Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?

Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?

Social Science & Medicine xxx (2014) 1e10 Contents lists available at ScienceDirect Social Science & Medicine journal homepage: www.elsevier.com/loc...

362KB Sizes 0 Downloads 2 Views

Social Science & Medicine xxx (2014) 1e10

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Questions of trust in health research on social capital: What aspects of personal network social capital do they measure? Richard M. Carpiano*, Lisa M. Fitterer Department of Sociology, University of British Columbia, 6303 Northwest Marine Drive, Vancouver, British Columbia V6T 1Z1, Canada

a r t i c l e i n f o

a b s t r a c t

Article history: Available online xxx

Health research on personal social capital has often utilized measures of respondents’ perceived trust of others as either a proxy for one’s social capital in the absence of more focused measures or as a subjective component of social capital. Little empirical work has evaluated the validity of such practices. We test the construct validity of two trust measures used commonly in health research on social capitaldgeneralized trust and trust of neighborsdwith respect to measures of people’s general network-, organization-, family-, friend-, and neighborhood-based social capital and the extent to which these two trust measures are associated with self-rated general health and mental health when social capital measures are included in the same models. Analyses of 2008 Canadian General Social Survey data (response rate 57.3%) indicate that generalized trust and trust of neighbors are both positivelydyet modestlydassociated with measures of several domains of network-based social capital. Both trust measures are positively associated with general and mental health, but these associations remain robust after adjusting for social capital measures. Our findings suggest that (a) trust is conceptually distinct from social capital, (b) trust measures are inadequate proxies for actual personal social networks, and (c) trust measures may only be capturing psychological aspects relevant todbut not indicative ofdsocial capital. Though links between perceived trust and health deserve study, health research on social capital needs to utilize measures of respondents’ actual social networks and their inherent resources. Ó 2014 Elsevier Ltd. All rights reserved.

Keywords: Canada Social capital Trust Social networks Theory Measurement Social determinants of health

The concept of social capital is credited as one of the most popular social science imports into public health (Kawachi, 2010). Since gaining popularity in health research more than a decade ago, the application of social capital to studying social determinants of health has resulted in the emergence of two theoretical perspectives. The social cohesion perspective, inspired by the scholarship of Coleman (1988) and Putnam (2000), conceptualizes social capital as the presence of trust, norms of reciprocity, and sanctions available to members of a group for influencing health (Kawachi, 2010). Empirical health research using this perspective often employs survey measures of perceived trust and reciprocitydto create individual- and community-level social capital indicators (see Kawachi, 2010). By contrast, the network perspective, influenced by the scholarship of Bourdieu (1986), emphasizes the health implications of actual or potential material, informational, and psychosocial resources rooted within the networks to which individuals

* Corresponding author. E-mail address: [email protected] (R.M. Carpiano).

are embedded. Empirical health research using this perspective often utilizes measures of individuals’ social ties and the extent that those ties give people potential access to various resources (e.g., Carpiano and Hystad, 2011; Moore et al., 2009). Reflecting these conceptual distinctions, some health scholars studying personal social capitaldi.e., the social capital that an individual possesses or has access via personal tiesdhave understandably aimed to be comprehensive in health surveys and included measures often categorized into two areas: “cognitive social capital,” which refers to people’s subjective values and perceptions and is often assessed using, among other items, attitudinal measures such as perceived trust of others in general and/or of one’s neighbors and “structural social capital,” which refers to what people do and is assessed using measures of network ties and group/organization participation as well as levels of engagement in religious and civic activities (e.g., Krishna and Shrader, 1999; Harpham, 2008; Hyyppä and Mäki, 2001; De Silva et al., 2005; Harpham et al., 2002; Ziersch et al., 2005). However, even when health studies are not explicitly aiming to assess cognitive social capital, measures of personal trust are the most commonly utilized measures of social capital, with numerous studies reporting

http://dx.doi.org/10.1016/j.socscimed.2014.03.017 0277-9536/Ó 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

2

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

correlations with health outcomes (Abbott and Freeth, 2008; Kim et al., 2008). These personal trust measures most commonly used are ones that assess two key conceptual domains of trust: generalized trust of others and particularized trust of specific persons, such as neighbors. Nevertheless, there is a fundamental question that is rarely acknowledged in health research on personal social capital: Is trust really social capital? Though health research has often classified trust as a component of social capital and measured it as such, many social capital scholars have presented arguments for a conceptual distinction between trust and personal social capital (e.g., see Portes, 1998; Lin, 2001; Glaeser et al., 2002; Field, 2003). Consistent with these debates, one review of social capital concluded that “one cannot assume that social trust is necessarily a product of social networks,” calling for further research on this matter (Policy Research Initiative, 2003, p. 9). But even if health researchers wish to remain agnostic about these debates and conceptually inclusive in selecting measures, it is important to recognize that these conceptual debates constitute more than academic disagreements of opinion. These concerns with theory translate into important concerns regarding the construct validity of measuresdparticularly when health studies only utilize trust measures as indicators of personal social capital. At the very least, such debates beg the empirical question: even if trust measures are simply proxies for social capital when more precise measures are unavailable in a dataset, do measures of perceived trust used commonly in health researchdnamely, generalized trust and particularized trust (of specific people)d adequately capture aspects of a person’s real life social relationships (or network ties) and their inherent resources that matter for one’s health? This measurement issue has received little attention in health research. The present study aims to contribute to this debate regarding trust as a measure of personal network social capital. Using Canadian national survey data, we evaluate the construct validity of two perceived trust measures used commonly in health research on social capitaldgeneralized trust and trust of neighbors. Specifically, we investigate the extent to which these measures are associated with (a) measures of several domains of actual personal social capital and (b) self-rated general and mental health once these social capital measures are included in the same models. 1. Background 1.1. Social capital In evaluating the validity of trust measures as indicators of social capital, we explicitly take a position supported by prior scholarship that personal social capitaldthe resources that one has access to via possessing social networksdis conceptually distinct from trust (e.g., Lin, 2001; Field, 2003; Glaeser et al., 2002). Individual social capital can exist without trust and possessing trust does not necessarily mean that one has social capital. Nevertheless, some social capital scholarship has included trust as a social capital component. Below, we provide a brief overview of social capital theory, focusing on how key scholars considered trust in relation to social capital. The seminal theoretical scholarship on social capital is typically attributed to Bourdieu (1986) and Coleman (1988). Bourdieu (1986, p. 248) defined social capital as: “the aggregate of actual or potential resources linked to possession of a durable network of more or less institutionalized relationships of mutual acquaintance and recognitiondor, in other words, to membership in a group.”

