Influence of individual and social contextual factors on changes in leisure-time physical activity in working-class populations: results of the Healthy Directions–Small Businesses Study

Abstract

Background

As part of the Harvard Cancer Prevention Program Project, we sought to address disparities reflected in social class and race/ethnicity by developing and testing a behavioral intervention model that targeted fruit and vegetable consumption, red meat consumption, multivitamin intake, and physical activity in working-class, multiethnic populations.

Methods

This paper examined the associations between change in leisure-time physical activity and individual and social contextual factors in participants employed in small businesses (n = 850) at both baseline and at 18-month final.

Results

In bivariate analyses, age, language acculturation, social ties, and workplace social capital were significantly associated with physical activity at final. In multivariable analyses, being younger and having high language acculturation were significantly associated with greater leisuretime physical activity at final; high workplace social capital was significantly associated with a decline in physical activity at final.

Conclusion

These findings have implications for understanding factors that are integral to promoting change in physical activity among working-class, multiethnic populations.

Introduction

Eliminating health disparities among racial/ethnic minorities and low socioeconomic groups represents a profound public health challenge. Social epidemiological findings show that these populations are more likely to engage in unhealthy behaviors, such as poor diet, tobacco use, and physical inactivity. For example, the prevalence of meeting recommended levels is lowest among those at a lower income and education and among blacks and Latinos [1–3]. These disparities are important because unhealthy behaviors are associated with increased risk for many health conditions such as cardiovascular disease, diabetes, obesity, and some cancers [4]. However, disparities are also socially patterned, such that those with greater access to social, educational, and economic resources experience better health than those with less access [5]. For example, few personal and social resources, such as limited social support and social and community ties (i.e., churches, schools, etc.), is associated with specific health conditions such as hypertension and cancer and increased mortality [6, 7]. Similarly, the social context of one’s neighborhood environment impacts behavior patterns. Exposure to environmental hazards (e.g., tobacco retail outlets, areas of economic deprivation; low walkability areas; fast food outlets) is associated with tobacco use, physical inactivity, and poor diet [8–13]. Strategies that address social contextual determinants of health-promoting behaviors among racial/ethnic minorities and those with low socioeconomic position are necessary to redress health disparities.

Few behavioral interventions have been specifically designed for socially and economically disadvantaged populations. In a previous study, we sought to address disparities reflected in social class and race/ethnicity as part of the Harvard Cancer Prevention Program Project by developing and testing a behavioral intervention model that targeted fruit and vegetable consumption, red meat consumption, multivitamin intake, and physical activity in a working-class, multiethnic population employed in small businesses (i.e., the Healthy Directions–Small Business Study) [14] and in health center patients (e.g., HealthyDirections–Health Centers Study) [15]. In brief, the intervention was designed to promote healthy behaviors by focusing on social contextual factors that could be explicitly changed by the intervention (e.g., social norms for physical activity) in addition to those that could not be changed directly but are important determinants of behavior (e.g., material circumstances such as access to a car), and delineated pathways through which sociodemographic characteristics (e.g., income, race/ethnicity, and acculturation) may be related to and influence health behaviors. By explicating these pathways, we were able to design and test an intervention that considered the individual and social contexts of participants’ lives that are meaningful and relevant to the intended audiences.

Overall, we found that these interventionswere effective in changing some of the targeted health behaviors [14, 15]. Compared with the workers in control sites, workers in the intervention sites significantly increased their use of multivitamins and physical activity; in the health centers study, the intervention led to increases in fruit and vegetable consumption, multivitamin intake, and reduced redmeat consumption.

Conceptual model

Figure 1 depicts the conceptual framework we developed to explicate the role of the social context in health behavior change [5]. We identified and defined a set of modifying conditions, factors that independently influence health outcomes but are not expected to be influenced by behavioral change interventions. Mediating mechanisms, those factors that lie along the causal pathway between the intervention and the outcomes, were also defined. These mediating mechanisms are important to behavioral change and are potentially modifiable within the context of a planned intervention and within the context of specific health promotion channels (i.e., small businesses). Social context may function as either modifying conditions or mediating mechanisms, depending on their location within or outside the causal pathway between the intervention and the outcomes.

