The Influence of Social Networks and the Built Environment on Physical Inactivity: A Longitudinal Study of Urban-Dwelling Adults

Abstract

Policies targeting the built environment to increase physical activity may be ineffective without considering personal social networks. Physical activity and social network data came from the Montreal Neighborhood Networks and Healthy Aging Panel; built environment measures were from geolocation data on Montreal parks and businesses. Using multilevel logistic regression with repeated physical inactivity measures, we showed that adults with more favorable social network characteristics had lower odds of physical inactivity. Having more physical activity facilities nearby also lowered physical inactivity, but not in socially-isolated adults. Community programs that address social isolation may also benefit efforts to increase physical activity.

Introduction

The concept of physical activity refers generally to bodily movement produced by the skeletal system leading to energy expenditure.¹ Physical inactivity is defined as activity level insufficient to meet current physical activity recommendations, and is thus seen as a risk category for various diseases.^1–4 Physical inactivity has long contributed to the prevalence and incidence of non-communicable disease in higher income countries, but it is also now a major contributor in middle- and low-income countries.⁵ In 2000, poor diet and physical inactivity were considered the second leading cause of death behind tobacco in the United States, and is expected to surpass tobacco in the future.⁶ More recently, physical inactivity has been ranked as the fourth leading cause of death worldwide, and contributes to millions of deaths globally.⁷

Social networks and the built environment are widely recognized as important determinants of health and health behaviors, including physical activity (PA).^8,9 Research has increasingly shown the importance of one’s social networks and environment, including family, community, and neighborhood settings on PA.¹⁰ The social ecological model, for example, posits the embedding of individuals within broader layers of the social environment, such as social networks and neighborhoods, and the influence of the broader environment on individual health behaviors and conditions. Standard recommendations for reducing individual physical inactivity may thus be ineffective without considering the social and built environmental factors influencing PA behaviors. Hence, research should shift from focusing primarily on individual behaviors to the role of social and built environmental contexts as key modifiable determinants of PA levels.¹¹

Social Environment: Social networks and capital

Research has shown the benefits of social networks for a range of health promoting behaviors such as PA.⁹ Social network analysis examines the pattern of social connections emerging from people’s social relationships.⁸ Social networks are considered meso-level characteristics, meaning that they shape downstream, micro-level health behaviors and conditions, while also being shaped by upstream, macro-level factors like social policies and socioeconomic factors.^8,9 Within a social network, a person may or may not have a connection or relationship to others in the network. Those without network ties to others are often considered to be socially isolated and, therefore, unable to access or leverage the various types of social resources (e.g., social capital, social support) that may be accessible to others in the network. Previous research has shown socially isolated adults to be at greater risk of a range of poor health behaviors, including physical inactivity.¹² Those persons who have social connections may have stronger relationships (e.g., be emotionally closer) with certain individuals and weaker ones to others. Strong ties may be important for a person’s PA behavior via social mechanisms, such as social support or influence, whereas weak ties may benefit PA behavior through other mechanisms, such as access to information. The composition of a person’s network may also shape PA behavior. For example, adults that have more active people in their social networks have been shown less likely to be physically inactive, suggesting the possible importance of social influence mechanisms within networks.¹³

Social capital refers to the resources to which individuals or groups have access through their social networks. The benefits of social capital tend to arise through a person’s weaker social ties and connections.¹² Researchers often describe social capital as having several different forms: social trust, social participation, and network social capital.¹³ Findings on social capital and PA have been mixed, which may be due in part to inconsistencies in how researchers have measured social capital and PA. For example, Lindstrom (2011) showed that low social capital in the form of trust was associated with lower odds of leisure time PA.¹⁴ In contrast, Legh-Jones et al. (2012) showed trust to be unrelated to physical inactivity, but greater network diversity to reduce and the lack of social participation to increase the odds of physical inactivity.¹⁶ Poortinga (2006) showed trust to be weakly associated in adults engaging in walking and sports, while high and medium social participation was associated with higher odds of overall activity.¹⁷ In addition, most research on social capital and PA has been cross-sectional, further suggesting the need to assess longitudinally the relationship between social capital and physical inactivity.