By contrast, Coleman (1988, p. S98) proposed that social capital is: “defined by its function. It is not a single entity but a variety of different entities, with two elements in common: they all consist of some aspect of social structures, and they facilitate certain actions of actorsdwhether persons or corporate actorsdwithin the structure. To Coleman, social capital exists in the relations among persons and derives its value from the function of the social structure that, in turn, is the resource that people use to achieve a desired outcome. Both theories consider social capital as resources linked to social networks, but differ regarding what constitutes “resources.” Whereas Bourdieu identifies economic, cultural or symbolic resources, Coleman’s resources include trustworthiness of the social environment, obligations, information, norms and effective sanctions. Of these two works, only Coleman’s scholarship explicitly mentions the word “trust.” Bourdieu’s thoughts on trust are the subject of debate (Field, 2002). Nevertheless, Coleman’s more abstract social capital definition and inclusion of many processes (some potential antecedents and consequences of social capital) has been argued to be the catalyst for later scholars classifying many different concepts under the label “social capital” (Portes, 1998). Coleman’s theory was particularly influential for Putnam (1995, p. 67), who expanded the social capital concept to describe collective feature of communities and civic life. Like Coleman, Putnam (1995, p. 67) included different processes under the term “social capital,” defining it as “features of social organization, such as networks, norms, and social trust, that facilitate coordination and cooperation for mutual benefit.” Though Putnam’s definition evolved in his later work to subsume trust (i.e., trustworthiness) under the broader conception of norms (Putnam, 2000, p. 19; Field, 2002), he reported survey-based trends in generalized trust as evidence for temporal declines in social capital (Putnam, 2000). Putnam’s theory has particular importance for the present discussion on health research. Other social capital theorists produced well-cited work that considered trust as an antecedent, component, or consequence of social capital (see Field, 2002). Putnam’s theory, however, became the most commonly cited within public health research (Moore et al., 2006) and arguably inspired health researchers to use perceived trust measures as indicators of social capital at not only aggregate (e.g., neighborhood, state) levels of analysis (see Kim et al., 2008), but, germane to the present study, the individual level, as either (a) lone indicators of personal social capital or (b) components of the cognitive social capital subconstruct. Despite this extensive focus on trust as a component of individual social capital, Abbott and Freeth (2008, p. 874) contend that, within the health literature, “Trust is usually treated simplistically, both conceptually and in relation to measurement.” Their suggestions for overcoming these limitations require attention to the nonhealth trust literature, which provides important insights about what different dimensions of trust and their measures maydand may notdtell us about personal social capital and its relation to health (see also Veenstra, 2002). 1.2. Trust and its relationship to social capital Though discussed and debated as a social capital component, the multidimensional concept of trust has been its own focus of study for several decades across the social sciences. Trust is defined as “the expectation of good will in others” (Glanville and Paxton,

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

2007) and is primarily conceptualized in three ways: generalized, particularized, and strategic (Smith, 2013). We focus on generalized and particularized trust, which are the most commonly used trust concepts in health research on social capital. 1.2.1. Generalized trust The conceptualization of trust receiving greatest attention in the social sciences is generalized trust (see Nannestad, 2008; Smith, 2013). Defined as “a belief in the benevolence of human nature in general and thus [.] not limited to particular objects” (Yamagishi and Yamagishi, 1994, p. 139), generalized trust concerns an individual’s evaluation of the trustworthiness of the average person and is not focused on a specific person (Glanville and Paxton, 2007). Thus, it is considered an internal characteristic or disposition and does not depend on reciprocity or evidence of another’s trustworthiness (Uslaner, 2002; Smith, 2013). Though generalized trust has been theorized to be either innate or learned early in life, evidence has supported a social learning theory-based explanation (Rotter, 1980) of generalized trust, whereby one’s localized experiences of trust, developed from interactions with specific groups or people, are extrapolated to form one’s generalized trust (Glanville and Paxton, 2007). It is often measured by a standard question asking respondents, “Generally speaking, would you say that most people can be trusted or that you cannot be too careful in dealing with people?” This item has been argued to have decent reliability and validity, but debate exists regarding whether it is predictive of trusting behavior (see Nannestad, 2008; Glaeser et al., 2000; Anderson et al., 2004). Substantively, generalized trust has been a focus of social science research because of its hypothesized potential for enabling people to connect in political and social spheres with people unlike themselves (Igarishi et al., 2008; Uslaner, 2002). Such connections are analogous to the term “bridging ties” in the social capital literature (e.g., Putnam, 2000), which have the potential for facilitating access to resources outside of one’s in-group. But is generalized trust associated with social ties? For informal social networks, studies have examined types of ties and network structure. Fischer’s (2005) analysis of the 1972e 2000 US General Social Surveys revealed that respondent trust of most people was (after controlling for age, education, marital status, and race) only weakly correlated with several common social capital indicators, including getting together with neighbors (.04) and friends living outside the neighborhood (.05). For network structure, findings from a cross-cultural ego-centric network analysis indicated that higher generalized trust was consistently related to higher density of connectedness among one’s network members (Igarishi et al., 2008). Another ego-centric network study, however, found generalized trust was not associated with having a confidante to discuss important matters (Moore et al., 2011). For formal group membership, several studies have reported positive associations with generalized trust (Brehm and Rahn, 1997; Claibourne and Martin, 2000; Wollebaek and Selle, 2002; Uslaner, 2002; Paxton, 2007). Some of these have also examined reciprocal associations, finding that the connection from participation to trust was either stronger than the reverse (Brehm and Rahn, 1997) or the only direction (Claibourne and Martin, 2000). Nevertheless, some studies have reported rather weak associations (Claibourne and Martin, 2000; Uslaner, 2002; Fischer, 2005) Uslaner (2002) found few associations between generalized trust and 20 measures of civic engagementdof which generalized trust was reciprocally associated with both charitable and volunteering acts, while trust influenced business, cultural, and ethnic group involvement. Overall, this collective evidence suggests that the generalized trust measure, while assessing one’s psychological orientation to

3

the world, does not permit clear assumptions about respondents’ informal and formal social network connections and the actual or potential resources that might be inherent in such tiesdand matter for health. 1.2.2. Particularized trust Particularized trust and its measurement concern trust in specified persons, offering the potential to assess trust of people who are part of one’s “in-group.” Such ties are akin to what is sometimes referred to in the social capital literature as “bonding ties” (Putnam, 2000), which entail group homogeneity and can potentially provide exclusive resources to its members. Though distinct from generalized trust (Abbott and Freeth, 2008), particularized trust across many life domains is important for shaping a person’s generalized trust (Glanville and Paxton, 2007). Health research on social capital, which has a strong focus on local communities, often uses measures of personal perceived trust of neighbors (e.g., Fujiwara and Kawachi, 2008; Sapag et al., 2010; Moore et al., 2011). But does this measure capture neighborhood ties? Studies examining associations between trust of neighbors and social capital, though rare, provide some insights. A UK study found that trust in neighbors does not necessarily correspond with the number of neighbors one knows and the frequency that one talks with neighbors (Coulthard et al., 2002). However, a Canadian study found that trust of neighbors was significantly associated with having a confidante to discuss important matters located in one’s neighborhood; yet neither measure was directly associated with self-rated health (Moore et al., 2011). For formal associations, Kankainen (2009) found that, among a sample of Finns, a person’s number of association memberships was associated with greater trust of neighbors (see also Veenstra, 2002). Hence, some evidence exists that trust of neighbors is associated with social capital. 1.3. Trust in health studies of cognitive and structural social capital In addition to the abovementioned research on trust, health studies examining both “cognitive” and “structural” social capital offer the potential for further informing how generalized and particularized trust correlate with personal network-based social capital and health outcomes (as well as other cognitive social capital measures like reciprocity). Yet, many of these studies do not report the correlations between their cognitive and structural social capital measures. Such omissions raise construct validity questions regarding the extent to which these cognitive and structural domain-specific measures may actually be capturing the same underlying latent construct of social capital. However, psychometric evaluations conducted across different cultural settings offer insights into these issues. Factor analyses of Vietnamese and Peruvian data revealed that generalized trust loaded with other measures of cognitive social capital on a single factor, but structural social capital items had very weak loadings on this same factordthereby suggesting not only very weak correlations between generalized trust and personal network social capital measures, but also that these items are capturing distinct constructs (De Silva et al., 2006). Furthermore, Mitchell and Bossert’s (2007) factor analyses of an extensive range of measures of cognitive and structural social capital among respondents living in disadvantaged communities in Nicaragua revealed that (a) generalized trust was more highly correlated with a factor underlying informal networks (e.g., number of people that could provide actual resources) than it was with a factor containing numerous measures of particularized trust in specific persons while (b) trust of neighbors did not produce substantial loadings on any of the model factors. Therefore, generalized