Fig. 1.

Conceptual model

We chose a social ecological framework to illustrate social contextual factors across multiple levels of influence [16–19]. Among these were individual factors, which include material circumstances such as owning one’s own car or having adequate resources for child care. Interpersonal factors, such as the presence of social ties, family roles and responsibilities, and social norms, are likely to be powerful correlates of health behaviors and may vary by factors reflecting cultural differences (race/ethnicity, acculturation). Organizational factors may reflect the work setting; for example, job stress and control, workplace social capital, and exposure to a hazardous work environment have been linked to tobacco use, depression, and physical inactivity [20]. Neighborhood and community factors measured on an individual level include access to a safe place to exercise. Prior research has shown that lower-income neighborhoods are less likely than higher-income neighborhoods to have access to health-promoting services such as physical activity facilities [21–23]. Finally, larger societal forces, such as racial discrimination, may also shape health behaviors and outcomes.

Social contextual factors in turn may influence health behaviors directly or indirectly through individual psychosocial factors. Social cognitive theory [24–27], the theory of reasoned action [28–30], and the transtheoretical model of behavior change [31–34] are guiding models that highlight specific individual psychosocial factors that predict a change in behavior. For example, self-efficacy, or the belief that one can perform a given behavior even in the face of challenges, has consistently been shown to be an important predictor of health behavior change. Psychosocial mediating variables in large part influence intentions to change behavior, which are highly associated with the likelihood of change [35, 36].

One’s social context and day-to-day realities are shaped by sociodemographic characteristics, which may influence a range of interrelated health behaviors. For example, socioeconomic position, race and ethnicity, nativity, gender, and age are important correlates of health outcomes. Identifying disparities in health behaviors across populations with these characteristics can inform priority-setting and guide policy decisions. In addition, culture, that is, learned and shared knowledge and beliefs used to interpret experiences, cuts across all domains in this model [37, 38].

Current study

A growing number of studies suggest that one’s social contextual environment may be important in determining health behavior outcomes, particularly physical activity [5, 39, 40]. Physical activity is a behavior that is intrinsically shaped by one’s social context in that most physical activity occurs within the bounds of families, communities, and neighborhoods [41]. A number of studies, for example, have shown that having social support and, to a lesser extent, many social ties are associated with increases in physical activity, especially among women and racial/ethnic minorities [36, 42–47]. Having multiple family roles such as financial and/or care giving responsibilities for small children or elderly family members [48–51] and neighborhood and organizational factors such as feeling unsafe in one’s neighborhood [52–54] and less social capital [55, 56] are negatively associated with physical activity, and thereby act to shape norms for physical inactivity. These factors may be particularly salient for lower-income racial/ethnic minorities who disproportionately live in environments that lack the necessary health-promoting resources for engaging in physical activity [21]. Taken together, social contextual factors are important parameters that help to either constrain or promote physical activity.

Using data fromthe “Harvard Cancer Prevention Program Project,” we examined the relationship between selected sociodemographic, individual, and social contextual factors and physical activity at baseline and at 18 months following the implementation of a health-promoting intervention. We also examined the extent to which these relationships were influenced by the intervention. Participants from the health centers had no improvements in physical activity, and as such, this paper focuses specifically on data fromparticipants in the Healthy Directions–Small Business Study.

Methods

This study used baseline and final data from the Healthy Directions–Small Business Study, a randomized controlled trial conducted between 1999 and 2003 that targeted health behaviors among employees in small businesses. Analyses for this study were performed in 2006. The unit of randomization and intervention was the workplace, and all workers were eligible to participate in intervention activities at a particular worksite. The unit of observation was a sample of workers at each worksite.

Study participants

Businesses enrolled in the study were identified using Standard Industrial Classification codes 20–39 (i.e., manufacturing codes) and were then limited to those located in and around the Boston area that employed between 50 and 150 workers. Eligible worksites included those sites that employed at least 25 % of workers who were first- or second-generation immigrants or racial/ethnic minorities, were in a position to decide whether to participate in the intervention study, and had a turnover rate of less than 20 % per year. Thirteen sites were randomized to the intervention group, and 13 sites were randomized to the minimal-intervention control group. Of the 26 worksites, the final was completed at 24 sites; one site was lost from each condition.