Built (Physical Activity) Environment

Recreational facilities and green spaces have been shown to provide health benefits to both youth and adult populations. Living in neighborhoods with a high density and a variety of non-residential land uses such as parks, play areas, and recreational facilities has been shown associated with higher rates of active transportation in children, overall PA in adults and children,^16–18 and a higher likelihood of meeting the 150 minutes per week recommendation.¹⁹ In Montreal, Canada, researchers have shown longitudinally that older adults residing in areas with more amenities were more likely to walk frequently over a three-year period.²⁰ Perceived access to open spaces, parks and sidewalks has been linked to increased PA, such as walking and vigorous activity.^21,22 Researchers have also suggested that the effects of the neighborhood built environment on health may depend in part on individual perceptions and experiences of the social environment.²⁴ For example, Carlson et al. (2012) showed a joint relationship between aesthetics of the built environment and psychosocial factors in PA among older adults, whereby individuals with social support and residing in a walkable environment were more likely to engage in weekly PA than those with either one or the other.²⁴ Adults with more social connections may be better able or willing to engage with local physical activity resources. Nevertheless, few longitudinal studies have assessed whether built and social environmental characteristics act independently or jointly to influence adult PA behaviors. A randomized control trial with a 3-month follow-up found that both social support and aspects of the built environment were independently and jointly related to walking time.²⁵ In addition, few studies have distinguished between the different types of PA-related structures, such as parks or green spaces (PoGs) and other recreation facilities, and whether those structures interact with a person’s social environment to influence their behavior.^26,27

Using three waves of data on physical inactivity over a five-year period, the objective of this study was to examine the social network and built environmental influences on physical inactivity among urban-dwelling adults, and whether these factors were independently or jointly associated with adult physical inactivity over time. We tested the following set of hypotheses:

1. Hypothesis one: Adults with larger or well-connected networks are less likely to be physically inactive over time.

   • a.
     Socially-isolated adults are more likely to be physically inactive.
   • b.
     Participants with more adults who exercise in their networks are less likely to be physically inactive.
   • c.
     Participants with greater social capital are less likely to be physically inactive.
2. Hypothesis two: Adults residing in areas with more physical activity resources are less likely to be physically inactive over time.

   • a.
     Participants who reside in areas with more physical activity-related facilities are less likely to be physically inactive.
   • b.
     Participants who reside closer to a park or green spaces (PoGs) are less likely to be physically inactive.

3. Hypothesis three: Adults who have larger social networks and reside in areas with more physical activity resources are less likely to be physically inactive over a five-year period compared to adults who have smaller networks and reside in areas with fewer resources, or those who have higher levels of one or the other.

Methods

Sample

Data came from the Montreal Neighborhood Networks and Healthy Aging Panel (MoNNET-HA), a cohort that consists of adults aged 25 years and older living in the Montreal Metropolitan Area (MMA). The MoNNET-HA study sample was recruited using a two-stage stratified cluster sampling design to recruit a representative sample of 2707 Montreal, Canada residents.²⁸ The inclusion criteria for the study were non-institutionalized adults that lived in their current address for at least one year, and able to complete the questionnaire in either French or English. Random digit dialing of listed telephone numbers was used to select households for participation. Questionnaires were administered using a computer-assisted telephone interviewing system with the response rate being 38.7% for wave one participants. Data from MoNNET participants were collected three times over 5 years: in 2008, 2010, and 2013. Comparisons of the 2008 MoNNET sample to 2006 Canada census data showed the sample to over-represent older adults (by design), females, individuals in households with less than $50,000 per year, and people who have resided in their home for more than five years.^16,28 Further information on the MoNNET-HA sampling design can be found elsewhere.²⁹