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

4

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

trust may be capturing social capital more closely in some contexts than others.

Table 1 Descriptive statistics for dependent and key independent variables (N ¼ 17,769). N

1.4. The present study Informed by this prior literature, the present study focuses on two issues regarding generalized trust and particularized trust (of neighbors) measures of personal network social capital. First, if trust measures are either valid indicators of social capital or simply proxy measures for social capital, then our analyses should find that (a) these trust measures are each substantially associated with measures of one or more domains of social capital and (b) their respective associations with health outcomes weaken substantially once the measures of personal social capital are controlled in multivariate models. Second, examining trust of neighbors (as a measure of a particularized trust) in addition to generalized trust allows us further insight into evaluating convergent and discriminant validity of trust as a social capital indicator or proxy. If trust of neighbors is measuring social capital, then we should expect this trust measure to correlate more strongly with measures of neighborhood-based social capital than with measures of social capital rooted in more general types of networks. 2. Methods 2.1. Data and sample We tested these expectations by analyzing data from the 2008 Canadian General Social Survey (GSS) Cycle 22, which had a thematic focus on social networks. The 2008 GSS is a national crosssectional computer-assisted telephone survey of persons 15 years of age and older in all ten provinces of Canada, excluding full-time residents of institutions (response rate 57.3%). Specific details of the sampling design have been discussed extensively elsewhere (see Statistics Canada, 2010). We limited our analysis to the 19,739 respondents aged 18 or older, yielding an analytic sample of 17,769 respondents who had complete (non-missing) data for all variables in our analyses. We use population-based sampling weights in all our analyses to account for non-response and sampling design (Statistics Canada, 2010). 2.2. Measures Table 1 reports the coding and descriptive statistics for the trust, social capital, and health outcome variables. 2.2.1. Trust measures We examine two trust measures. Generalized trust was measured using a single item asking respondents “Generally speaking, would you say that most people can be trusted or that you cannot be too careful in dealing with people?” and was coded 0 ¼ “You cannot be too careful in dealing with people” and 1 ¼ “People can be trusted.” Trust of neighbors was measured using an item asking respondents whether they trust many of the people, most of the people, a few of the people, or nobody else in their neighborhood. This variable was coded 0 ¼ nobody/a few in the neighborhood and 1 ¼ many/most people in the neighborhood. The tetrachoric correlation (rt) between these two trust measures was .53, indicating that they are correlated, but empirically distinct. 2.2.2. Social capital measures Our social capital measures comprise two domains: general network social capital, which concerns social capital rooted in a

Perceived trust Generalized trust (people can be trusted) 8611 Trust of neighbors (trust many/most 10,737 people in the neighborhood) Health outcome variables Very good/excellent general health 8954 Very good/excellent mental health 10,739 General network social capital Network diversity, mean (SD); weighted mean (SD) 10.09 (4.52) Any group participation 11,643 Geographic-based social capital Close relatives in city/local community None 5285 1e5 9278 6e10 2208 Over 10 998 Close friends in city/local community None 2938 1e5 10,826 6e10 2898 Over 10 1107 People in neighborhood R knows None 809 A few 8352 Many 2541 Most 6067 People in neighborhood R knows well enough to ask for a favor None 2208 1e5 9611 6e10 3240 Over 10 2710

Weighted % 47.89 57.76

52.26 62.08 10.07 (4.45) 65.48

28.67 52.96 12.89 5.48 16.84 60.79 16.46 5.91 4.70 49.04 14.52 31.75 12.72 55.77 18.21 13.30

NOTE: SD ¼ Standard deviation.

respondent’s total network and that is not necessarily restricted to social ties in a specific locality, and geographic-based social capital, which measures network-based social capital located within the respondent’s city/local community and neighborhood. 2.2.2.1. General network social capital measures. Network diversity, the extent of accessibility that each respondent has to persons occupying different social positions, was assessed using a position generator (PG)da commonly used social capital instrument in sociological research that has also been used in health research (e.g., Moore et al., 2009; Carpiano and Hystad, 2011). The PG is useful for measuring general network-based social capital because it samples positions in a hierarchical (in this case, occupational) structure, rather than sampling interpersonal ties (Lin et al., 2001). Also, the PG in the GSS does not distinguish between family/friends and acquaintances, which enables broad assessment of different ties possessed by a respondent. The PG asked respondents if they know someone “by name and by sight and well enough to talk to” in each of 18 different occupational positions representing a variety of sectors and occupational prestige levels (e.g., farmer, social worker, police officer or firefighter, manager in sales, marketing or advertising, computer programmer, engineer, delivery or courier driver, nurse, and accountants or auditors). Network diversity was computed as the sum total of occupational positions for which a respondent reported knowing someone, thereby ranging from 0 to 18. Membership/participation in groups was computed as single dichotomous measure (any group membership ¼ 1; no group membership ¼ 0) using a series of questions that asked respondents if, in the past 12 months, they were a member of or participant of a union/professional association; political party/ group; sports/recreational organization; cultural, educational or hobby organization; religious-affiliated group; school group,