Data were collected from individual employees by interviewer- administered surveys at baseline and at 18-month final. Interviews were administered on company time in the language preferred by respondents. Individual workers were eligible for the survey if they were permanent employees, worked over 20 h a week, worked on-site, and spoke English, Spanish, Portuguese, or Vietnamese. Baseline and final surveys were conducted at approximately the same time of year to avoid seasonal differences in physical activity patterns. Of the 2,069 eligible employees, 1,684 completed the baseline survey (84 % response rate). Participation in the final survey was not contingent on participation in the intervention. The response rate in the 24 sites completing the study and final survey was 77 % (n = 1,408). An embedded cohort of 850 participants completed both baseline and final surveys; we present analyses from this group in order to track individual behavior changes.

Intervention

Details regarding the design of the intervention are presented more fully elsewhere [57]. The intervention strategies were based on (1) principles of employee participation and (2) a social contextual framework that targeted multiple levels of influence on behavior (i.e., individual/interpersonal, social contextual, and environmental/organizational), with special emphasis on low literacy levels and the unique and shared features of culture across ethnic groups [14]. Because the unit of intervention was the worksite, intervention activities were targeted and delivered at the worksite level rather than to individual workers. Over the 18-month intervention period, we delivered 1 monthly intervention activity focused on individual behavior change and made an average of 1 monthly contact with management environmental support and organization change in each of the 12 intervention worksites. The environmental/organizational interventions aimed to create a workplace environmental supportive of healthful eating, physical activity patterns, tobacco control, and reduction in hazardous occupational exposures through policy changes, that is, creating a smoke-free workplace and working with worksite management. Participants in both the minimal-intervention control groups and the intervention groups took part in smoking-cessation programs. There were an average of 16.9 events and 19.4 consultations contacts with management representations [57]. All questionnaires and intervention activities were translated into Spanish, Portuguese, and Vietnamese.

Measures

Physical activity

Leisure-time and occupational physical activities were assessed using a semi-quantitative physical activity questionnaire at baseline and 18-month final. Participants were asked to report how often in the last 4 weeks they engaged in eight types of leisure-time activities: walking for exercise; jogging; running; bicycling; aerobics or aerobic dancing; playing soccer, rugby, basketball, or lacrosse; playing baseball, football, bowling, or lifting weights; and other activities in which one gets at least a little bit out of breath. Occupational activity was assessed by asking participants how many hours per week they spent doing heavy physical activities at work, such as moving heavy equipment or lifting heavy boxes. Participants were asked to select from eight responses categories (never to more than 6 h per week), and responses were recoded to equivalent minutes per week and summed to yield total minutes or hours of physical activity per week. The semi-quantitative activity questionnaire was originally developed and validated in a population of nurses and in a separate population of health professionals [58, 59]. The questionnaire was modified based on a pilot test of a subsample of our study population and had good correlation (r = 0.53) with activity assessed via an accelerometer.

Sociodemographic characteristics

Participants were asked their date of birth, sex, highest level of education completed, and marital status. Those who said they were not married or living with a partner, but was living with at least one child under the age of 18, were classified as single heads of household. Participants were asked to identify all the racial/ethnic groups to which they belonged. Participants who reported being of Hispanic or Latino origin were coded into the Hispanic group, regardless of any other race or ethnic group noted. For the rest, those who reported only one race were categorized into that group (white, black, Asian, Native American). Respondents who selected more than one group were classified as having a mixed heritage. Because of small numbers in some categories, we combined Native Americans, blacks, and those of mixed heritage into a single category of other/mixed.