Outcome

The main outcome for this study was physical inactivity. Physical inactivity was measured using an adapted version of the short International Physical Activity Questionnaire (IPAQ) and calculated using IPAQ analysis algorithms and recommended cutoffs.²⁶ These have been shown to have acceptable measurement properties, particularly among 18–65 years old.²⁷ The IPAQ integrates questions about the total volume of PA and the number of days per week the activities were conducted in order to calculate the energy costs of activities as the metabolic equivalent of task (MET). Vigorous, moderate, and walking activities are converted at 8.0 MET, 4.0 MET, and 3.3 MET, respectively. For each activity level, respondents were asked, “During the last 7 days, on how many days did you do this type of activity?” and “how much time did you spend doing this activity on one of those days?” An activity had to have been carried out for at least 10 minutes to be recorded as such. The total MET for each person was the sum of METs for each activity multiplied by their frequency (minutes and days). IPAQ cutoff guidelines were used to classify respondents into high, moderate or low/inactive PA levels. High PA was defined as at least 3 days of vigorous activity on at least 3 days accumulating at least 1500 MET-min/week, or 7 or more days of any combination of walking, moderate or vigorous activity achieving a minimum of at least 3000 MET-min/week. To be moderately active, one of three criteria had to be met: (1) three or more days of vigorous activity of at least 20 min per day, (2) five or more days of moderate activity or walking of at least 30 min per day, or (3) five or more days of any combination of walking, moderate or vigorous activities achieving a minimum of at least 600 MET-min/week. For each wave, individuals were classified as being physically inactive if they did not achieve the criteria for moderate or high PA.²⁶ For this analysis, a binary variable was created to indicate physical inactivity for each of the three waves.

Exposures

The main exposures for this study were a person’s social networks and their built environment in 2008 (i.e., wave one of the MoNNET study). The social network exposures were: (1) social isolation, (2) the number of adults in a person’s core networks that exercised regularly, and (3) three different measures of social capital (i.e., network capital, generalized trust, and social participation). For the built environment, our measures focused on physical activity-related locations. Two built environment exposures were thus examined: (1) residential exposure to recreational- and sports-related facilities and (2) geographical proximity to PoGs. Although characteristics of a person’s social networks and built environment may change over time, our study uses network and built environment data from one time point – 2008 – to examine their relationship with physical inactivity over a relatively short five-year period.

Social networks:

To capture strong, core networks, participants were asked to name up to three people (i.e., their alters) with whom they had discussed important matters in the past six months. If a person did not name anyone, we repeated the question to confirm they had not discussed important matters with anyone in the past six months. Those who confirmed their response of not having spoken to anyone were considered social network isolates. This measure has been used in other studies to indicate social isolation.³² If a person named one, two or three discussant alters, we asked whether their alters did or did not exercise regularly. “Exercising alters” represents the number of alters (min. 0; max 3) whom were reported to exercise regularly.

To capture the network, cognitive, and structural dimensions of social capital, we used three different social capital measures. For the network dimension, we used a position generator.^33,34 Position generators ask individuals to indicate whether or not they know someone who holds a certain occupation. The occupations listed on the position generator represented a random selection of occupations from high to low prestige in Canadian society. Three network measures were calculated using the position generator: diversity (i.e., the number of different occupations accessed), reach (i.e., the highest prestige occupation accessed), and range (i.e., the difference between the highest and lowest prestige occupation accessed). Principal components analysis was used to estimate network capital from these three measures with the first component used to represent the network capital score. Further details on the construction of the network social capital measure can be found in previous publications.^16,29 Cognitive dimensions of social capital were measured with a question on generalized trust. The question read, “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” High generalized trust was restricted to those participants reporting that “Most people can be trusted.” To capture structural social capital, we asked participants if they had been active as a member or officer in a neighborhood group or association, for any other group or association. Social participation was treated as a categorical varible consisting of (1) no social participation, (2) low participation, i.e., involvement in either a neighbourhood or other voluntary association or group, and (3) high participation, i.e., involvement in both a neighbourhood and other voluntary association or group. These three different dimensions and measures of social capital have been used in previous studies of social capital and health, including physical activity.^14,15

Built environment:

Information on the built environment was subcontracted from researchers housing the MEGAPHONE geographic information system (GIS).²⁹ Business records were classified using the Standard Industrial Classification (SIC) codes and used to identify sports and recreation clubs (#7997) and physical fitness facilities (#7991). Business locations were geocoded using GeoPinPoint software. A previous study using onsite field visits to validate the location of the food outlets in the registry showed good agreement (0.77) and positive predictive value (0.90).³⁰ GIS was used to identify and sum the number of PA-related facilities, like sports clubs, dance locations, and bowling alleys, within a 500-meter circular buffer of each participant’s residence. Previous research in Canada has also applied a 500-meter buffer to define the neighborhood zone.³⁶ GIS tools were also used to calculate the distance from a person’s place of residence (based on their 6-digit postal code) and the PoG nearest to them. Two measures of the built environment were thus calculated and used in these analyses: (1) the number of PA-related facilities within a 500-meters buffer zone of a person’s residence and (2) the geographical distance from a person’s residence to the nearest PoG.

