout min/week > = 150 min/week

^cNPAQ= summed total walking and total biking min/week; IPAQ= summed total walking and transportation biking min/week

SD = Standard Deviation

NPAQ = Neighborhood Physical Activity Questionnaire

IPAQ = International Physical Activity Questionnaire

MVPA = Moderate-to-vigorous Physical Activity (assessed by accelerometer ‘bou’ minutes)

Aim 1: Associations between Physical Activity and Worksite Walkability

Accelerometer-Measured Outcomes:

During the 2-week period of accelerometer data collection (the Baseline Phase) participants recorded an average of 161.5 MVPA bout min·wk⁻¹ (SD 125.6) and 13,735,454 summed VM counts/week (SD 3,947,535) (Appendix Table 1).

Accelerometer-measured MVPA Bout Minutes per Week.

Results of LRTs for the four negative binomial regression models indicated that the model fit with background covariates only (Model 2) was significantly better than the null model (Model 1) (p<0.001; Appendix Table 2). Neither worksite walkability (Model 3) nor home walkability (Model 4) improved model fit significantly (ps > .20). In every case, a comparison of sequential model BIC values favored the less complex model (ΔBIC values=+9.22, +4.65, and +6.15, respectively; Appendix Table 2), indicating that when taking model complexity into account, covariates in Models 2-4 did not improve fit for Model 1. Across Models 2-4, sex and age were significantly related to MVPA min·wk⁻¹. Women engaged in less MVPA than men and relatively older participants engaged in less MVPA than relatively younger participants (ps < 0.01). The results are consistent with the LRTs. Neither worksite nor home walkability was significantly related to MVPA min·wk⁻¹ (ps=0.218 and 0.927, respectively).

Vector Magnitude (VM) Counts per Week

Both LRTs and ΔBIC (Appendix Table 3) indicated that adding background covariates (Model 2) improved model fit over the null model (LRT p<0.001; ΔBIC = −39.92). However, the background covariates did not improve model fit for worksite walkability (Model 3, LRT p=0.944; ΔBIC = +6.15) and home walkability (Model 4) did not improve fit (LRT p = 0.479; ΔBIC = +5.66). Across Models 2-4, VM counts/week was significantly related to sex (counts for women lower than for men, ps<0.01), cohabitating with a partner (counts for cohabitating individuals higher than for non-cohabitating, ps<0.05), age (counts for relatively older participants lower than relatively younger participants, ps<0.05), distance to worksite (counts for participants living farther from their workplaces lower than those living relatively closer, ps<0.05), and accelerometer wear time (higher wear time associated with higher counts, ps<0.01). Neither worksite walkability (Model 3) nor home walkability (Model 4) was significantly associated with VM counts/week.

Self-Reported PA Outcomes

Based on the walking item from the IPAQ-long (Appendix Table 1), participants (n=512) reported an average of 105.4 min·wk⁻¹ (SD=162.3) of total walking, and among only those participants who reported any walking in a week (i.e., > 0 min·wk⁻¹; n=414) the average was 130.3 min·wk⁻¹ (SD=171.2).

Using the NPAQ, participants (n=512) reported an average of 118.7 min·wk⁻¹ (SD=205.0) across all neighborhood settings, and among participants who reported any walking (> 0 min·wk⁻¹; n=385) the average was 157.8 min·wk⁻¹ (SD=223.0). Participants reported an average of 65.6 min·wk⁻¹ (SD=130.3) inside their home neighborhoods and an average of 53.1 min·wk⁻¹ (SD=127.3) outside their home neighborhoods, and among those reporting any (> 0 min·wk⁻¹) walking inside (n=310) and outside (n=214) their home neighborhood averages were 108.3 (SD=153.0) and 127.1 (SD=171.5) min·wk⁻¹, respectively.

Participants (n=512) reported an average of 5.1 min·wk⁻¹ (SD=25.1) of transportation bicycling on the IPAQ-long questionnaire. Among the 34 participants who reported any transportation bicycling, the average was 76.1 min·wk⁻¹ (SD=64.7). Using the NPAQ measure of bicycling (whether for fitness, recreation, or transportation), participants (n=512) reported an average of 16.6 min·wk⁻¹ (SD=76.0). Among those reporting any bicycling (n=82), the average was 103.8 min·wk⁻¹ (SD=165.2).