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

neighborhood, civic, or community association; service club or fraternal organization; any other group (for which the respondent was asked to list the total number of other groups). The prevalence of participation in each group ranged from approximately 3.2% (any other group) to approximately 30.1% (union/professional association). Also, due to the potential of membership/participation in a “school group, neighborhood, civic, or community association” to reflect geographically-specific (versus more geographically-diffuse) ties, we re-computed our dichotomous group membership variable with this specific item excluded. A cross tabulation of the two variables indicated that they are essentially the same measure. 2.2.2.2. Geographic-based social capital measures. For city/local community social capital, we created two measures that respectively focus on the total number of (a) close family and (b) friends in the respondent’s city/local community whom the respondent feels “at ease with, can talk to about what is on your mind, or call on for help.” Both measures were modeled as four dummy variables (1e5, 6e10, and over 10, with None as the referent) in order to be consistent with the neighborhood social capital measures described below. Neighborhood social capital was assessed using two items. The first item asked respondents “Would you say that you know most, many, a few or none of the people in your neighborhood?” We modeled this response scale as four dummy variables with “none” as the excluded referent category. The second item asked “About how many people in your neighborhood do you know well enough to ask for a favor?” Examples of favors were “picking up the mail, watering plants, shoveling, lending tools or garden equipment, carrying things, feeding pets when neighbors go on holiday, and shopping.” We modeled its four-point response scale (None, 1e5, 6e10, and over 10) as four dummy variables with “none” as the excluded referent. 2.2.3. Health outcome measures Our two health outcome variables are self-reported health and self-reported mental health (hereafter referred to as “general health” and “mental health,” respectively). Each variable is based on a single item that asked respondents to respectively rate (“in general”) their health and mental health using a five-point scale ranging from “excellent” to “poor.” Both variables were recoded into separate dichotomous measures where 0 ¼ good/fair/poor and excellent/very good ¼ 1 and their rt correlation was .72. 2.2.4. Control variables Our analyses control for an extensive range of sociodemographic factors and (self-rated) health now compared to five years ago. Table 2 details these items. Household income categories were computed using a formula utilized previously in analyses of Canadian national data (see Ross, 2002; Carpiano and Hystad, 2011), which accounts for the number of people in the household and produces five categories coded from “lowest” (the referent category) to “highest.” We also included a sixth category for respondents who had missing household income data. Sensitivity analyses (not presented here) revealed that this household member-adjusted income measure was a stronger predictor of health outcomes than the unadjusted income measure. Furthermore, the general pattern of results was the same for analyses that excluded cases with missing income data. For urban location, we used the urban/rural indicator in the GSS public use file, which originally coded each respondent’s residential location as: (1) “Large urban centres” (Census Metropolitan Area [CMA]/Census Agglomeration [CA] Area), (2) Rural and small town (non-CMA/CA), and (3) Prince Edward Island. We recoded this item as 0 ¼ nonurban (categories 2 and 3) and 1 ¼ urban.

5

Table 2 Descriptive statistics for sociodemographic and health-related control variables (N ¼ 17,769)d2008 Canadian General Social Survey. N Male sex Age 18e29 30e44 45e64 65 or older Education Less than secondary graduation Secondary graduation Some post-secondary education Diploma/certificate from community college or trade/technical school Bachelor’s or graduate degree Household income Lowest Lower-middle Medium Upper-middle Highest Missing income data Visible minority status Aboriginal status Immigrant Marital status Married/common-law Divorced/separated/widowed Single Length of residence in neighborhood Less than 1 year 1e5 years 5 years to less than 10 years 10 years and over Urban (versus non-urban) residence Health now compared to 5 years ago Worse than before Same as before Better than before

Weighted % 7671

49.24

2318 4665 7021 3765

20.80 27.88 35.73 15.59

3039 2551 2516 5107

14.95 14.16 15.80 28.21

4556

26.88

370 746 2490 4975 6220 2968 1435 642 2931

1.27 3.05 11.88 26.74 40.04 17.02 20.14 3.32 20.14

10,575 3693 3501

66.50 11.75 21.75

1262 4054 2905 9548 13,380

7.67 24.10 16.85 51.38 81.01

5362 8232 4175

28.98 47.09 23.93

2.3. Analyses Our analyses proceed in two steps. First, we investigate the convergent and discriminant validity of both trust measures with all personal network-based social capital variables. Because the two trust measures were substantially correlated, we utilized seemingly unrelated bivariate probit regression, a multivariate technique that enabled the joint estimation of generalized trust and trust of neighbors on the social capital variables. This model consisted of two equationsdone for each trust variabledthat specified, as independent variables, the social capital variables as well as controlled for sociodemographic and prior health confounding variables. The equation for generalized trust also controlled for trust of neighbors. Due to convergence problems, generalized trust was not included as a control variable for the equation estimating trust of neighbors. Nevertheless, this exclusion is consistent with prior theoretical and empirical research that finds that particularized trust is a foundation for generalized trust (Glanville and Paxton, 2007). Also, supplementary analyses (not shown) using a bivariate probit model, as well as binary probit and logistic models that controlled for generalized trust revealed results that were similar to what are reported here for trust of neighbors. Second, we investigate the personal social capital that potentially underlies the relationship between each trust measure and health. Because this analysis required the inclusion of both correlated trust measures as well as the correlated health measures, we utilized bivariate probit regression to simultaneously estimate equations for both health outcomes (general health and mental health) in a series of models that systematically include the trust and social capital

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

6

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

variables while controlling for sociodemographic factors. After presenting models for the association between each health outcome and (a) each type of trust, (b) both types of trust modeled together, and (c) only the social capital variables, we then estimate full models to determine the extent to which the initial trustehealth associations remain after controlling for the social capital variables. Diagnostic analyses (not shown) indicated that there were no substantial correlations between any of the variables that could present potential multicollinearity in our analytic models. Given the complex sampling design of the GSS, we utilized survey weights for all point estimates and computed standard errors using the 500 mean bootstrap weight variables provided in the GSS dataset (Statistics Canada, 2010). Analyses were conducted using Stata 13 (StataCorp LP, College Station, Texas). For each independent variable, we report bivariate probit slope estimates, as well as average marginal effects (AMEs) for the univariate (marginal) predicted probability that a dependent variable equals 1 [i.e., Pr(Yx ¼ 1)]. Compared to probit coefficients, the AME estimates better facilitate comparisons across models to interpret how much an initial association increases/decreases once other variables are included in these binary outcome models (see Mood, 2010). These AMEs can be interpreted as risk differences. For all models, we report p values .05 as statistically significant. 3. Results 3.1. Social capital correlates of general trust and trust of neighbors Table 3 presents results of the seemingly unrelated bivariate probit model that regresses generalized trust and trust of neighbors on the social capital variables while controlling for all confounding variables and (for the generalized trust equation) the trust of neighbors variable. These results indicate that both trust measures are significantly associated to a similar degree of magnitude with both measures of general network social capital: an approximate risk difference of .5% for each unit increase in network diversity and an approximate 5% risk difference for group participation versus no participation. However, the trust measures differ with respect to their correlations with other domains. Neither trust variables is significantly associated with close family. Close friends living in one’s city/local community shows stronger associations with generalized trust (5.3%e12.3%) versus trust of neighbors (3.1%e8.8%). All neighborhood-based social capital variables, however, show stronger associations for trust of neighbors (14.2%e30.2%). Overall, it is important to examine the magnitudes of these estimates, which collectively indicate that all of the associations for generalized trust and the majority of the associations for trust of neighbors were rather modest. Though both types of trust were significantly associated with at least one type of social capital, no probit estimate exceeded .878dthe association between trust of neighbors and having 10 or more (versus no) people in one’s neighborhood that one knows well. Furthermore, most probit estimates were less than .50 (including the highest estimate observed for generalized trust, which was .489). By comparison, probit estimates of .878 and .50 (a) respectively convert to odds ratios (ORs) of approximately 4.07 and 2.23 (computed by multiplying the probit coefficients by 1.6 to convert to logit coefficients [see Amemiya, 1981; Gelman, 2006], then exponentiating these products to obtain ORs) and (b) approximate rt correlations of .51 and .31 using the equation rt ¼ cos[p/(1 þ OR1/2)] (Bonett, 2007). Therefore, even though trust of neighbors showed moderate correlations with two of the three categories of knowing neighbors well enough to ask a favor (6e10, over 10 people), the collective evidence in Table 3 indicates that both trust measures are still