We also assessed household income, birthplace, acculturation, and occupational class. Household income was assessed in $10,000 increments, from less than $10,000 per year to $50,000 or more. We combined the responses to this item with the number of people supported by the income and the ages of household members in order to categorize respondents according to the federal poverty guidelines [60]. Participants were classified as being below the poverty level, above the poverty level but below 185 % of the guideline, or above 185 % of the poverty level. We used three categories to assess birthplace: participant born outside the United States, participant born in the United States who had one or both parents born outside the United States, and participant and both parents born in the United States [61]. In addition, according to the method described by Marin [62], participants were classified as low acculturated, moderately acculturated, and highly acculturated based on their scored responses to questions regarding language preference for reading, language spoken at home, and first language. Occupational class (manager status) was based on whether employees managed or supervised others [63, 64] according to information provided by the worksites.

Modifying conditions

We measured several social contextual characteristics that would likely influence physical activity in our sample.

Individual factors

We assessed four aspects of material circumstances: current financial situation, whether food ran out in the last year, access to a car, and the crowdedness of the respondent’s home. Respondents were asked to choose one of the four categories that reflect their current financial situation: “comfortable, with some extras,” “enough, but no extras,” “have to cut back,” and “cannot make ends meet.” They were also asked whether they had run out of food in the last year because they did not have money to buy more [65]. We also asked participants whether they own a car or whether one is regularly available to them. We asked about the number of rooms in their home and the number of people who live there. Households with three or more rooms per person were defined as a low crowding; medium crowding was defined as households with 1.5–3 rooms per person; and high crowding was defined as 1.5 rooms or fewer per person.

Interpersonal factors

We examined two dimensions of social networks: social ties and diversity of friendship patterns [66]. Social ties were assessed by four factors: whether participants have a spouse or partner, relatives they feel close to, close friends they have, and whether they are an active member of any groups or clubs. The social ties variable was computed as the number of these ties they have (0–4). As a measure of social network diversity, diverse friendship patterns impacts information transfer, attitude formation, and adoption of health behaviors [67–69]. To evaluate diversity of friendship patterns, we asked participants about the ethnicity of the participants’ close friends (i.e., diversity of friends), with five response categories ranging from “only my own ethnic group” to “only from other ethnic groups.”

Participants were also asked about the number of family roles they have. Response options were “earning money to support the family,” “taking care of children,” “food shopping and cooking,” “taking care of the home,” and “taking care of another household.” The measure of multiple roles was computed as the number of family roles that the respondent has most or all of the responsibility for (0–5).

Organizational factors

We asked participants about job strain and social capital at work. Job strain was assessed using items adapted from the Job Content Questionnaire [70]. We used two items to assess psychological job demand (job makes conflicting demands and job requires working fast): one item to assess participants’ authority to make decisions (lots of input about what happens on the job) and the other to assess skill discretion (job requires learning new things and job involves doing the same things over and over) [70]. Job control was computed as a weighted sum of decision authority and skill discretion. A worker was considered to be under job strain if his or her psychological demand was greater than the national median and his or her job control was below the national median. National medians were obtained [71, 72] and rescaled to adjust for the number of items used in our study. This variable was further dichotomized into yes/no for job strain.

To assess social capital at work, we assessed participants’ perceptions of support, reciprocity, and trust. We asked participants to report their agreement with three statements: “the people I work with are willing to help each other,” “I trust the people I work with,” and “the managers of this company look out for the people who work here.” Agreement with the three items was averaged; summary scores ranged from 1 to 3, with higher scores indicating greater social capital at work [73–76].

Neighborhood/community factors

We asked respondents to rate the safety of walking in the daytime in their neighborhood as “safe,” “a little unsafe,” or “not at all safe” [77].

Societal factors

Participants were asked whether they had experienced discrimination in their lifetime. Responses were grouped into five categories: those who experienced discrimination never, a few times, sometimes, fairly often, and all the time. [78–80].

Mediating mechanisms

Social support

A single item was used to assess three of the four domains of social support based on the Inventory of Social Supportive Behaviors [81]. Three domains—emotional, instrumental, and financial support—were assessed by asking participants one question about the support available to them from each domain. Participants responded to each item as “always available,” “sometimes available,” “available but wouldn’t accept help,” “don’t need help,” and “not available.” The responses were combined, and the social support scale score was computed as the number of domains always or sometimes available to the participant, with a range from 0 to 3, excluding “don’t need help” and “available but wouldn’t accept.”