IPAQ Total Walking

Neither background covariates (Model 2, p=0.098) nor worksite walkability (Model 3, p=0.062) significantly improved model fit (Appendix Table 4) for predicting IPAQ total walking (LRT ps > 0.062; ΔBIC = +26.74 and ΔBIC = +2.76, respectively). However, home walkability (Model 4) improved model fit (LRT p = 0.001; ΔBIC = −4.00) relative to Model 3. The positive association between home walkability and IPAQ total walking was the only significant association (p = 0.002) found in Models 2-4.

NPAQ Total Walking

Adding background covariates (Model 2, Appendix Table 5) to the model predicting NPAQ Total Walking significantly improved model fit (LRTp = 0.028), but the BIC value increased markedly (ΔBIC = +23.32). Worksite walkability (Model 3) did not improve model fit (LRT p = 0.443; ΔBIC = +5.65). The addition of home walkability (Model 4) yielded mixed results, with a significant LRT (p = 0.015, indicating improved fit) but also a small increase in BIC (ΔBIC = +0.31). In Models 2 and 3 (but not Model 4), sex was significantly related to NPAQ Total Walking (ps = 0.030 and 0.049, respectively), such that women reported less walking than men, and home walkability was positively related to NPAQ Total Walking (p = 0.016).

NPAQ Walking Inside Home Neighborhood

Fit for models of Walking Inside the Home Neighborhood did not improve with the addition of covariates, with no significant LRTs (all ps > .10; Appendix Table 6) and an increase in BIC at each step (ΔBICs = +4.03 – +28.34). Model 3 indicates that worksite walkability does not explain additional variance in NPAQ Walking Inside the Home Neighborhood. No significant associations were found between covariates and NPAQ Walking Inside Home Neighborhood.

NPAQ Walking Outside Home Neighborhood

As with models for walking inside the home neighborhood, covariates did not improve the fit of models for NPAQ Walking Outside the Home Neighborhood (LRT ps > 0.079; ΔBIC = +3.16 – +30.67; Appendix Table 7). No covariate was significantly related to walking outside of the home neighborhood.

NPAQ Total Bicycling

Addition of covariates did not improve the fit of models for NPAQ Total Bicycling at any step (LRT ps > 0.75; ΔBIC = +3.07 – +31.56; Appendix Table 8), and none of the covariates were significantly related to total bicycling.

IPAQ Transportation Bicycling

As with NPAQ Total Bicycling models, IPAQ Transportation Bicycling did not show improved fit at any step (LRT ps > .07; ΔBIC = +2.99 – +34.80; Appendix Table 9). No covariate was significantly related to bicycling for transportation.

Aim 2: Moderation of Relationships Between Physical Activity and Worksite Walkability:

The study’s second aim was to examine the potential moderation of the relationships between PA outcomes and worksite walkability. The moderation was stratified by each participant’s background characteristics (age, sex, race/ethnicity, income, number of children in the household, ratio of cars to drivers) and home walkability levels. Moderator variables x Worksite Walkability interactions were tested using separate GZLMs and estimated for each moderator and PA outcome combination.

Both sex and the number of children in the household interacted with worksite walkability in predicting accelerometer-measured MVPA, but no other interactions were detected (Appendix Table 10). The form of the significant Sex x Worksite Walkability interaction (B = 0.056, p = 0.042, Table 2) in predicting accelerometer-measured MVPA was such that for women, the conditional effect of worksite walkability on MVPA was positive (Effect = 6.52, p = 0.04, Figure 1), but not significantly different from zero for men (Effect = −1.38, p = 0.67; Table 2 and Figure 1). The form of the interaction between number of children under 18 in the household and worksite walkability in predicting MVPA (B = −0.025, p = 0.014;Table 3) was such that for adults with no children under 18 years old in the household, the conditional effect of worksite walkability on MVPA was positive (Effect = 7.55, p = 0.006), but not related to MVPA for those with 1 child (Effect = 2.36, p = 0.29) or 2 children (Effect = −4.16, p = 0.18; Figure 1).

Table 2.

Negative Binomial Regression for Sex Moderation Model on MVPA Bout Minutes/week.

