If You Build It, Will They Come? A Quasi-experiment of Sidewalk Improvements and Physical Activity

Abstract

Improving sidewalks could lead to more physical activity through improved access, while providing a safe and defined space to walk. Yet, findings on the association between sidewalks and physical activity are inconclusive.

PURPOSE

The purpose of this study was to examine changes in self-reported and accelerometer-derived physical activity associated with living near recently improved sidewalks in a diverse, community-based sample from the Houston Travel Related Activity in Neighborhoods (TRAIN) Study.

METHODS

Data are from 430 adults and include baseline and first annual follow-up (2014–2017). Fully adjusted, two-step regression models were built to test the hypothesis that living near (within 250-meters) an improved sidewalk was associated with greater levels of physical activity than not living near an improved sidewalk.

RESULTS

The majority of participants were female, non-Hispanic black, low income, low education, and nearly half lived near at least one improved sidewalk. After adjustment, among participants reporting some physical activity, living near two sidewalk improvements was associated with 1.6 times more minutes per week of walking and leisure-time physical activity than those not living near a sidewalk improvement (p<0.05). Based on accelerometry, which does not specifically quantify domain-specific physical activity, there were no significant associations.

CONCLUSION

Although these mixed findings warrant further research, results suggest that improving sidewalks may have an effect on participants’ physical activity. Nonspecific definitions of sidewalk improvements could be contributing to type 1 error. Future work should also examine behavioral interventions alongside changes to the built environment to determine the effects on physical activity.

Introduction

Few Americans engage in sufficient physical activity to realize health benefits.(1) Health benefits extend beyond weight control and improving the cardio-metabolic profile to prevention and management of diseases such as dementia and some cancers.(2, 3) There is also evidence that habitual physical activity improves functional ability and reduces the risk of falls and fractures in older adults.(4) Physical inactivity is said to be a pandemic(5) and the result of inadequate policy development and implementation.(6) This may be due, in part, to the failure to scale many physical activity interventions; that is to create viable, long-term programs with population-wide effects.(6) Modifications to the built environment, such as building and improving sidewalks, are considered to be scalable because they can be far reaching and sustainable.(7) Additionally, enhanced sidewalk infrastructure may lead to physical activity by increasing safety and access to key destinations and and improving aesthetics. The Centers for Disease Control’s (CDC) Guide to Community Preventive Services highlights the need for pedestrian friendly built environments that promote active transportation, and/or leisure-time physical activity.(7)

However, the association between sidewalk infrastructure and physical activity is not sufficiently clear and a causal relation has yet to be established. Early cross-sectional analyses found positive associations between sidewalks and physical activity.(8-11) Recently, prospective study design has been undertaken,(12, 13) yet techniques to measure physical activity remain limited. In their review, McCormack and Shiell (14) examined the relation between the built environment and physical activity and found that self-reports alone (without device-based assessments) were the most common method to assess physical activity. Rarely were the measurement properties of the self-report instruments provided. These methodological flaws have made it difficult to draw an uncontested conclusion on the relation between sidewalk infrastructure and physical activity and makes clear the need for further research on the topic.

The purpose of this study is to examine the association of living near recently installed or improved sidewalks with changes in self-reported and accelerometer-derived physical activity among a diverse, community-based sample from the Houston Travel Related Activity in Neighborhoods (TRAIN) Study. We addressed major gaps in the literature by utilizing robust estimates of physical activity in a natural experiment of the association between improving sidewalk infrastructure and changes in physical activity.