Social norms for physical activity

We used one item to assess social norms for physical activity [82]. Participants were asked to state how many of their coworkers engage in at least a half-hour of exercise each day. Response options were few or none, some, most or all, or don’t know.

Motivation to perform behavior

Motivation to change was assessed by asking participants whether they were planning to be physically active within the next 30 days or 6 months. Responses were categorized into pre-contemplation, contemplation, and preparation [83].

Self-efficacy

Self-efficacy was assessed by asking participants one question: “If you thought you needed to improve at least one of the target health habits, e.g., physical activity, how sure are you that you could do something about it in the next 30 days?” Response categories were “sure,” “maybe,” and “not at all sure.”

Statistical analysis

To evaluate the effect of the variables in the model on leisure-time physical activity, we computed a least squares linear regression analysis of physical activity at the final survey based on baseline characteristics of the respondents. We analyzed the two samples separately and used data from the respondents who completed both the baseline and final surveys, that is, an embedded cohort. In all analyses, we controlled for baseline physical activity as well as the design effects, intervention condition, and unit of randomization (worksite) [84].

We first tested the association between each baseline variable alone and leisure-time physical activity at final, controlling for baseline activity, intervention group, and randomization unit. We then did a multivariable analysis by selecting all variables that were individually significantly associated with physical activity at final. To determine the effect of intervention on change, we tested the interaction between statistically significant main effects and the intervention group. We also tested all pairwise interactions between significant main effects. Each interaction of interest was added to the main effects model alone.

We report the slope coefficients from the resulting analyses. This represents the change in final physical activity per unit change in the explanatory variable, holding baseline activity constant. For categorical explanatory variables, we report the statistical significance test that all slopes for that variable are 0. All tests were Wald tests based on the type III sums of squares [84]. p values at the 0.05 level were considered statistically significant.

Results

Participant characteristics and change in physical activity at final

Overall, 65.8 % of the participants were men, 46.5 % had a high school education or less, 32.3 % were foreign-born, and 82.7 % were employed in non-managerial positions; mean age was 44 years (standard deviation = 11 years). Other participant characteristics are shown in Table 1. Findings from the trial overall found that an increase in physical activity from baseline to 18-month final was significantly greater in the intervention group (64 % baseline; 72 % final) compared with the control group (76 % baseline; 66 % final) when controlling for poverty (p = .02) [14]. However, in this embedded cohort of 850, workers in the intervention worksites increased their physical activity an average of 0.31 h per week, compared with a decrease of 0.30 h per week among workers in the control worksites, resulting in an overall increase of 0.6 h per week more in the intervention group than the control group (p = 0.19). Consequently, although there was not a significant increase in physical activity in the embedded cohort, there was a change in physical activity, and this paper examines the factors associated with change at final.

Table 1.

Selected characteristics of study participants at baseline (n = 850): Healthy Directions–Small Business Study

Characteristic	n	%
Education
HS or less	449	46.5
Some post-HS	328	34.0
BA/BS or more	189	19.6
Manager status
Non-manager	805	82.7
Manager	169	17.4
Poverty
Above 185 %	835	86.7
Above poverty, below 185 %	102	10.6
Below poverty	26	2.7
Race/ethnicity
Asian/Pacific Islander	65	6.7
Hispanic	107	11.1
Multi-racial/ethnic/other	128	13.3
White	666	68.9
Language acculturation
Low	160	18.8
Moderate	129	15.1
High	563	66.1
Place of birth
Not born in USA	313	32.3
Born in USA, parents not	90	9.3
Born in USA, parents too	567	58.5
Sex
Men	641	65.8
Women	333	34.2
Single head of household
No	944	96.9
Yes	30	3.1
Discrimination
Never	614	64
Only a few times in my life	242	25.2
Sometimes	83	8.7
Often	16	1.7
All the time	5	0.5
Current money situation
Comfortable	608	63.3
Enough, no extras	225	23.4
Have to cut back	101	10.5
Cannot make ends meet	26	2.7
Diversity of friends
Only own group	211	22.6
More own group	317	33.9
Half and half	354	37.9
More/all other groups	53	5.7
Multiple roles
Low	288	29.7
Medium	546	56.3
High	136	14.0
Job strain
No	698	73.3
Yes	254	26.7
Neighborhood safety
Safe	932	96.1
A little safe	29	3.0
Not at all safe	9	0.9
Crowding
Low	176	18.1
Medium	482	49.5
High	316	32.4
Access to car
No	73	7.6
Yes	894	92.5
Social norms
Few or none	188	20.0
Some	240	25.5
Most or all	119	12.6
Don’t know	396	42.1
Self-efficacy
No	280	29.7
Yes	664	70.3
Motivation
Pre-contemplation	196	20.6
Contemplation	77	8.1
Preparation	680	71.4
	Mean	SD
Age (years)	44	11
Social capital	2.8	0.5
Social ties	3.3	0.7
Social support	5.0	1.5