Confounders

Ages were grouped into the following categories: 25–34 years old, 35–44, 45–54, 55–64, 65–74, and 75 years or older. Respondents reported their marital status (as married/common law, divorced/separated, single or widowed), household language, and years at the current residence. Principal components analysis was used to estimate participants’ overall socioeconomic status (SES) from their responses about educational attainment, household income, and employment status. Further information on the SES measures of MoNNET participants may be found elsewhere.²⁵ Respondents also self-assessed and reported their health using one of five categories: poor, fair, good, very good, or excellent. At the neighborhood level, the study also adjusted for the average socioeconomic level of the census tract in which participants were residing using a composite measure of 2006 Canada census data on census tract education and median income.

Statistical Analyses

Our analyses examined whether wave-one social network and built environment exposures were related to changes in participant’s physical inactivity over time. Hierarchical logistic regression was used for this analysis. Repeated measures of physical inactivity were nested within individuals, and individuals nested within the census tracts from which they were sampled. Although the sample size decreased over the 5-year period, each individual had between 1 and 3 measures of physical inactivity. Individuals were excluded from the analyses if the outcome or exposure variables were missing at baseline (n = 11).

The analyses proceeded in the following order: The differences between those adults who were physically active and those inactive were assessed using a Chi-Square test for categorical data and t-test for continuous data. Next, the independent association between each exposure variable and physical inactivity was assessed in the bivariate model. We then examined in separate models whether each of the social network variables modified the association between the built environment and physical inactivity. Finally, in the fully-adjusted multivariate model, we assessed whether the bivariate associations remained significant after adjusting for all exposures and confounding variables. Each model included a time variable representing the data collection period to assess whether the odds of physical inactivity increased or decreased across waves. In ancillary analyses, we tested whether time moderated the relationship between each wave-one exposure and changes in physical inactivity to examine whether the effect of the network and built environment exposures differed between waves. Since no time by exposure interactions were significant in fully adjusted models, they were not included in the final results. All analyses were conducted in Stata 14 using the Generalized Linear Latent and Mixed Models (GLLAMMs) with adaptive quadratures.

Results

The wave-one sample size was 2,696 individuals, with 1,381 and 971 of those providing further data at waves two and three respectively, resulting in a total of 5,048 observations. Table 1 provides a summary of the sample characteristics at wave one, as well as the prevalence of physical inactivity at each wave. The prevalence of physical inactivity gradually increased from 17% at wave one to 24% at wave three. Approximately 73% of the participants had at least one alter that exercised regularly at wave one, with more physically active adults tending to have at least two exercising alters. Physically inactive adults tended to have no exercising alters (42.5% versus 24.1%). At wave one, the physically inactive group had a greater number of older adults and women, and tended to have lower SES, trust, and participation (p < 0.001).

Table 1.