Methods

Study Design, Setting, and Participants

We examined baseline (2014–2016) and first follow-up (2015–2017) assessment data of the Houston TRAIN Study (median interval = 14.3 months). Briefly, the TRAIN Study is a prospective cohort study designed to evaluate the effect of a public light rail transit system on transit use and various domains of physical activity among adults living in Houston, Texas.(15) Eligible participants were at least 18 years of age, resided in the study area (3-mile Euclidean buffer around the light rail system), and did not reside with another TRAIN participant at enrollment. Study materials were available in English and Spanish, and participants were provided incentives. The TRAIN Study methodology has been described by Durand et al.(15) For this study, the analytic sample included participants with complete, self-reported physical activity at baseline and follow-up (n=430). At enrollment, all participants were offered to participate in the accelerometer protocol (wear an accelerometer on the hip during waking hours for seven consecutive days) in addition to completing the questionnaire. Approximately 53% (228/430) of participants opted-in and had returned an accelerometer at baseline and follow-up. These participants were included as an analytic subsample.(16)

Data Collection

Transportation-related and leisure-time physical activity

The Self-Administered Modifiable Activity Questionnaire (S-MAQ) was used to estimate transportation-related and leisure-time physical activity over the previous 7-days (i.e., week). The S-MAQ is derived from the Modifiable Activity Questionnaire (MAQ), which is a reliable and valid measure of physical activity.(17, 18) For both transportation and leisure-time domains, the metabolic equivalent of task (MET)·minutes per week (m·wk⁻¹) was calculated.

For the transportation-related domain, participants reported the total duration, including each one-way trip of walking, bicycling, and other activities (e.g., skateboarding) each day. To obtain the MET·m·wk⁻¹ summary estimate for each activity type, the reported duration was multiplied by a standardized MET value(19) (i.e., 3.5, 6.8, and 5.0 METs for walking, bicycling, or other activity, respectively) and summed across activities.

For the leisure-time domain, 38 common activity types were included in the S-MAQ. For each type, the participant reported if s/he did that activity in the previous 7-days, and if so, additionally recorded the total duration (number of minutes) on each day. For each activity type, the reported duration was multiplied by a standardized MET value(19) to obtain the MET·m·wk⁻¹ estimate.

Transportation and leisure walking and their totals were reported in minutes per week, independent of the total transportation-related and leisure-time estimates.

Accelerometer-derived physical activity

Participants that opted-in to the accelerometer protocol were asked to wear accelerometers on the right hip, secured by an adjustable elastic belt, during their waking hours for 7 consecutive days. Accelerometer data were sampled at 40 Hz, and data were reintegrated to 60-second epochs for further processing. Choi algorithms were used to screen for wear time, and valid wear was considered as ≥4 of the 7-days, ≥10 hours per day.(16) The primary summary estimates, from the vertical axis, were the average accelerometer counts per minute per day (ct·min·d⁻¹). Summed daily counts detected over wear periods, and time spent per day in minutes per day (m d⁻¹) in different intensity levels, were estimated using count threshold values proposed by Freedson et al.(20) In addition, we used vector magnitude (m·d⁻¹), an estimate reflecting the square root of the sum of squares from each of the three axes over detected wear periods.

Sidewalk infrastructure

Data on sidewalk infrastructure improvement for the City of Houston were obtained from the ArcGIS website (http://cohgis.mycity.opendata.arcgis.com/). The City of Houston provides publicly available geospatial information systems (GIS) datasets on areas of interest. For this analysis, we used a sidewalk inventory feature class that detailed active sidewalk projects at various stages, including programmed, design, construction, pending construction, and complete. The dataset was accessed and downloaded on September 14, 2016. To establish the exposure variable, we first created a 250-meter pedestrian network buffer around the participant’s residence. The network buffer was based on the road centerline file, excluding highways and ramps.(21) Others have suggested that 250-meter buffer lengths may be more useful than the usual 800-meter–1-kilometer distance.(13, 22) Second, an ordinal count of the number of sidewalk segments within the buffer labeled as a completed project were dichotomized as 0, 1, 2, and 3 or more improvements. For this analysis, a sidewalk segment was defined as an uninterrupted stretch of sidewalk between street crossings.

Covariates

We selected covariates based upon previous work on the correlates of physical activity related to the built environment.(14, 23) Sociodemographic data were self-reported at baseline. These included sex, age, race/ethnicity, household income, and highest level of education achieved. Self-reported height and weight were used to calculate body mass index (kg/m²). Other variables included as possible covariates were transit user status, seasonality (time of year) to account for possible seasonal differences in physical activity levels,(24) automobile ownership, and distance in meters between a participant’s home and the nearest transit stop.