				90% CI for IRR
Parameter	B	p	IRR	Lower	Upper
Age, mean centered	−0.014	0.002	0.99	−0.022	−0.005
Sex (female)	−0.379	0.000	0.68	−0.526	−0.232
Race a (White)	0.026	0.747	1.03	−0.132	0.184
Cohabitating	−0.088	0.378	0.92	−0.285	0.108
Number Under 18, centered	0.026	0.633	1.03	−0.081	0.133
Annual household income, centered	−0.011	0.649	0.99	−0.058	0.036
Wear-time week average, centered	<0.01	0.076	1.00	<−0.01	<0.01
Cars: People in household, centered	−0.073	0.285	0.93	−0.207	0.061
Distance to works, centered	<−0.01	0.975	1.00	<−0.01	<0.01
Household number, centered	−0.003	0.944	1.00	−0.095	0.089
Tenure at Residence, centered	0.012	0.028	1.01	0.001	0.023
Home walkability 500-m, centered	0.001	0.947	1.00	−0.029	0.031
Worksite walkability	−0.009	0.644	0.99	−0.048	0.030
Interaction Worksite Walkability x Sex	0.056	0.042	1.06	0.002	0.110

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

Figure 1.

Conditional Effect of Worksite Walkability on MVPA by Sex and Number of Children Under 18 in the Household

Table 3.

Negative Binomial Regression for Number of Household Members Under 18 years Moderation Model on MVPA Bout Minutes/week.

				90% CI for IRR
Parameter	B	p	IRR	Lower	Upper
Age, mean centered	−0.012	0.005	0.99	−0.021	−0.004
Sex (female)	−0.367	0.000	0.69	−0.516	−0.217
Race a (White)	0.034	0.669	1.03	−0.124	0.193
Cohabitating	−0.075	0.452	0.93	−0.272	0.121
Number Under 18, centered	0.012	0.824	1.01	−0.095	0.119
Annual household income, centered	−0.011	0.658	0.99	−0.057	0.036
Wear-time week average, centered	<0.01	0.100	1.00	<−0.01	0.000
Cars: People in household, centered	−0.059	0.390	0.94	−0.193	0.075
Distance to worksite, centered	<0.01	0.899	1.00	<−0.01	<0.01
Household number, centered	0.001	0.977	1.00	−0.091	0.094
Tenure at Residence, centered	0.010	0.083	1.01	−0.001	0.020
Home walkability 500-m, centered	0.004	0.810	1.00	−0.026	0.034
Worksite walkability	0.044	0.010	1.04	0.011	0.078
Interaction Worksite Walkability x # Children <18	−0.025	0.014	0.98	−0.046	−0.005

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

Discussion

This study examined the relationships between worksite walkability and self-reported and accelerometer-measured PA outcomes. Regardless of the analyses performed, results failed to support the hypothesis that worksite walkability was related to PA measured by self-report or accelerometry. Our results reflect the general findings described in Lin et al.’s (2020) meta-analysis that found out of 14 analyses of associations between PA outcomes and workplace walkability composite indices, 10 (71%) found non-significant associations (Lin et al., 2020).

Our analysis also showed limited support for associations between PA and home neighborhood walkability. Only two of the eight PA outcomes (i.e., total self-reported walking PA from the IPAQ-long and NPAQ measures) were significantly associated with home neighborhood walkability. This pattern of results differs from previous findings reporting associations between home neighborhood walkability and PA (McCormack and Shiell, 2011) (Adlakha et al., 2015; Sallis et al., 2020), particularly for accelerometer-measured PA and NPAQ Walking Inside the Home Neighborhood. Although Individuals’ eligibility to participate in the study was based on insufficient activity, defined as less than 150 minutes per week, determined through self-report during study screening, in the sample for this analysis, 28.1% were sufficiently active based on NPAQ, 24.0% based on IPAQ, and 45.8% based on accelerometer-measured MVPA minutes (see Table 1) during the baseline/lead-in phase. For the large proportion of individuals who do not already engage in sufficient PA, home and worksite walkability may not be relevant influences due to them not utilizing these particular environments to engage in PA without intervention. Studies with samples that include individuals who regularly engage in PA in outdoor environments may reveal more consistent associations of worksite and home walkability with PA.