Table 3 Bivariate probit estimates and average marginal effects (marginal probabilities) for general trust and trust of neighbors by social capital variables (N ¼ 17,769)d2008 Canadian General Social Survey. Trust of neighbors Generalized trust Coeff.

Pr(Y ¼ 1) Coeff.

Social capital Social network diversity .016** .005** .013** Any group participation .143** .047** .138** Close relatives in city/local community None Referent Referent Referent 1e5 .002 .001 -.025 6e10 .066 .021 .007 Over 10 .041 .013 .041 Close friends in city/local community None Referent Referent Referent 1e5 .092** .031** .144** 6e10 .269** .088** .387** Over 10 .222** .073** .337** People in neighborhood R knows None Referent Referent Referent A few .057 .020 .093 Many .466** .162** .285** Most .574** .196** .281** People in neighborhood R knows well enough to ask for a favor None Referent Referent Referent 1e5 .399** .142** .221** 6e10 .764** .266** .458** Over 10 .878** .302** .489** Trust of neighbors Nobody/a few in the neighborhood Referent Many/most people in the neighborhood -.669** atanh (rho) [standard error] 1.205** [.338] rho .835

Pr(Y ¼ 1) .005** .050** Referent -.009 .002 .015 Referent .053** .142** .123** Referent .034 .104** .102** Referent .080** .167** .178** Referent -.239**

*p  .05; **p  .01; NOTE: Both models adjust for the control variables detailed in Table 2. Atanh(rho) is the Fisher’s z transformation of rho, which is the correlation of the error terms in the two equations. Atanh(rho) is directly estimated; rho is computed as the hyperbolic tangent of atanh(rho) (Buis, 2011).

empirically quite distinct from any of the social capital measures. While this is particularly the case for generalized trust, the trust of neighbors measure demonstrated stronger associations with social capital specific to neighborhood versus other group tiesdthereby indicating greater convergent and discriminant validity (Campbell and Fisk, 1959) with, respectively, social capital measures of similar and dissimilar domains. 3.2. Generalized trust, trust of neighbors, social capital, and health Next, we examine the extent to which both trust measures covary with social capital measures in regression models estimating the odds of very good/excellent general and mental health. Table 4 shows results for five bivariate probit models that simultaneously estimate each health outcome: Models 1 and 2 respectively test the health associations for generalized trust and trust of neighbors, Model 3 includes both trust measures, Model 4 tests health associations for only the social capital variables, and Model 5 includes all trust and social capital variables. All models control for the confounding factors in Table 2. For Models 1 and 2, respondents who respectively reported feeling that most people can be trusted (versus otherwise) and trusting many/most (versus nobody/a few) neighbors have significant and, in terms of magnitude, generally similar average marginal risk differences for reporting very good/excellent general health and mental health (approximately 5e7% higher probability). When both trust measures are simultaneously entered into Model 3, both trust items are still significantly associated with both health outcomes, but their estimates are comparably smaller to that observed in the prior models, which can be expected given their correlation reported earlier.

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

7

Table 4 Bivariate probit estimates and average marginal effects for very good/excellent general and mental health by trust and social capital variables (N ¼ 17,769)d2008 Canadian General Social Survey. 1 Coeff.

2 Pr(Y ¼ 1)

General health Generalized trust Cannot be too careful in dealing with people Referent Referent People can be trusted .155** .053** Trust of neighbors Nobody/a few in the neighborhood Many/most people in the neighborhood Social capital Social network diversity Any group participation Close relatives in city/local community None 1e5 6e10 Over 10 Close friends in city/local community None 1e5 6e10 Over 10 People in neighborhood R knows None A few Many Most People in neighborhood R knows well enough to ask for a favor None 1e5 6e10 Over 10 Mental health Generalized trust Cannot be too careful in dealing with people People can be trusted Trust of neighbors Nobody/a few in the neighborhood Many/most people in the neighborhood Social capital Social network diversity Any group participation Close relatives in city/local community None 1e5 6e10 Over 10 Close friends in city/local community None 1e5 6e10 Over 10 People in neighborhood R knows None A few Many Most People in neighborhood R knows well enough None 1e5 6e10 Over 10 atanh (rho) [standard error] rho

Referent .186**

Coeff.

Referent .200**

3 Pr(Y ¼ 1)

Referent .068**

Referent .066** Referent .202**

Referent .072**

4

Coeff.

Pr(Y ¼ 1)

Referent .103** Referent .165**

5 Coeff.

Pr(Y ¼ 1)

Referent .035**

Referent .096**

Referent .032**

Referent .056**

Referent .112**

Referent .037**

Coeff.

Pr(Y ¼ 1)

.007* .093**

.002* .031**

.006 .083**

.002 .028**

Referent .008 .012 .086

Referent .003 .004 .029

Referent .007 .015 .083

Referent .002 .005 .028

Referent .001 .053 .075

Referent .000 .018 .025

Referent .007 .029 .054

Referent .002 .010 .018

Referent .037 .073 .066

Referent .013 .024 .022

Referent .041 .050 .041

Referent .014 .017 .014

Referent .130** .175** .175**

Referent .044** .059** .059**

Referent .109* .133* .132*

Referent .037* .045* .044*

Referent .137**

Referent .048**

Referent .129**

Referent .045**

Referent .156**

Referent .055**

Referent .094**

Referent .033**

.006 .013

.002 .005

.005 .003

.002 .001

Referent .042 .097* .138*

Referent .015 .034* .048*

Referent .043 .095* .136*

Referent .015 .033* .047*

Referent .029 .145** .132*

Referent .010 .051** .046*

Referent .020 .119** .108

Referent .007 .041** .038

Referent .158* .252** .308**

Referent .058* .090** .110**

Referent .156* .232** .287**

Referent .056* .083** .102**

to ask for a favor

.788** [.019] .657

.786** [.019] .656

.785** [.019] .655

Referent Referent .008 .003 .045 .016 .110* .038* .788** [.019] .657

Referent Referent .012 .004 .004 .001 .067 .023 .784** [.019] .655

*p  .05; **p  .01; NOTE: All models adjust for the control variables detailed in Table 2. Atanh(rho) is the Fisher’s z transformation of rho, which is the correlation of the error terms in the two equations. Atanh(rho) is directly estimated; rho is computed as the hyperbolic tangent of atanh(rho) (Buis, 2011).