MoNNET-HA Participant Characteristics at baseline, 2008; n = 2,696

Variable n (%)	All n = 2,696	Physically Inactive n = 461	Physically Active n = 2,235
Number of Participants
Wave 1 (baseline)	2,696	461 (17.1)	2,235 (82.9)
Wave 2	1,381	280 (20.3)	1,101 (79.7)
Wave 3	971	229 (23.6)	742 (76.42)
Number of Exercising Alters
0	734 (27.2)	196 (42.5)	538 (24.1)***
1	697 (25.9)	120 (26.0)	577 (25.8)
2	761 (28.2)	102 (22.1)	659 (29.5)
3	504 (18.7)	43 (9.3)	461 (20.6)
Socially Isolated	374 (13.9)	117 (25.4)	257 (11.5)***
Generalized Trust
Low	1,540 (57.5)	308 (67.5)	1,232 (55.5)***
High	1,137 (42.5)	148 (32.5)	989 (44.5)
Participation
No	1,711 (63.5)	331 (71.8)	1,380 (61.7)***
Low	694 (25.7)	101 (21.9)	593 (26.5)
High	291 (10.8)	29 (6.3)	262 (11.7)
Social Capitala	0.001 (0.97)	0.26 (1.06)	0.06 (0.94)***
Distance to nearest PoG (meters)a	570.4 (827.3)	546.1 (569.7)	575.4 (871.1)
Number of PA facilities
0	2,197 (81.5)	385 (83.5)	1,812 (81.1)
1 – 3	464 (17.2)	71 (15.4)	393 (17.58)
4+	35 (1.3)	5 (1.1)	30 (1.3)
Census Tract SESa	−0.001 (0.91)	0.03 (0.90)	0.01 (0.92)
Gender (Females)	1,744 (64.7)	332 (72.0)	1,412 (63.2)***
Age Category (years)
25 – 34	395 (14.7)	39 (8.5)	356 (15.9)***
35 – 44	474 (17.6)	59 (12.8)	415 (18.6)
45 – 54	540 (20.0)	76 (16.5)	464 (20.8)
55 – 64	439 (16.3)	68 (14.8)	371 (16.6)
65 – 74	565 (21.0)	130 (28.2)	435 (19.5)
75+	283 (10.5)	89 (19.3)	194 (8.7)
Self-Reported Health
Excellent	555 (20.6)	47 (10.2)	508 (22.7)***
Very Good	920 (34.1)	118 (25.6)	802 (35.9)
Good	839 (31.1)	180 (39.1)	659 (29.5)
Fair	288 (10.7)	79 (17.1)	209 (9.4)
Poor	90 (3.3)	37 (8.0)	53 (2.4)
Annual Household Income (Canadian $)
< $28,000	563 (20.9)	155 (33.6)	408 (18.3)***
$28,000 – $49,000	761 (28.2)	137 (29.7)	624 (27.9)
$50,000 – $74,000	725 (27.9)	103 (22.3)	622 (27.8)
$75,000 – $100,000	343 (12.7)	36 (7.8)	307 (13.7)
> $100,000	304 (11.3)	30 (6.5)	274 (12.3)
Educational attainment
Less than High School	320 (12.0)	106 (23.3)	214 (9.6)***
High School/Trade	780 (29.2)	142 (31.1)	638 (28.8)
Some College	553 (20.7)	91 (20.0)	462 (20.8)
University degree	1,022 (38.2)	117 (25.7)	905 (40.8)
Employed	1,466 (54.4)	163 (35.4)	1,303 (58.4)***
Marital Status
Single	548 (20.5)	91 (19.8)	457 (20.6)
Married/Common Law	1,453 (54.2)	226 (43.2)	1,227 (55.3)
Divorced/Separated	400 (14.9)	55 (12.0)	345 (15.5)
Widowed	279 (10.4)	87 (19.0)	192 (8.6)
Household Language
French	2,099 (78)	345 (74.8)	1,754 (78.7)
English	368 (13.7)	65 (14.1)	303 (13.6)
Other	224 (8.3)	51 (11.1)	173 (7.76)
Residential Duration, yearsa	14 (12.9)	15.1 (13.4)	13.9 (12.8)

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

Abbreviations: PoG – Parks or green space

The counts and proportions are related only to the subpopulation in the particular column.

^aValues reported as mean (standard deviation). Social capital is based on principal components analysis.

Table 2 reports the estimates of the multilevel logistic models. The first column reports the unadjusted estimates of the main exposures (i.e., the social network and PA environment variables) from the bivariate models. The variables representing social network characteristics – social isolation, exercising alters, network capital – and the number of PA facilities were each negatively related to physical inactivity. Park distance was not related to physical inactivity. In Model 2 analyses, only the number of PA facilities was shown to modify the association between social isolation and physical inactivity (p = 0.008). After adjusting for all confounders, having exercising alters [OR = 0.74 (0.66, 0.84)], greater network capital [OR=0.86 (0.76, 0.97)], and high social participation [OR = 0.67 (0.46, 0.98)] reduced the odds of physical inactivity. As shown in Figure 1, effect modification of the association between social isolation and physical inactivity by the number of PA facilities remained significant (p = 0.04). Trust was not associated with physical inactivity in the full model.