Ding et al. highlighted the need to investigate potential moderators of the built environment and physical activity relationship (Ding and Gebel, 2012). Examining the potential moderating effects of the relationship between PA and worksite walkability yielded few significant findings. However, emerging interactions suggest possible future avenues for investigation. The association between worksite walkability and MVPA was significant and positive for women but was near zero for men. These results are consistent with Marquet et al.’s (2020) findings that, for women, MVPA conducted while at work is positively associated with workplace walkability (Marquet et al., 2020). Also, we found that worksite walkability and MVPA, while significantly and positively related among participants with no children in the household, were unrelated among those with at least one child. These results are similar to those of Bopp et al., who found that an all-female population of a similar average age, income, and educational level to the current study exhibited an inverse relationship between the number of children and active commuting to work (Bopp et al., 2014). A possible explanation for these results is that adults without children may have more discretionary time. In contrast, people with children are more likely to have responsibilities before and after work, such as transporting kids to and from daycare or school. Liu et al., found a positive association between discretionary time and time spent engaging in PA/MVPA (Liu et al., 2020). Accordingly, people with more discretionary time might be more likely to walk to a nearby restaurant or store in their worksite neighborhood or use active transportation to commute to work.

The strengths of this study are the combined use of self-reported and objectively-measured PA and GIS-measured walkability of workplace and home neighborhoods. Furthermore, the method of measuring MVPA was specialized in that measurements of MVPA were linked to each participant’s VO₂ data based on a treadmill test aimed at measuring metabolic MVPA, allowing more exact measurements of MVPA in participants’ everyday routines. Further strengths were using surveys with domain- and context-specific measures of PA behaviors, allowing for a more nuanced view into the location and purpose of self-reported PA outcomes.

A limitation of this study is that the data collection came from a population selected specifically to participate in the WalkIT AZ Intervention. This study design might have influenced these analyses, as an insufficiently active sample might be less likely to use their surrounding environments to be physically active. However, approximately a quarter (or more) of the lead-in/baseline phase sample was sufficiently active. Secondly, the number of participants reporting bicycling activity was very low, with only 82 individuals reporting any bicycling on the NPAQ and 34 individuals reporting transportation bicycling on the IPAQ. This small sample size likely made it difficult to reveal significant patterns, resulting in no significant likelihood ratio tests or parameter estimates for any models with bicycling outcomes (Appendix Tables 8 and 9). A similar study conducted in Sweden by Eriksson et al. in 2012 also had lower numbers of participants reporting IPAQ transportation bicycling, with only 7-13% during certain months (Eriksson et al., 2012). Eriksson et al. used observations from the IPAQ-long transportation bicycling questions only in months with higher reporting (20-32%). Still, they found no associations between the IPAQ-long transportation bicycling questions and walkability parameters (residential density, street connectivity, and land use mix). This finding suggests that walkability might not be a driving indicator of bicycling activity or larger street-network buffers may be necessary to investigate bicycling behavior accurately (Sallis et al., 2013). Finally, about 12% of our sample recruited between 2016 and 2018 worked from home. The COVID-19 pandemic accelerated a decoupling of work from the worksite. In 2022, about a third of employed individuals worked from home on some or all days (Statistics, 2023), suggesting that the concept of worksite needs to be broadened to consider working from home along with a potentially more dynamic measure of neighborhood exposure based on whether the individual is working from home or an external site.

A suggestion for future research would be to examine the combined effects of multiple levels of the Social-Ecological Model by collecting data about both the external worksite walkability (built environment) and the internal workplace culture (social environment) on PA behaviors. Potential avenues for analysis could be an investigation of the external worksite neighborhood and the internal social support for PA in the workplace (Hipp et al., 2015), including the extent to which the workplace promotes PA through programs, the culture, or even through the allowance of employees to take breaks to walk outside. As suggested by Van Holle et al., the combined effect of psychosocial factors, such as social support, and neighborhood walkability may result in positive, synergistic effects on PA (Van Holle et al., 2015). This suggestion is supported by Carlson et al.’s investigation into the interaction of walkability and social support on MVPA (Carlson et al., 2012). Additionally, insight could be gained through investigating the nature of individuals’ work: indoors vs. outdoors or sedentary vs. physically demanding, which might affect the use of the worksite neighborhood for PA. These avenues could help illuminate other factors about the workplace that influence the choice to use the worksite neighborhood as an environment for PA.

Lastly, COVID-19 pandemic has disassociated work from worksite (Statistics, 2023). Changes in options for working from home due to the COVID-19 pandemic and technology allow people to work remotely and offer an opportunity for an investigation into how PA behaviors are affected by working from home and the impact of home neighborhoods compared to working in an external workplace neighborhood environment.