Data Management and Analysis

Participants’ baseline characteristics were reported as the count and proportion of each categorical variable. The extent of missing values was assessed for each variable. Bivariate analysis of the relation between physical activity and sidewalk infrastructure improvements was stratified by transportation-related, leisure-time, walking behaviors, and accelerometer-derived activity estimates to analyze activity level differences. Hosmer-Lemeshow tests and chi-square tests for heterogeneity were used to determine differences in the median and proportion estimates, respectively. Physical activity change was calculated as the difference in estimates from baseline to follow-up and reported as the mean (±SD) value across all participants. Multivariable regression models were built to test the hypothesis that exposure to sidewalk improvements is related to change in an individual’s physical activity, when accounting for possible confounders. The follow-up assessment estimate of physical activity was modeled as the outcome with the baseline assessment locked in the model as a covariate. This method was chosen, rather than modeling the change score, based on a correlation, below 0.70 threshold, between the baseline and follow-up assessments.(25)

For the self-reported physical activity models, a two-step method was used to account for the overdispersion of zero values in the outcome estimates.(26) The two-step method first modeled a probit mean response function of observations where the outcome was recoded as zero for those with no reported physical activity at follow-up and one for those with any non-zero reported physical activity (participation portion). Then, a separate ordinary least-squares regression model with only those non-zero value observations was constructed (duration portion). The two separately derived models used different model-building techniques and therefore contained different covariates. For accelerometer-derived estimate models, there was not an overdispersion of zeros in the outcome. Therefore these models were constructed as multivariable ordinary least-squares linear regressions. Data management and statistical analysis were performed using StataSE 14.1 (StataCorp, College Station, TX).

The Houston TRAIN Study protocol and this secondary analysis were reviewed and approved by The University of Texas Health Science Center (UTHealth) at Houston Committee for the Protection of Human Subjects (UTHealth IRB). All participants provided consent to participate.

Results

Participant Characteristics

Table 1 details the participants’ characteristics and shows that the majority did not live near improved sidewalks, were female and had low incomes and education levels. Most owned vehicles and were also transit users.

Table 1.

Baseline characteristics of Houston TRAIN study participants (n=430), 2014–2017.

Characteristic at baseline	N	%
Sidewalk improvements near home1
No improvements	234	54.4
1 improvement	77	17.9
2 improvements	46	10.7
≥3 improvements	73	17.0
Female	287	66.7
Race/ethnicity
White	143	33.3
Black or African American	180	41.9
Hispanic or Latino	79	18.4
Other2	25	5.8
Household income (in thousands)
Under $20	168	39.1
$20–$39	64	14.9
$40–$79	71	16.5
≥$80	89	20.7
Don’t know	29	6.7
Educational attainment
High school or less	210	48.8
College degree	201	46.7

Notes:

The following variables contained missing data: race/ethnicity, n=3 (0.7%); household income, n=9 (2.1%); educational attainment, n=19 (4.4%).

¹Indicates the number of sidewalk improvements within a 250-meter pedestrian network buffer around the participant’s home address at baseline.

²Other race/ethnicity include: Two or more races not including Hispanic and white or black, American Indian or Alaskan Native, Native Hawaiian or other Pacific Islander, Asian or East Indian, other race.

Physical Activity and Sidewalk Improvements

For a participant exposed to improved sidewalks, Table 2 shows by intensity category the median (IQR) duration of time that s/he was physically active, measured by self-report and accelerometer at baseline and follow-up. Among exposed participants, there were no significant differences in any estimate of physical activity from baseline to follow-up.

Table 2.