Table 2.

Adjusted and Unadjusted Odds Ratios and 95% confidence intervals of physical inactivity from multilevel logistic regression analyses, n = 2,696

Variable	Model 1	Model 2
	Bivariate Model	Multivariate Model
Wave	1.14 (1.09, 1.19)***	1.49 (1.33, 1.67)***
Social Networks
Number of Exercising Alters	0.63 (0.57, 0.70)***	0.74 (0.66, 0.84)***
Network Capital	0.70 (0.63, 0.78) ***	0.86 (0.76, 0.97) *
Social Participation
No	1.00	1.00
Low	0.75 (0.59, 0.97) *	0.96 (0.75, 1.24)
High	0.49 (0.33, 0.70) ***	0.67 (0.46, 0.98) *
Generalized Trust
Low	1.00	1.00
High	0.58 (0.46, 0.72) ***	1.03 (0.82, 1.29)
Social Isolation
No	1.00	1.00
Yes	2.71 (2.01, 3.67) ***	0.83 (0.56, 1.23)
Built Environment
Distance to PoG	0.99 (0.99, 1.00)	−
PA Facilities (500 m)	0.92 (0.87, 0.97) **	0.92 (0.86, 0.98)**a
Variance Components
Level 2: ID		1.82 (0.31)
Level 3: Census Tract		0.03 (0.08)

^*p<0.05;

^**p<0.01;

^***p<0.001.

PA Facilities is the number of physical activity facilities within 500 meters of the individual’s residence. The bivariate model presents the association between each variable with the outcome. The Multivariate model (Full model) adjust for sex, age, self-reported health status, SES, household language, marriage status, residential duration, and wave, and includes any significant interaction terms.

^aNote, an interaction term was significant; this value will vary over levels of social isolation.

Figure 1.

Odds of physical inactivity for adults with and without social ties residing in areas with low, moderate, and high number of physical activity locations. Footnotes: Each odds ratio compares the odds of having x physical activity locations to the odds of having zero physical activity locations within 500 meters.

Discussion

This longitudinal study examined the relationship among physical inactivity, social networks, and the PA environment, and whether social and built environment features acted synergistically to influence physical inactivity. Using physical inactivity data collected at three time points, we tested three sets of hypotheses. First, we hypothesized that participants with larger or more resourceful networks were less likely to be physically inactive over time. Our study mostly supported each of the three sub-hypotheses on social networks and physical inactivity, thereby highlighting the overall importance of social networks on PA behavior. Social isolation – not having network connections – was associated with higher odds of physical inactivity (H1.a). Research has shown socially-isolated adults to report multiple unhealthy behaviors, including physical inactivity.⁸ The effects of social isolation on physical inactivity may operate through the lack of social support or social cues for healthy behavioral choices, particularly among older adults.¹² Our study also showed that having exercising alters in one’s core network lowered the odds of physical inactivity(H1.b). Being exposed to a greater number of friends or family that exercise may reduce a person’s odds of being physically inactive via a number of social mechanisms, including support, social control, role modeling or behavioral guidance.⁸ Whereas social isolation highlights the absence of network ties for support in general, having exercise alters captures the positive modeling and support that networks might provide in support of healthy behaviors. In addition to the importance of core ties, this study also showed network social capital was associated with lower odds of physical inactivity (H1.c). Access to social resources, more generally, may provide individuals with greater knowledge of the benefits of healthy behaviors, including PA. While cross-sectional research has shown the importance of network social capital¹², this study improves the previous findings to report the effects of network social capital on physical inactivity over time. High social participation was also associated with lower odds of physical inactivity levels. Social participation represents greater exposure to social contacts and opportunities.¹⁴ Participation may also reflect a personality trait that increases the likelihood of being socially active. For example, research has shown that college students that engage in social educational settings tended to have an extroverted and open personality.³⁷ In a study of older adults, greater social participation was associated with higher vitality, walking, general health, and perceived access to resources.³³ These findings suggest that social networks likely work through a range of social mechanisms, including social influence, support, and social capital. Overall, adults in our sample tended to become more physically inactive over time, regardless of their environments. Future research might examine these distinct mechanisms to assess whether and in which contexts they play the greatest role.