Conclusion

This study found that in a sample of employed, relatively healthy, insufficiently active adults, worksite walkability was not directly related to any self-reported or accelerometer-measured physical activity (PA) outcomes after adjusting for demographic covariates. Home neighborhood walkability improved models’ fit for self-reported IPAQ and NPAQ Total Walking. Significant positive associations between worksite walkability and accelerometer-measured MVPA for women and those without young children highlight the need for nuanced research into the social and environmental factors that influence PA behaviors. This research can help design or modify urban environments, including those around workplaces, to optimize the equitable promotion of physical activity.

Appendix

Appendix Table 1.

Dependent Variables Descriptive Statistics by Analysis Type.

	Self-Report PAa (n = 512)						Accelerometer PA (n = 472)
	IPAQ Total Walking min/week	NPAQ Total Walking min/week	NPAQ Walking Inside Home Neighborhood min/week	NPAQ Walking Outside Home Neighborhood min/week	NPAQ Total Biking min/week	IPAQ Transportation Biking min/week	MVPA bout min/week	VM counts/week
Minimum	0	0	0	0	0	0	0	181,188
Maximum	1800	1945	1680	1570	990	300	824	26,203,478
Mean	105.4	118.7	65.6	53.1	16.6	5.1	161.5	13,735,454.4
(SD)	(162.3)	(205.0)	(130.3)	(127.3)	(76.0)	(25.1)	(125.6)	(3,947,535.4)
Median	60	60	30	0	0	0	136.5	13,565,386.5
IQR	118.8	125	80	60	0	0	152.3	5,232,038
	Self-Report any PAb (n = 414)	(n = 385)	(n = 310)	(n = 214)	(n = 82)	(n = 34)
Minimum	10	10	10	10	10	14
Maximum	1800	1945	1680	1570	990	300
Mean	130.3	157.8	108.3	127.1	103.8	76.1
(SD)	(171.2)	(223.0)	(153.0)	(171.5)	(165.2)	(64.7)
Median	80	90	60	68.5	60	55
IQR	120	140	90	110	90	75

SD = Standard deviation

PA = Physical activity

IPAQ = International Physical Activity Questionnaire

NPAQ = Neighborhood Physical Activity Questionnaire

min/week = minutes per week

IQR = Interquartile Range

MVPA = Moderate-to-Vigorous Physical Activity

VM = Vector Magnitude

^aTotal sample, including those who reported 0 minutes per week for each outcome.

^bIncluding only those who reported any minutes per week for each outcome.

Appendix Table 2.

Negative Binomial Regression Models: Accelerometer-Measured MVPA Bout Minutes per Week

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.353	−0.498	−0.209	<0.01	−0.371	−0.518	−0.223	<0.01	−0.372	−0.521	−0.222	<0.01	0.69
Race a (White)	0.045	−0.114	0.204	0.577	0.044	−0.115	0.203	0.588	0.044	−0.115	0.203	0.590	1.04
Cohabitates with partner	−0.036	−0.206	0.134	0.676	−0.034	−0.204	0.135	0.691	−0.036	−0.210	0.138	0.683	0.97
Age	−0.011	−0.019	−0.003	0.005	−0.011	−0.019	−0.003	0.005	−0.011	−0.019	−0.003	0.005	0.99
Distance to worksite	<−0.01	<−0.01	<0.01	0.758	<−0.01	<−0.01	<0.01	0.926	<−0.01	<−0.01	<0.01	0.917	1.00
Accelerometer Wear Time (min per week)	<0.01	<−0.01	<0.01	0.116	<0.01	<−0.01	<0.01	0.120	<0.01	<−0.01	<0.01	0.119	1.00
Annual household income	−0.018	−0.064	0.029	0.462	−0.019	−0.066	0.028	0.424	−0.019	−0.066	0.028	0.421	0.98
Worksite walkability 500-m					0.017	−0.010	0.044	0.218	0.017	−0.010	0.044	0.217	1.02
Home walkability 500-m									−0.001	−0.030	0.027	0.927
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 5703.011)	5712.227				5716.878				5723.027
LR Test b,c	χ2 = 33.883				χ2 = 1.506				χ2 = 0.008
	p = <0.001				p = 0.220				p = 0.929
Degrees of Freedom	7				8				9

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 3.