Model 4, which examines only social capital variables, indicates a more divergent pattern of findings across the two health outcomes. For general health, only network diversity, group participation, and knowing people in one’s neighborhood well enough to

ask them for a favor are associated with significantly higher risk differences of very good/excellent heath. By contrast, mental health is significantly associated with having six or more relatives or friends (versus none) that one feels close to who also live in one’s

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

8

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

city/local community, just knowing an increasing number of people in one’s neighborhood, and knowing more than 10 people in one’s neighborhood well enough to ask them a favor. Model 5, which includes all variables, shows that both trust variables remain significantly associated with general and mental health once the social capital variables are also included in the same model. Comparing marginal effects before and after controlling for the social capital variables indicates that the initial Model 3 marginal effects for generalized trust are only reduced by approximately 8.5% for general health and 6.3% for mental health. For trust of neighbors, the Model 3 marginal effects for general and mental health are respectively reduced by 34% and 40%. Collectively, this evidence suggests that, for both health outcomes, the social capital variables account for substantially more of the observed associations for trust of neighbors than they do for the general trust associations. Also, consistent with the trust findings reported in Table 3, the social capital variables that seem to have the most substantial change in estimates from Model 4 to Model 5 are the neighborhoodspecific social capital variables. To check the robustness of these changes in estimates across models, we conducted supplementary analyses (not shown) using the “KHB” method Stata program for comparing coefficients between two nested nonlinear probability models (see Kohler et al., 2011). Though this program could not perform the KHB method calculations from the results of bivariate probit models, we instead based the calculations on standard probit models for each health outcome that simultaneously included both trust measures and were specified in a similar manner to Models 3 and 5 above. The percent change in average marginal effects from those analyses almost exactly replicated the findings reported here and indicated that the degree to which any changes were simply due to scaling differences between models was minimal. 4. Discussion For well over a decade, the voluminous health research literature on social capital has commonly utilized attitude-based measures of generalized and particularized trust to measure personal social capitaldarguing (when critically evaluated at all) that trust is either a theoretical component of social capital or, at the very least, a proxy measure for actual social capital. Our study evaluated the validity of such claims for two commonly used measures of trustdgeneralized trust and trust of neighbors (a form of particularized trust important to examine due to the common focus on neighborhood or local community social capital in health research). Using a high quality, nationally representative Canadian dataset that included measures of several domains of respondents’ actual social capital, we find limited evidence for the argument that either trust measure is a component of or valid proxy for actual social capital. In interpreting our findings, we return to the two issues that motivated our analyses. 4.1. Analytic issue 1: associations between trust, social capital, and health measures If trust measures are either valid indicators of or simply proxy measures for social capital, then our analyses should have found that (a) these trust measures are each substantially associated with measures of one or more domains of social capital and (b) their respective associations with health outcomes weakened substantially once the measures of personal social capital are controlled in multivariate models. We found that both trust measures correlated in the expected directions with several domains of social capital measures, but the magnitudes of these relationships were rather modest. Likewise,

the respective relationships that each trust measure had with both health measures (findings previously observed in other studies) remained statistically significant and were only modestly reduced once the social capital measures were introduced as control variables in the model. Hence, the social capital measures did not substantially explain the associations that each trust measure had with both general and mental health. When considered together, these findings make it difficult to conclude that perceived trust is a component of social capital. At best, these findings provide evidence that some social capital domains (a) are associated with the distinct constructs of generalized trust and trust of neighbors, (b) may mediate the association between these two types of trust and health and/or (c) may have indirect influences on health via these trust items. Such conclusions are consistent with Lin’s (2001) argument that trust may promote social capital or vice versa as well as Glaeser et al.’s (2002) rationale for analyzing trust and personal social capital separately. 4.2. Analytic issue 2: differential findings for generalized trust and trust of neighbors In considering the validity of trust as a social capital indicator or proxy, we proposed that, if trust of neighbors is measuring social capital, then we would expect this trust measure to correlate more strongly with measures of neighborhood-based social capital than with measures of social capital rooted in more general network ties or relationships. Likewise, we would expect the correlation between trust of neighbors and health to be substantially reduced in magnitude once such neighborhood-based social capital measures were controlled. Consistent with this expectation, our findings indicated that trust of neighbors was more strongly associated with social capital measures focused on neighborhood ties than with measures of other social capital domains. These findings corroborate prior work that highlights how particularized trust captures in-group ties or ties within one’s community (Uslaner, 2002). Trust of neighbors focuses on particular people with whom one may interact regularly in the conduct of everyday routines and, thus, would be expected to have less direct application to social capital located beyond one’s neighborhood. Furthermore, we observed that these associations with neighborhood-based social capital measures (knowing neighbors in general and well enough to ask favors) were substantially stronger for trust of neighbors than for generalized trust. These findings corroborate prior research whereby generalized trust had weak associations with neighbor ties (Fischer, 2005; Moore et al., 2011). 4.3. Summary of collective results Our findings suggest that these two trust measures are capturing some construct (or even constructs) that are related todbut conceptually distinct fromdseveral domains of personal social capital and, hence, raise questions about the adequacy of trust measures as proxy measures for personal social capital; that is, people’s actual social connections or networks and their inherent resources. As such, these findings are consistent with arguments that, while a person’s trust may be representing moral and/or psychological orientations that are relatively distinct from her/his actual social ties and activities (Uslaner, 2002), it still has potential health implications (Kawachi et al., 2008). 4.4. Study strengths and limitations Our study has several strengths. In addition to using high quality, nationally representative survey data, our study was able