Bivariate model

Table 2 presents the bivariate association between each sociodemographic, individual and social contextual variable at baseline and leisure-time physical activity at final, controlling for physical activity at baseline, intervention condition, and randomization unit. Gender was not associated with a change in physical activity, so the analyses were not gender stratified. Findings show that increasing age (p =<.01) was significantly associated with less improvement in physical activity, while having high language acculturation (p = .02), having more social ties (p = .04), and high workplace social capital (p = .03) were significantly associated with increased physical activity at final.

Table 2.

Effect of each variable on change in leisure-time physical activity hours per week at final, controlling for baseline activity, intervention condition, and randomization unit

	n = 850
	Slope	p value
Age (+10 years)	−0.5	<0.01
Education		0.15
HS or less	−0.54
Some post-HS	0.00
4-year college or more	0
Manager status		0.32
No	−0.35
Yes	0
Poverty		0.20
Above 185 %	1.52
Above poverty, below 185 %	1.62
Below poverty	0
Race/ethnicity		0.12
Asian/PI	1.21
Hispanic	−0.25
Mixed/other	−0.28
White	0
Language acculturation		0.02
Low	−0.39
Medium	−1.13
High	0
Place of birth		0.32
Not born in USA	−0.44
Born in USA, parents not	0.09
Born in USA, parents too	0
Sex		0.42
Male	−0.25
Female	0
Social ties (+1 unit)	0.31	0.04
Diversity of friends		0.65
Only own group	−0.01
More from own	−0.40
Half and half	−0.39
Only/more from other groups	0
Multiple roles		0.82
Low	−0.27
Medium	−0.16
High	0
Single head of household		0.56
No	0.45
Yes	0
Neighborhood safety		0.73
1 = safe	1.26
2 = a little unsafe	1.20
3 = not at all safe	0
Crowding		0.88
1 = low	0.20
2 = medium	0.06
3 = high	0
Access to a car		0.53
No	−0.35
Yes	0
Discrimination		0.20
Never	−0.46
Few times	−0.47
Sometimes	0.45
Often	1.20
All the time	0
Social norms		0.14
Few or none	0.54
Some	0.81
Most or all	0.48
Don’t know	0
Support (+1 unit)	0.01	0.90
Self-efficacy		0.15
No	−0.41
Yes	0
Motivation to change		0.79
Pre-contemplation	0.09
Contemplation	−0.29
Preparation	0
Job strain		0.66
No	0.14
Yes	0
Social Capital (+1 unit)	−0.63	0.03
Occupational activity (+1 h/week)	−0.01	0.44

Bold values represent statistically significant p-values

The slopes for the categorical variables presented in the table represent the difference in change in hours of physical activity between each category and the reference group. For example, those with medium language acculturation were less active at final reporting −1.13 fewer hours per week of physical activity and those with low language acculturation reported −0.39 fewer hours per week than those with high language acculturation at final. Number of social ties was associated with 0.31 h more of physical activity at final and social capital at work was associated with a decline of physical activity at final (-0.63 h). No other social contextual variables were associated with a change in physical activity in bivariate analyses.