Negative Binomial Regression Models: Accelerometer-Measured Vector Magnitude Counts per Week

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	0.106	0.052	0.160	<0.01	0.106	0.052	0.161	<0.01	0.105	0.050	0.160	<0.01	1.11
Race a (White)	−0.008	−0.067	0.051	0.802	−0.007	−0.067	0.052	0.804	−0.008	−0.067	0.051	0.798	0.99
Cohabitates with partner	0.072	0.008	0.136	0.028	0.072	0.008	0.136	0.028	0.068	0.003	0.133	0.041	1.07
Age	−0.003	−0.006	<−0.01	0.044	−0.003	−0.006	<−0.01	0.044	−0.003	−0.006	<−0.01	0.036	1.00
Distance to worksite	<−0.01	<−0.01	<−0.01	0.040	<−0.01	<−0.01	<−0.01	0.041	<−0.01	<−0.01	<−0.01	0.035	1.00
Accelerometer Wear Time (min per week)	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	<0.01	1.00
Annual household income	−0.011	−0.029	0.007	0.217	−0.011	−0.029	0.007	0.219	−0.012	−0.030	0.006	0.197	0.99
Worksite walkability 500-m					<0.01	−0.011	0.010	0.944	<−0.01	−0.010	0.010	0.966	1.00
Home walkability 500-m									−0.004	−0.015	0.007	0.479
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 15,741.234)	15,701.315				15,707.467				15,713.122
LR Test b,c	χ2 = 83.018				χ2 = 0.005				χ2 = 0.501
	p = <0.001				p = 0.944				p = 0.479
Degrees of Freedom	7				8				9

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 4.

Negative Binomial Regression Models: IPAQ Total Self-Reported Walking.

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.263	−0.541	0.015	0.064	−0.223	−0.503	0.057	0.118	−0.192	−0.471	0.087	0.177	0.80
Race a (White)	−0.179	−0.486	0.128	0.254	−0.152	−0.460	0.156	0.333	−0.096	−0.403	0.211	0.541	0.86
Cohabitates with partner	−0.159	−0.485	0.167	0.339	−0.137	−0.462	0.188	0.408	−0.016	−0.347	0.315	0.924	0.87
Age	−0.011	−0.026	0.004	0.163	−0.012	−0.028	0.003	0.118	−0.009	−0.025	0.006	0.234	0.99
Distance to worksite	<0.01	<−0.01	<0.01	0.642	<0.01	<−0.01	<0.01	0.871	<0.01	<−0.01	<0.01	0.339	1.00
Annual household income	−0.020	−0.113	0.074	0.682	−0.013	−0.106	0.080	0.782	−0.013	−0.106	0.081	0.789	0.99
Worksite walkability 500-m				−0.048	−0.099	0.003	0.062	−0.043	−0.094	0.008	0.102	0.95
Home walkability 500-m									0.083	0.032	0.134	0.002
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 5534.821)	5561.556				5564.314				5560.312
LR Test b,c	χ2 = 10.695				χ2 = 3.481				χ2 = 10.240
	p =0.098				p = 0.062				p = 0.001
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 5.

Negative Binomial Regression Models: NPAQ Self-Reported Total Walking.

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.354	−0.673	−0.034	0.030	−0.327	−0.654	−0.001	0.049	−0.261	−0.591	0.069	0.121	0.72
Race a (White)	−0.166	−0.515	0.183	0.351	−0.161	−0.511	0.188	0.366	−0.175	−0.524	0.174	0.325	0.85
Cohabitates with partner	−0.314	−0.700	0.072	0.111	−0.292	−0.682	0.098	0.142	−0.196	−0.589	0.197	0.329	0.75
Age	0.001	−0.016	0.018	0.896	0.001	−0.016	0.018	0.923	0.004	−0.013	0.022	0.632	1.00
Distance to worksite	<−0.01	<−0.01	<0.01	0.358	<−0.01	<−0.01	<0.01	0.287	<−0.01	<−0.01	<0.01	0.684	1.00
Annual household income	−0.050	−0.155	0.054	0.345	−0.049	−0.153	0.056	0.361	−0.041	−0.146	0.063	0.439	0.95
Worksite walkability 500-m					−0.024	−0.086	0.038	0.444	−0.028	−0.091	0.034	0.369	0.98
Home walkability 500-m									0.079	0.015	0.143	0.016
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 5425.094)	5448.413				5454.063				5454.369
LR Test b,c	χ2 = 14.111				χ2 = 0.588				χ2 = 5.932
	p = <0.028				p = 0.443				p = 0.015
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 6.

Negative Binomial Regression Models: NPAQ Self-Reported Walking Inside Home Neighborhood.