Unadjusted absolute change in physical activity among those living near (within 250-meter pedestrian network buffer) a recent sidewalk infrastructure improvement, Houston TRAIN Study, 2014-2017

	n	Physical activity estimate, median (IQR)		Absolute change, mean (±SD)	p value1
		Time point 1	Time point 2
Self-reported physical activity
Transportation-related, MET m wk−1	193	0.0 (0.0-315.0)	0.0 (0.0-378.0)	−31.8 (633.5)	0.45
Leisure-time, MET m wk−1	196	753.0 (127.5-1856.3)	742.5 (117.5-1772.5)	−26.2 (2114.0)	1.0
Walking for transportation, m wk−1	196	0.0 (0.0-80.0)	0.0 (0.0-80.0)	−8.4 (161.7)	0.34
Walking for leisure, m wk−1	196	4.0 (0.0-120.0)	0.0 (0.0-120.0)	8.9 (278.5)	0.84
Total walking, m wk−1	196	80.0 (0.0-210.0)	70.0 (0.0-210.0)	0.41 (315.9)	0.34
Accelerometer-derived estimates
Wear-time, minutes	72	825.9 (755.9-899.4)	845.4 (775.8-883.9)	−12.9 (125.1)	0.44
Average counts, ct min d−1	72	208.6 (149.2-282.5)	184.9 (130.7-258.1)	−15.5 (88.0)	0.22
Average vector magnitude2	72	453.8 (355.3-554.8)	407.0 (332.9-520.1)	−32.0 (182.1)	0.35
Sedentary, m d−1	72	569.1 (504.4-661.2)	592.2 (521.1-656.0)	2.5 (132.3)	0.18
% time sedentary, mean (±SD)	72	68.8 (9.7)	70.5 (10.0)	1.7 (9.8)	0.24
Active, m d−1	72	257.0 (202.5-317.2)	248.9 (194.7-286.2)	−15.4 (79.1)	0.72
% time active, mean (±SD)	72	31.2 (9.7)	29.5 (10.0)	−1.7 (9.8)	0.24
MET m wk−1	72	1324.5 (1194.4-1469.4)	1323.8 (1217.4-1452.1)	−29.7 (201.8)	0.72

Abbreviations: IQR: interquartile range; SD: standard deviation; MET: metabolic equivalent of task; m wk-1, minutes per week;

¹Notes: Nonparametric K-sample test on the equality of medians

Multivariable analysis

The adjusted linear associations between sidewalk improvements near participants’ homes and changes in physical activity are presented in Table 3. For the self-reports of physical activity (transportation-related, leisure-time, and walking), the participation portion of the two-step model (modeling the association between living near a sidewalk and any non-zero reported physical activity) was not statistically significant (p>0.05). For transportation-related physical activity, among participants reporting any physical activity (duration portion), living near sidewalk improvements was not statistically significant (p>0.05).

Table 3.

Multivariable linear relation between changes in self-reported physical activity and improvements to sidewalk infrastructure within 250 meters of residence, Houston TRAIN Study, 2014–2017

Outcomes (covariates in model)2	n	Sidewalk infrastructure improvements, counts1
		1 sidewalk		2 sidewalks		≥3 sidewalks
		β (SE)	p value	β (SE)	p value	β (SE)	p value
Transportation-related (MET m wk−1)3
Participation predictors: B, G, I, J	400	0.02 (0.19)	0.90	0.34 (0.22)	0.13	0.33 (0.19)	0.09
Duration predictors: A, G, J	170	−0.07 (0.22)	0.74	−0.29 (0.25)	0.25	0.33 (0.21)	0.12
Leisure-time (MET m wk−1)3
Participation predictors: C, F, H, I	404	0.18 (0.20)	0.36	0.20 (0.24)	0.42	0.14 (0.20)	0.47
Duration predictors: D, F	305	0.14 (0.14)	0.32	0.46 (0.16)	0.005	0.13 (0.14)	0.35
Walking (m wk−1)4
Participation predictors: A, C	426	0.20 (0.18)	0.26	−0.09 (0.21)	0.68	0.06 (0.17)	0.74
Duration predictors: C, F, I, J	206	−0.12 (0.15)	0.44	0.50 (0.20)	0.02	−0.02 (0.16)	0.91

Abbreviations: MET m wk⁻¹: metabolic equivalent of task minutes per week; SE: standard error.