Multivariable model

To select additional adjustment variables for the multivariable model, we first identified all variables that had p values less than or equal to 0.20. The main effects of model included age, education, poverty, race/ethnicity, language acculturation, social ties, self-efficacy, and workplace social capital. To this model, we added physical activity at work at baseline, which was not associated with physical activity at final when other variables were controlled. We then determined which effects had a significant interaction (Table 3). Only those significant interactions are presented. In comparing the intervention group with the control group, the average increase in leisure-time physical activity from baseline to final was about 20 min more per week (0.34 h) and was 45 min higher in the intervention group than the control group (0.76 h).

Table 3.

Multivariable effects of covariates on change in hours of leisure-time physical activity at final

	SB n = 653
	Slope	p value
Intervention condition		0.09
Intervention	0.76
Control	0
Baseline PA by race		0.02
Asian	0.19
Hispanic	1.12
Mixed/other	0.38
White	0.51
Education		0.67
Les than HS	−0.15
Some post-HS	0.16
BA/BS or more	0
Poverty		0.19
Above 185 %	1.73
Above poverty	1.70
Below poverty	0
Self-efficacy		0.22
No	−0.39
Yes	0
Age(+10 years)	−0.43	<0.01
Language acculturation		0.02
Low	−0.63
Moderate	−1.35
High	0
Social ties (+1 unit)	0.36	0.08
Social Capital	−0.80	0.01

Bold values represent statistically significant p-values

Although the association between race and physical activity was not significant in the bivariate analyses, the interaction of race/ethnicity and baseline activity was significant in the multivariable analyses (p = .02). For Hispanics, 1 h more of activity per week at baseline was associated with more than 1 h more activity per week at final. For all other participants, 1 h more of baseline physical activity per week was associated with between 11 and 30 min more physical activity per week at final. As with the bivariate analyses, in the multivariable analyses, change in physical activity was greater for those who were younger (p =<.01), more acculturated (p = .02) and reported low social capital at work (p = .01). No other social contextual variables were associated with a change in physical activity in multivariable analyses (data not shown).

Discussion

This study demonstrated a relationship between specific sociodemographic, individual and social contextual factors and a prospective change in physical activity in a working-class, multiethnic population. Many cross-sectional studies have also found associations between the factors we studied and physical activity; however, analyses conducted in this paper attempted to understand social contextual factors that are instrumental in promoting a change in physical activity behavior by examining these relationships prospectively. Our findings demonstrate that few social contextual factors were independently associated with improvements in physical activity.

Age and low language acculturation were associated with a decline in physical activity at final. The findings on age are generally consistent with studies that show declines in physical activity with increasing age [1, 85]. In our multivariable analyses, low and medium language acculturation were inversely associated with a change in physical activity at final, representing 25–68 fewer minutes of physical activity per week, respectively, compared with those with high language acculturation. In our working-class population, more than one-third of the participants were born outside the United States or spoke a language other than English. These findings are similar to those of other cross-sectional studies, which show that low language acculturation is negatively correlated with leisure-time physical activity [86–90]. Several mechanisms between acculturation and physical activity are likely at play including lack of time for physical activity due to job demands/work schedules among those less acculturated, or economic and social barriers, that is, small social networks comprised of English-speaking or physically active members [88]. It has also been suggested that one of the likely reasons for the relationship between acculturation and physical activity is physical activity messages from the media [86]. Those who can read or understand English are more likely to benefit from print and television messages encouraging physical activity. Although our surveys and intervention materials were in English, Spanish, Portuguese, and Vietnamese, our findings suggest a different response to the intervention based on one’s level of acculturation and show that non-English speakers have difficulty in increasing their levels of physical activity. Interestingly, the greatest differences were between the most acculturated and the moderate acculturation group. Some evidence suggests that moderately acculturated individuals have poorer health outcomes [91–93]. Moderately acculturated individuals may represent a unique group that is caught between two cultures and may have difficulty in balancing both as it relates to engaging in physical activity.