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.349	−0.750	0.052	0.088	−0.346	−0.751	0.060	0.095	−0.285	−0.697	0.128	0.177	0.71
Race a (White)	−0.156	−0.601	0.288	0.491	−0.155	−0.600	0.290	0.494	−0.173	−0.619	0.272	0.446	0.86
Cohabitates with partner	−0.210	−0.706	0.287	0.408	−0.205	−0.709	0.299	0.426	−0.144	−0.652	0.364	0.579	0.81
Age	0.013	−0.010	0.035	0.263	0.013	−0.010	0.035	0.268	0.015	−0.008	0.038	0.189	1.01
Distance to worksite	<−0.01	<−0.01	<0.01	0.244	<−0.01	<−0.01	<0.01	0.243	<−0.01	<−0.01	<0.01	0.413	1.00
Annual household income	−0.037	−0.169	0.096	0.587	−0.036	−0.169	0.096	0.592	−0.028	−0.162	0.106	0.681	0.96
Worksite walkability 500-m					−0.005	−0.086	0.076	0.906	−0.007	−0.089	0.074	0.859	1.00
Home walkability 500-m									0.064	−0.021	0.149	0.141
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 4428.878)	4457.215				4463.439				4467.473
LR Test b,c	χ2 = 9.093				χ2 = 0.014				χ2 = 2.204
	p = 0.168				p = 0.906				p = 0.138
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 7.

Negative Binomial Regression Models: NPAQ Self-Reported Walking Outside Home Neighborhood

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.394	−0.969	0.181	0.179	−0.342	−0.931	0.247	0.255	−0.292	−0.888	0.304	0.337	0.71
Race a (White)	−0.235	−0.854	0.385	0.457	−0.223	−0.842	0.397	0.482	−0.229	−0.848	0.390	0.469	0.80
Cohabitates with partner	−0.425	−1.098	0.249	0.216	−0.393	−1.069	0.284	0.255	−0.247	−0.931	0.438	0.480	0.68
Age	−0.011	−0.041	0.019	0.460	−0.012	−0.041	0.018	0.444	−0.007	−0.037	0.023	0.639	0.99
Distance to worksite	<−0.01	<−0.01	<0.01	0.834	<−0.01	<−0.01	<0.01	0.647	<0.01	<−0.01	<0.01	0.897	1.00
Annual household income	−0.070	−0.255	0.114	0.455	−0.071	−0.254	0.113	0.450	−0.067	−0.252	0.117	0.474	0.93
Worksite walkability 500-m					−0.045	−0.151	0.062	0.413	−0.051	−0.159	0.056	0.348	0.96
Home walkability 500-m									0.098	−0.014	0.210	0.086
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 3396.284)	3426.952				3432.515				3435.674
LR Test b,c	χ2 = 6.762				χ2 = 0.675				χ2 = 3.079
	p = 0.343				p = 0.411				p = 0.079
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 8.

Negative Binomial Regression Models: NPAQ Self-Reported Total Biking

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.503	−1.599	0.594	0.369	−0.326	−1.407	0.755	0.554	−0.239	−1.340	0.861	0.670	0.72
Race a (White)	−0.232	−1.471	1.007	0.714	−0.282	−1.432	0.868	0.631	−0.320	−1.456	0.817	0.581	0.75
Cohabitates with partner	−0.005	−1.309	1.300	0.994	0.057	−1.158	1.272	0.927	0.091	−1.127	1.310	0.883	1.06
Age	−0.002	−0.065	0.061	0.951	0.002	−0.056	0.061	0.939	0.007	−0.051	0.065	0.812	1.00
Distance to worksite	<−0.01	<−0.01	<0.01	0.932	<−0.01	<−0.01	<0.01	0.599	<−0.01	<−0.01	<0.01	0.746	1.00
Annual household income	−0.301	−0.641	0.040	0.083	−0.322	−0.647	0.003	0.052	−0.292	−0.619	0.035	0.080	0.72
Worksite walkability 500-m					−0.159	−0.335	0.018	0.079	−0.160	−0.338	0.018	0.079	0.85
Home walkability 500-m									0.081	−0.124	0.287	0.439
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 1455.491)	1487.052				1490.118				1495.740
LR Test b,c	χ2 = 5.869				χ2 = 3.172				χ2 = 0.616
	p = 0.438				p = 0.075				p = 0.433
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio.

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 9.

Negative Binomial Regression Models: IPAQ Self-Reported Transportation Biking.