Notes: For each of the self-reported physical activity estimates, the participation portion is a probit model of a 0,1 outcome where 0 is a zero outcomes estimate (no reported physical activity), and 1 is non-zero outcome estimate; the duration portion is an ordinary least squares model that includes only those observations with a non-zero outcome.

¹Counts of sidewalk infrastructure improvements are within a 250 meter road network buffer from the participants’ home.

²Covariates in the adjusted models: A, sex; B, age in years (18-44, 45-64, ≥65); C, Race (non-Hispanic white, black, Hispanic, other); D, income (≤ 100% Federal Poverty Threshold to 199% FPT, ≥200% FPT); E, education (less than high school/GED, high school/GED, more than high school/GED); F, body mass index (<18.5-24.9, 25-29.9, ≥30); G, transit user; H, season (summer months [May-October], non-summer months [November-April]); I, automobile ownership; J, distance to nearest transit stop from home in meters.

³Transportation-related and leisure-time MET values were assigned to the specific activity reported based on 2011 Compendium of Physical Activities.

⁴Walking minutes was the sum of the reported time walking for exercise/leisure and transportation.

Considering self-reported, leisure-time activity in the duration portion of the model, living near two sidewalk improvements was associated with 1.58 times (58%) more self-reported MET minutes per week of leisure-time physical activity than those not living near a sidewalk improvement (β = 0.46, p=0.005). This statistically significant linear relation did not persist for those living near three or more sidewalks.

We found a similar result for the reported minutes per week spent walking for transportation and leisure. Among participants who reported any walking (duration portion), living near two improved sidewalks was associated with 1.64 times more self-reported walking minutes per week than those not living near an improved sidewalk (β = 0.50, p=0.02). Based on accelerometry, living near an improved sidewalk was not associated with physical activity (see Table 4).

Table 4.

Multivariable linear relation between changes in accelerometer-derived physical activity and improvements to sidewalk infrastructure within 250 meters of residence, Houston TRAIN Study, 2014–2017

Outcomes (covariates in model)	n		Sidewalk infrastructure improvements, counts1
			1 sidewalk		2 sidewalks		≥3 sidewalks
			β (SE)	p value	β (SE)	p value	β (SE)	p value
Active (m d−1)2
Predictors: Age, income level		152	−0.10 (0.07)	0.16	−0.04 (0.09)	0.64	−0.03 (0.07)	0.73
Counts (mean ct d−1)
Predictors: Age, transit user status		152	−0.12 (0.09)	0.20	−0.05 (0.12)	0.65	−0.04 (0.10)	0.67
Vector magnitude3
Predictors: Age		152	−0.09 (0.08)	0.27	0.04 (0.10)	0.70	−0.04 (0.09)	0.64

Abbreviations: MET m wk⁻¹: metabolic equivalent of task minutes per week; SE: standard error.

¹Notes: Counts of sidewalk infrastructure improvements are within a 250 meter road network buffer from the participants’ home.

²Accelerometer-derived time spent active was the mean daily time active based on Freedson cutpoints for moderate-to-vigorous physical activity (≥1952 counts).

³Mean value of the square root of the total sum of squares from each of the three axes over detected wear periods

Discussion

In this study, we prospectively evaluated the association of living near sidewalk improvements with changes in physical activity levels. Few studies have combined self-reported and device-assessed physical activity data within natural experiments of the built environment. While our findings in unadjusted models were not statistically significant, in a multivariable analysis we found self-reported leisure-time physical activity and time spent walking was significantly greater (~60%) among those who reported some activity and lived near two sidewalk improvements. Self-reported, transportation-related physical activity and accelerometer estimates, on the other hand, indicated that there were no associations with sidewalk improvements. These mixed results could reflect the differing constructs measured with self-reports and device-based assessment of physical activity. The S-MAQ primarily operationalizes the time spent in structured, higher intensity physical activities, like leisure walking or jogging on a sidewalk. Alternatively, accelerometry captures all ambulation, including, structured and unstructured activity, the most of which does not take place on a sidewalk and is accumulated in small bursts (walking around the home/office). Therefore, the accelerometer estimates, although less prone to self-reporting bias, may be diluted by activities unrelated to sidewalk infrastructure improvements. Nonetheless, future work should continue to combine self-reports with device-based assessments of physical activity, and should consider incorporating geospatial data (i.e., GPS) to further elucidate sidewalk versus non-sidewalk (e.g., indoor) physical activity.