Interestingly, income, education and race/ethnicity, and common predictors of physical activity were not independently associated with a change in physical activity. Although not statistically significant in either bivariate or multivariable models, being above the poverty level was associated with a positive change in physical activity in both bivariate and multivariable analyses. Those who were above the poverty level engaged in more than 1.5 h more activity per week at final compared with those below the poverty level; this is equivalent to roughly four 30 min bouts of activity per week. The lack of statistical significance may be due to the small number of participants who were below the poverty level in our sample of workers. Race/ethnicity was not associated with change in physical activity at baseline; however, there was a significant interaction between race/ethnicity and baseline activity at final. For everyone, engaging in at least some physical activity at baseline resulted in an increased change in physical activity at final. This finding highlights the importance of getting sedentary racial/ethnic minorities, particularly Hispanics, to begin a physical activity program, so they can benefit from future physical activity intervention and health promotion efforts.

When examining social contextual predictors, we found that having more social ties and social capital at work were associated with physical activity at final—positively and negatively, respectively; however, in multivariable analyses, the significant relationship between social ties and physical activity became marginally significant (p = 0.08). Social support was not associated with physical activity, but the number of people one reported feeling close to and participation in social activities were associated with increased activity. Few studies have evaluated the role of adult social ties in behavioral change interventions, and therefore, less is known about the association between the size, density, and uniformity of one’s social ties for physical activity promotion. Future work should examine social ties in worksite settings.

Our intervention capitalized on the social interaction and environmental support of the worksites to promote physical activity among workers; this was particularly effective for increasing physical activity among non-managers [14]. It is therefore surprising that reporting high workplace social capital was associated with a decline in physical activity at final. The literature has pointed out that there may also be a potential downside to social capital in that under certain circumstances it may diffuse unhealthy behaviors, such as preference for sedentary behaviors among group members [94]. It is possible that these individuals were less responsive to the intervention for this reason. Also, those who report high social capital at work may also be more involved in the workplace, that is, organizational activities, longer hours, thereby reducing time for physical activity participation [95].

It is unclear why we did not find stronger associations between many of the social contextual variables and change in physical activity, as we have found for fruit and vegetable consumption in this sample [96]. As stated earlier, associations between these variables and physical activity have mainly been conducted using cross-sectional study designs signaling the need to better understand their influence on intervention uptake and change over time. Also, some of our variables, particularly organizational factors, were specific to the work environment. For example, among studies that demonstrate an association with social capital and behavior, most have done so in the context of the neighborhood environment [97–99]; fewer studies have examined social capital at work as we did in this study [74, 100].

This study had several strengths and limitations. Because participants in both the intervention and control conditions reported high levels of physical activity at baseline (5.32 h/week), which were higher than expected and higher than national estimates for physical activity, only small increases in physical activity were achieved likely due to a ceiling effect for individuals who were already engaging in the recommended amount of physical activity. We conducted a validation study from a subsample of respondents comparing self-report findings of physical activity with objective measures of physical activity using accelerometers and found similar estimates [14]. While we used a wide array of items to measure social contextual factors, in all cases, the measures were not used in their validated form. This decision was juxtaposed with the desire to reduce respondent burden, as there were over a dozen constructs being measured in the study, and the overall goal was to enhance intervention effects. There are a growing number of studies examining the influence of social environmental factors on physical activity using cross-sectional data [39]; however, we were specifically seeking to understand factors that influence change in activity. Notwithstanding the limitations, we developed a theoretically and empirically based conceptual model that outlined the influence of individual and social contextual factors on behavior that demonstrated validity in better understanding change in physical activity.

In conclusion, our study findings illustrate the importance of sociodemographic and social contextual factors in promoting physical activity change. In our study, we found that age, language acculturation, and workplace social capital were strong determinants of a change in physical activity among working-class, multiethnic adults. This study explicitly targeted the social context as an important determinant of behavior change and was successful in increasing physical activity in this population, although this difference was not statistically significant in the embedded cohort. The social contextual model of behavior change developed for this intervention provides a useful tool for designing and testing interventions for working-class, multiethnic populations; future studies should examine many of these social contextual factors longitudinally in order to increase our knowledge of their influence on behavior change.