Second, we hypothesized that adults who resided in areas with more PA-related amenities or closer proximity to PoGs were less likely to be physically inactive over time. Using two distinct measures of the PA environment, our bivariate analyses showed only one of them - the number of recreational and sports-related facilities – related to lower odds of physical inactivity. Distance to PoGs was not associated with physical inactivity, although previous research has shown the importance of local parks for PA behavior.³³ Park distance differs from these previous measures of perceived access or park density, which may help explain some differences in findings. However, it may also be the case that parks are not as important for reducing physical inactivity in our sample of urban-dwelling adults. Yet, further research is needed to assess whether our findings are particular to this cohort or extend more broadly to other populations and cities in North America.

Finally, our study tested the hypothesis that the built PA environment modified the association between social networks and physical inactivity. Previous cross-sectional research has suggested that built environment aesthetics and personal psychosocial characteristics may work jointly to improve PA behavior in older adults.²⁴ Our study supported such ideas by showing that the PA environment reduced the odds of physical inactivity only in those adults who were socially-connected. Socially-isolated adults were at greater overall risk of physical inactivity, but the PA environment neither increased nor decreased their odds of inactivity. This finding highlights the differential role that the built environment may play in population health and suggests that some degree of social connectivity may be needed for groups to take advantage of local built environment resources. Thus, programs and policies designed to increase PA through the built environment may fail to reach more at-risk groups (e.g., socially-isolated adults) and, thus, be less effective than those that incorporate network-related components.

Limitations

There were a number of limitations that should be considered. First, the PA of the participants as well as their knowledge of their alters’ exercising habits were self-reported. Previous studies have compared objectively measured and self-reported PA and found that self-reported responses can over or underestimate a person’s actual PA behavior.^34,35 Nevertheless, our study highlights the importance of ego’s perceptions of the alters’ PA behavior as associated with the ego’s odds of physical inactivity. Second, the current study suggests that social networks are related to PA via different social and psychosocial mechanisms, including social influence, support, and social capital. Yet, disentangling network influences as well as environmental influences from social selection biases remain a challenge. Selection biases suggest, in this case, that adults who were physically active may be more likely to choose friends who are also physically active or to live in places with more physical activity resources. Longitudinal studies compared to cross-sectional studies are better fit to disentangle selection from influence, but limitations remain. Third, as with most longitudinal studies, the MoNNET panel experienced attrition over time. Previous research has shown recurrent MoNNET participants tending to be primarily French speakers, higher educated, middle-aged, and residing in their current place of residence longer.²⁹ Attrition was also affected by levels of social connectivity, with less connected or isolated individuals less likely to participate in the later waves. Finally, we recorded participants’ PA levels in different seasons (summer 2008; winter 2010; spring 2013). It may be that the PA built environment, including distance to PoG, would have had a stronger association with physical inactivity if the data all came from warmer seasons.

Conclusion

Despite its limitations, this longitudinal study is unique in its examination of the relationship among physical inactivity, social networks, and the PA environment, in a representative sample of urban-dwelling adults. We tested three sets of hypotheses on the direct and indirect influences of these social and environmental factors on physical inactivity. Few studies have assessed these relationships with repeated measures of physical inactivity over time. Findings from this study can inform programs and policies aimed at increasing PA in adults. Programs, such as neighborhood “buddy systems” and walking groups, may be effective in leveraging the importance of social networks for reducing physical inactivity, while also taking advantage of local PA built environment amenities.⁴¹

Highlights.

• Having diverse networks reduces physical inactivity.

• Having family and friends who exercise reduces physical inactivity.

• Living in areas with physical activity facilities reduces physical inactivity.

• Built physical activity resources play no role for social isolates.