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

to: (a) focus on two measures that respectively captured generalized trust and particularized trust (in neighbors), (b) include many domains of personal social capital considered in health researchdfrom general networks to more local, geographicallybounded networks, and (c) analyze the data using multivariate techniques that could better account for the correlations between the trust variables as well as the health variables than could more traditional binary regression models. Nevertheless, some limitations need to be considered. First, the social capital measures have limitations in terms of precision and the types of network ties they aim to capture. Likewise, our analysis did not include measures used in some other social capital studies including participation measures (e.g., hours spent volunteering and religious service attendance) and behavioral trust measures (e.g., whether respondents leave their doors purposely unlocked) (e.g., Anderson et al., 2004). Given the range of social capital measures we examined though, it is still difficult to argue that the rather small reductions in the trustehealth associations produced by these measures are due to the trust measures being stronger proxies for some other types of social capital not included in this study. Second, the cross-sectional nature of our data prevents us from making any causal claims. We were able to control for prior health status to some degree, but it is still not possible to completely rule out that health status may be affecting one’s trust of others and social relationships. Also, our study only provides evidence about measures of personal social capital. Many health studies examine neighborhoodand country-level social capital, but we are unable to extrapolate our findings to such higher levels of analysis; as doing so risks making atomistic fallacies. 4.5. Implications for future health research on social capital Our findings have important implications for future research. First (and perhaps stating the obvious), health research on social capital needs to include measures of people’s actual social capital. Trust exists in many social relationships, but, it is neither a necessary nor sufficient factor for the generation of social capital (see Field, 2003; Glaeser et al., 2002). Though one might argue that generalized and particularized trust are capturing the potential or willingness of a person to establish social ties and generate social capital with similar or dissimilar othersdimportant research issues in their own rightdthis psychological state or disposition to engage with others is conceptually different from actually possessing social capital. Second, caution should be exercised when using these trust measures. While various forms of trust (e.g., generalized, particularized) may have health implications via biological, psychological, and sociological pathways, careful attention must be paid to (a) what aspects of trust are specifically being assessed by particular measures (essentially, the concordance between constructs and measures) and (b) what hypothesized pathways might link a specific trust form to specific health outcomes. For example, generalized trust may be overlapping with hostilityda personality trait that has implications for some (stress-related) health outcomes (Kawachi et al., 2008). Third, studies need to utilize more precise measures of personal social capital. Comprehensive efforts to include measures of cognitive and structural social capital in health surveys (e.g., Krishna and Shrader, 1999; Harpham, 2008) have been an important start. However, ego-centric network instruments, such as the Name Generator (Moore et al., 2009; Song and Chang, 2012), Position Generator (Carpiano and Hystad, 2011), and Resource Generator (van der Gaag and Weber, 2008), offer great utility for creating measures of social capital that a person can access via

9

different tiesdand thus can enable greater understanding of how networks may facilitate or undermine one’s health. 5. Conclusion For health research on social capital to make important contributions to knowledge, it is essential that researchers utilize measures that capture valid aspects of social capital. While using perceived trust measures to assess personal social capital may have appeal due to the lack of more desirable or comprehensive measures in a health dataset (and/or be legitimated in their use based on prior practices in published studies), authors, peer reviewers, and readers should recognize the conceptual and methodological implications of such practices for making valid inferences. To be explicit, we are, in no way, arguing for a doctrinaire (arguably anti-intellectual) stance that trust measures should be eliminated from social capital research. As our findings indicate, trust is associated with some domains of social capital. Rather, we simply stress that the use of trust measuresdlike any other measuresdshould be guided by strong theoretical considerations motivating one’s research approach (Carpiano and Daley, 2006a, 2006b; see also Lindström, 2004). In closing, we strongly encourage continued theoretical and empirical scrutiny of the constructs and measures utilized by health researchers in studying social capital as a social structural determinant of health. Ethics statement This paper uses publicly available data and thus is exempt from review as per the guidelines of the University of British Columbia Behavioral Research Ethics Board. Acknowledgments Richard Carpiano conducted this research while receiving funding from Investigator Awards from the Michael Smith Foundation for Health Research (_501100000245) and Canadian Institutes of Health Research (_501100000024). The authors express their thanks to Ralph Matthews, University of British Columbia, Thomas Abel, University of Bern, and Jennifer E.V. Lloyd, University of British Columbia for providing scholarship and/or comments that informed the preparation of this manuscript, as well as Joy Wang at StataCorp for providing software technical support that enabled the analyses for this paper. An earlier draft of this paper was presented at the 2013 International Conference on Social Stratification and Health, Tokyo, Japan. References Abbott, S., Freeth, D., 2008. Social capital and health: starting to make sense of the role of generalized trust and reciprocity. Journal of Health Psychology 13, 874e 883. Amemiya, T., 1981. Qualitative response models: a survey. Journal of Economic Literature 19 (4), 1483e1536. Anderson, L.R., Mellor, J.M., Milyo, J., 2004. Social capital and contributions in a public-goods experiment. American Economic Review 94 (2), 373e376. Bonett, D.G., 2007. Transforming odds ratios into correlations for meta-analytic research. American Psychologist 62 (3), 254e255. Bourdieu, P., 1986. The forms of social capital. In: Richardson, J.G. (Ed.), Handbook of Theory and Research for the Sociology of Education. Greenwood, New York, pp. 241e258. Brehm, J., Rahn, W., 1997. Individual-level evidence for the causes and consequences of social capital. American Journal of Political Science 41 (3), 999e1023. Buis, M., 2011. Stata tip 97: getting at r’s and s’s. The Stata Journal 11 (2), 1e3. Campbell, D., Fisk, D., 1959. Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin 56 (2), 81e105.

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017

10

R.M. Carpiano, L.M. Fitterer / Social Science & Medicine xxx (2014) 1e10

Carpiano, R.M., Daley, D.M., 2006a. A guide and glossary on postpositivist theory building for population health. Journal of Epidemiology and Community Health 60 (7), 564e570. Carpiano, R.M., Daley, D.M., 2006b. Theory building on the high seas of population health: Love Boat, Mutiny on the Bounty, or Poseidon Adventure? Journal of Epidemiology and Community Health 60 (7), 574e577. Carpiano, R.M., Hystad,, P.W., 2011. “Sense of community belonging” in health surveys: What social capital is it measuring? Health & Place 17 (2), 606e617. Claibourne, M.P., Martin, P.S., 2000. Trusting and joining? an empirical test of the reciprocal nature of social capital. Political Behavior 22 (4), 267e291. Coleman, J., 1988. Social capital in the creation of human capital. American Journal of Sociology 94, 95e120. Coulthard, M., Walker, A., Morgan, A., 2002. People’s Perception of Their Neighbourhood and Community Involvement: Results From the Social Capital Module of the General Household Survey 2000. Stationery Office, London. De Silva, M.J., Harpham, T., Tuan, T., Bartolini, R., Penny, M.E., Huttly, S.R., 2006. Psychometric and cognitive validation of a social capital measurement tool in Peru and Vietnam. Social Science & Medicine 62 (4), 941e953. De Silva, M.J., McKenzie, K., Harpham, T., Huttley, S.R.A., 2005. Social capital and mental illness: a systematic review. Journal of Epidemiology & Community Health 59, 619e627. Field, J., 2002. Social Capital. Routledge, New York. Field, J., 2003. Social Capital. Taylor & Francis, New York. Fischer, C.S., 2005. Bowling Alone: what’s the score? Social Networks 27 (2), 155e 167. Fujiwara, T., Kawachi, I., 2008. Social capital and health. A study of adult twins in the U.S. American Journal of Preventive Medicine 35 (2), 139e144. Gelman, A., 2006. Take logit coefficients and divide by approximately 1.6 to get probit coefficients. In: Statistical Modeling, Causal Inference, and Social Science. Retrieved October 24, 2013, from. http://andrewgelman.com/2006/06/06/take_ logit_coef/. Glaeser, E.L., Laibson, D., Sacerdote, B., 2002. An economic approach to social capital. The Economic Journal 112 (483), F437eF458. Glaeser, E.L., Laibson, D., Scheinkman, J.A., Soutter, C.L., 2000. Measuring trust. Quarterly Journal of Economics 115, 811e846. Glanville, Paxton, P., 2007. How do we learn to trust? A confirmatory tetrad analysis of the sources of generalized trust. Social Psychology Quarterly 70 (3), 230e242. Harpham, T., 2008. The measurement of community social capital through surveys. In: Kawachi, I., Subramanian, S.V., Kim, D. (Eds.), Social Capital and Health. Springer, New York, pp. 51e62. Harpham, T., Grant, E., Thomas, E., 2002. Measuring social capital within health surveys: key issues. Health Policy and Planning 17 (1), 106e111. Hyyppä, M.T., Mäki, J., 2001. Individual-level relationships between social capital and self-rated health in a bilingual community. Preventive Medicine 32, 148e 155. Igarishi, T., Kashima, Y., Kashima, E.S., Farsides, T., Kim, U., Strack, F., Werth, L., Yuki, M., 2008. Culture, trust, and social networks. 11, 88e101. Kankainen, T., 2009. Voluntary associations and trust in Finland. Research on Finnish Society 2, 5e17. Kawachi, I., 2010. Social capital and health. In: Bird, C.E., Conrad, P., Fremont, A.M., Timmermans, S. (Eds.), Handbook of Medical Sociology, sixth ed. Vanderbilt University Press, Nashville, TN, pp. 18e32. Kawachi, I., Subramanian, S.V., Kim, D., 2008. Social capital and health: a decade of progress and beyond. In: Kawachi, I., Subramanian, S.V., Kim, D. (Eds.), Social Capital and Health. Springer, New York, pp. 1e26. Kim, D., Subramanian, S.V., Kawachi, I., 2008. Social capital and physical health: a systematic review of the literature. In: Kawachi, I., Subramanian, S.V., Kim, D. (Eds.), Social Capital and Health. Springer, New York, pp. 139e190. Kohler, U., Karlson, K.B., Holm, A., 2011. Comparing coefficients of nested nonlinear probability models. The Stata Journal 11 (3), 420e438. Krishna, A., Shrader, E., 1999. Social Capital Assessment Tool. The World Bank, Washington, DC. Available at. http://siteresources.worldbank.org/ INTSOCIALCAPITAL/Resources/Social-Capital-Assessment-Tool-SOCAT-/ sciwp22.pdf (accessed 05.11.11.). Lin, N., 2001. Social Capital: a Theory of Social Structure and Action. Cambridge University Press, Cambridge.