		Model 2d				Model 3				Model 4
Parameter	B	95% CI		p	B	95% CI		p	B	95% CI		p	IRRe
Sex (female)	−0.426	−2.213	1.360	0.640	0.292	−1.571	2.156	0.759	0.395	−1.551	2.341	0.691	1.34
Race a (White)	1.122	−1.775	4.019	0.448	2.068	−0.976	5.112	0.183	1.925	−1.132	4.982	0.217	7.91
Cohabitates with partner	−0.499	−2.552	1.555	0.634	0.007	−2.111	2.125	0.995	0.050	−2.086	2.186	0.964	1.01
Age	−0.037	−0.158	0.083	0.542	−0.094	−0.235	0.047	0.192	−0.090	−0.230	0.051	0.210	0.91
Distance to worksite	<−0.01	<−0.01	<0.01	0.907	<−0.01	<−0.01	<0.01	0.518	<−0.01	<−0.01	<0.01	0.474	1.00
Annual household income	−0.306	−0.913	0.301	0.324	−0.468	−1.134	0.199	0.169	−0.531	−1.288	0.226	0.169	0.63
Worksite walkability 500-m					−0.359	−0.762	0.044	0.081	−0.375	−0.784	0.033	0.072	0.70
Home walkability 500-m									−0.074	−0.499	0.352	0.735
Model Fit Information	Model 2c				Model 3				Model 4
BIC (null = 670.242)	705.044				708.036				714.160
LR Test b,c	χ2 = 2.628				χ2 = 3.246				χ2 = 0.115
	p = 0.854				p = 0.072				p = 0.735
Degrees of Freedom	6				7				8

BIC = Bayesian Information Criterion.

LR = Likelihood Ratio (−2*log likelihood).

IRR = Incidence Rate Ratio

^aIndividuals who only reported Caucasian/White race/ethnicity compared to individuals who reported any other race/ethnicity.

^bLikelihood ratio test using −2 Log Likelihood for comparison of model fit to fit of the previous model; for Model 2, the previous model (“Model 1”) was a model without any predictors (a “null” model).

^cOne-tailed tests for chi square p-values in LR Tests.

^dModel 1 is not shown as it was the null model.

^eIncidence Rate Ratios displayed for Model 3, with worksite walkability, as it is the main variable of interest.

Appendix Table 10.

Moderator x Worksite Walkability Interactions Tested for 8 Physical Activity Outcomes

Moderators
	Age		Sex		Race/ethnicity		Annual Household Income		# children <18 years in household		Ratio of cars to drivers		Home walkability
PA Outcomes	B	p	B	p	B	p	B	p	B	p	B	p	B	p
MVPA Bout Minutes per Week	0.001	0.550	0.056	0.042	−0.009	0.746	0.005	0.531	−0.025	0.014	0.000	0.995	0.000	0.976
VM Counts per Week	0.000	0.435	>0.001	0.996	−0.010	0.351	0.004	0.177	−0.003	0.412	0.003	0.755	−0.003	0.089
IPAQ Total Walking	0.000	0.912	−0.011	0.836	0.037	0.534	0.020	0.207	−0.001	0.943	0.019	0.712	−0.015	0.109
NPAQ Total Walking	0.000	0.973	0.019	0.760	0.001	0.985	0.018	0.307	−0.003	0.911	−0.015	0.788	−0.001	0.936
NPAQ Walking Inside Home Neighborhood	−0.001	0.861	−0.008	0.925	0.030	0.740	0.031	0.164	0.009	0.778	−0.008	0.917	−0.014	0.389
NPAQ Walking Outside Home Neighborhood	0.000	0.935	0.023	0.829	−0.026	0.833	−0.003	0.935	−0.014	0.731	−0.015	0.874	0.013	0.471
NPAQ Total Biking	0.005	0.694	0.084	0.671	−0.066	0.760	−0.071	0.264	−0.072	0.366	−0.171	0.415	0.016	0.675
IPAQ Transportation Biking	−0.015	0.607	0.563	0.229	0.429	0.390	−0.033	0.794	−0.150	0.340	0.976	0.127	0.012	0.909

MVPA = Moderate-to-vigorous Physical Activity (assessed by accelerometer ‘bout’ minutes)

VM = Vector Magnitude

IPAQ = International Physical Activity Questionnaire

NPAQ = Neighborhood Physical Activity Questionnaire