The results of our study have implications for health promotion professionals. Most notably, the data suggest that improving sidewalks was not sufficient to increase physical activity among those who are inactive. Whereas, for those who are already active, living near two improved sidewalks was associated with increases in reported leisure-time and walking activity. This distinction could be important. The 2008 Physical Activity Guidelines for Americans states that any physical activity is beneficial, and at least 150 minutes of moderate-to-vigorous aerobic physical activity provides a variety of health benefits.(4) This could indicate that sidewalks provide opportunities for already-active people to be even more physically active, especially if they seek opportunities around their home. It is interesting that the significance of this finding did not hold for three or more sidewalk improvements. We hypothesize a few possible reasons for this finding. This may be the result of a smaller sample (n=46) of participants living near 2 sidewalk improvements, or a possible threshold effect that occurs as the result of 3 or more sidewalk improvements being inhibitive of physical activity (e.g., increased construction). There is also the possibility that the significant findings for 2 sidewalks is a type 1 error.

Although we did not evaluate sidewalk use, our results can be compared with other prospective evaluations of the pedestrian environment and physical activity changes. First, Evenson et al.(27) found that respondents who live within 2 miles of a newly constructed walking/biking trail were not more likely to increase their physical activity. Specifically, the authors found that those who increased their walking from baseline to follow-up were 57% less likely to have reported using the walking/biking trail, compared with those who showed no change in physical activity. Additionally, those who decreased their bicycling were more likely (~4.0 times) to have used the trail. Second, Burbridge and Goulias(12) conducted a similar analysis of how a newly constructed trail affected physical activity among adults who lived within 1-mile of it. After controlling for relevant factors, physical activity bouts were found to significantly decrease from baseline to year 3 follow-up among those living near the new trail. These two studies, along with our results, indicate that improving or building new sidewalks/trails may not be enough of an incentive to the people who live nearby. The CDC’s Guide to Community Preventive Services recommends changing the built environment to support and provide opportunities for the community to be physically active, including connected sidewalks and streets but recognizes the need for complementary behavioral and social approaches.(28) Our results underscore this recommendation. Pairing infrastructure changes with behavioral economic theories, such as asymmetric paternalism,(29) could be an area for future investigation.

The current study has its limitations. Only those sidewalk projects that had a completed construction date before participant follow-up (2016) were included in the current analysis. As such, the results should be interpreted as the change in physical activity being associated with sidewalk infrastructure improvement near the home. Additionally, there is no clear definition from the City of Houston of a sidewalk improvement. This could include construction of a new sidewalk, a crack repair, or a curb cut to facilitate disability access. Finally, although quasi-experimental design helped elucidate changes in physical activity over a 1-year period, a longer follow-up period could provide a better understanding of the relation and whether the increases in physical activity diminish in subsequent years or perhaps require a critical mass to become a social norm.(30)

Nonetheless, this study utilized a natural experiment and a prospective design to examine the changes in physical activity associated with an improved built environment. Additionally, a robust measurement of physical activity that combined domain-specific assessments of self-reported physical activity with accelerometry was used along with a novel statistical approach to highlight the public health implications of the findings. The results indicate that participants who engage in some physical activity and live near two sidewalk improvements report significantly more walking and leisure-time physical activity than active participants who do not live near an improvement. This association was not observed among accelerometry participants. In order to examine short- and long-term effects on physical activity, future studies should utilize temporally structured and detailed assessments of sidewalk infrastructure and attempt to examine the effects of theory-driven behavioral intervention alongside changes to the built environment.