Lin, N., Fu, Y., Hsung, R.M., 2001. The position generator: measurement techniques for investigations of social capital. In: Lin, N., Cook, K., Burt, R.S. (Eds.), Social Capital: Theory and Research. Transaction Publishers, New Jersey, pp. 57e81. Lindström, M., 2004. Social capital, the miniaturization of community and selfreported global and psychological health. Social Science & Medicine 2004 (59), 595e607. Mitchell, A.D., Bossert, T.J., 2007. Measuring dimensions of social capital: evidence from surveys in poor communities in Nicaragua. Social Science & Medicine 64, 50e63. Mood, C., 2010. Logistic regression: why we cannot do what we think we can do, and what we can do about it. European Sociological Review 26 (1), 67e82. Moore, S., Bockenholt, U., Daniel, M., Frohlich, K., Kestens, Y., Richard, L., 2011. Social capital and core network ties: a validation study of individual-level social capital measures and their association with extra- and intra-neighborhood ties, and self-rated health. Health & Place 17 (2), 536e544. Moore, S., Daniel, M., Paquet, C., Dube, L., Gauvin, L., 2009. Association of individual network social capital with abdominal adiposity, overweight and obesity. Journal of Public Health 31, 175e183. Moore, S., Haines, V., Hawe, P., Shiell, A., 2006. Lost in translation: a genealogy of the social capital concept in public health. Journal of Epidemiology & Community Health 60, 729e734. Nannestad, P., 2008. What have we learned about generalized trust, if anything? Annual Review of Political Science 11, 413e436. Paxton, P., 2007. Association memberships and generalized trust: a multilevel model across 31 countries. Social Forces 86 (1), 47e76. Policy Research Initiative, 2003. Social Capital: Building on a Network-Based Approach. Draft Discussion Paper. http://publications.gc.ca/pub? id¼264914&sl¼0 (Last accessed 5 Dec 2011). Portes, A., 1998. Social capital: its origins and applications in modern sociology. Annual Review of Sociology 24, 1e24. Putnam, R.D., 1995. Bowling alone: America's declining social capital. Journal of Democracy 6 (1), 65e78. Putnam, R.D., 2000. Bowling Alone: the Collapse and Revival of American Community. Simon & Schuster, New York. Ross, N., 2002. Community belonging and health. Health Reports 13 (3), 33e39. Rotter, J.B., 1980. Interpersonal trust, trustworthiness, and gullibility. American Psychologist 35 (1), 1e7. Sapag, J.C., Poblete, F.C., Eicher, C., Aracena, M., Caneo, C., Vera, G., Martínez, M., Hoyos, R., Villarroel, L., Bradford, E., 2010. Tobacco smoking in urban neighborhoods: exploring social capital as a protective factor in Santiago, Chile. Nicotine & Tobacco Research 12 (9), 927e936. Smith, S.S., 2013. Race and trust. Annual Review of Sociology 36, 453e475. Song, L., Chang, T., 2012. Do resources of network members help in help seeking? Social capital and health information search. Social Networks 34 (4), 658e669. Statistics Canada, 2010. Canadian General Social Survey. Available: http://www. statcan.gc.ca/cgibin/imdb/p2SV.pl? Function¼getSurvey&SDDS¼4503&lang¼en&db¼imdb&adm¼8&dis¼2. Uslaner, E.M., 2002. The moral foundations of trust. In: Paper Presented at The Symposium, “Trust in the Knowledge Society,” University of Jyvaskyla, Jyvaskala, Finland 20 September, 2002. van der Gaag, M., Weber, M., 2008. Measurement of individual social capital: questions, instruments, and measures. In: Kawachi, I., Subramanian, S.V., Kim, D. (Eds.), Social Capital and Health. Springer, New York, pp. 29e49. Veenstra, G., 2002. Explicating social capital: trust and participation in the civil space. Canadian Journal of Sociology 27 (4), 547e572. Wollebaek, D., Selle, P., 2002. Does participation in voluntary associations contribute to social capital? The impact of intensity, scope, and type. Nonprofit and Voluntary Sector Quarterly 31, 32e61. Yamagishi, T., Yamagishi, M., 1994. Trust and commitment in the United States and Japan. Motivation and Emotion 18 (2), 129e166. Ziersch, A.M., Baum, F.E., MacDougall, C., Putland, C., 2005. Neighbourhood life and social capital: the implications for health. Social Science & Medicine 60 (1), 71e 86.

Please cite this article in press as: Carpiano, R.M., Fitterer, L.M., Questions of trust in health research on social capital: What aspects of personal network social capital do they measure?, Social Science & Medicine (2014), http://dx.doi.org/10.1016/j.socscimed.2014.03.017