Occupational inequalities in volunteering participation: Using detailed data on jobs to explore the influence of habits and circumstances

Volunteering is an important component of the social fabric of contemporary wealthy countries. Services supported by volunteers provide contributions to a wide range of welfares, and are proactively encouraged by national and local governments (e.g., Ishkanian & Szreter, 2012). Voluntary organizations are also thought to enhance social cohesion and trust (e.g., Paxton, 2007), and to benefit volunteers, for instance with improved employability (e.g., Ellis Paine, McKay, & Moro, 2013; Kamerade & Ellis Paine, 2014; Wilson, Mantovan, & Sauer, 2019; Wilson & Musick, 2003) and well-being (e.g. Binder & Freytag, 2013; Konrath, 2012; Nichols & Ralston, 2011; Wheeler, Gorey, & Greenblatt, 1998). Nevertheless voluntary participation is socially structured. In general, formal volunteering— “giving unpaid help through groups, clubs or organisations to benefit other people or the environment” (Rochester, Received: 27 June 2019 | Revised: 6 March 2020 | Accepted: 11 March 2020 DOI: 10.1111/1468-4446.12756

to the UK Citizenship Survey definition)-is more often undertaken by those with higher levels of education and those in more favorable socio-economic circumstances; by the younger and older rather than mid-aged; by women rather than by men; by host-country nationals rather than immigrants or ethnic minorities; and by those not living with disabilities (e.g., Alesina & La Ferrara, 2000;Dean, 2016;Lindsey & Mohan, 2018;Musick & Wilson, 2008;Rochester et al., 2010, c4;Shandra, 2017;Valentova & Alieva, 2018;Voicu & Serban, 2012;Wilson, 2000;Wilson & Musick, 1997). Some authors have highlighted issues of social inclusion in voluntary participation and explored strategies to reduce social inequalities in engagement (e.g., Rochester et al., 2010, c14).
Amongst other social inequalities, volunteering is associated with occupations. Wilson and Musick (1997) demonstrate that volunteering is more common for people in public sector occupations, for those in relatively more socio-economically advantaged occupations, and for those in occupations that provide skills relevant to the needs of specific volunteering roles (see also Kamerade 2009;Lengfeld & Ordemann, 2016;Paxton, 2007;Rotolo & Wilson, 2006). Based upon occupational indicators, many studies find that more advantaged jobs are associated with greater levels of volunteering, explained through greater flexibility and autonomy (Musick & Wilson, 2008); human capital (Wilson et al., 2019); and social connections (Wiertz, 2015).
Previous studies, however, have used quite simplistic measures of occupational status, most often operationalized as simply three or four categories based on high-level occupational classifications. Relatively more sociological information is captured by occupational data--in particular, recent interest in occupational "microclass" inequalities has shown that quite specific occupational differences are related to an array of social outcomes, grounded in the premise that occupations and their host organizations instil distinctive but consequential cultures and experiences upon their incumbents in a fine-grained manner (e.g., Weeden & Grusky, 2012). In the analysis that follows we show that enhanced insight about social inequalities in volunteering can be gained from using occupational information in more extended depth than has previously been common. In particular, we argue that data on occupations can help us better evaluate the relative roles of "circumstances" and "habits" as influences upon volunteering.

| HAB ITS , CIRCUMS TAN CE S AND O CCUPATI ON S A S INFLUEN CE S UP ON VOLUNTARY PARTICIPATION
Recent longitudinal evidence has suggested that volunteering is often a stable, lifetime orientation, that is relatively robust to the vagaries of other life changes (e.g., Dawson, Baker, & Dowell, 2019;Lancee & Radl, 2014;McCulloch, 2014;Mohan & Bulloch, 2012;Taylor-Collins, Harrison, Thoma, & Moller, 2019). Likewise, sociological theories of volunteering often emphasize the influence of structural circumstances that may support or discourage volunteering and that tend to be largely stable over time-for instance, people's "human capital," such as their qualifications and occupational skills, and their "social resources," such as their networks and contacts (e.g., Wilson, 2000). In spite of this, policy-oriented literatures (and voluntary organizations themselves) often focus on the ways in which more localized and transient factors might encourage or discourage volunteering (e.g., Rochester et al., 2010, c14). As a device to foreground this divergence-and because we will argue that occupational data provide a helpful tool to study it-we suggest a conceptual distinction between "circumstances" and "habits" as influences upon voluntary participation.
"Circumstances" might be thought of as direct, tangible factors that can inhibit or enable an individual's ability to volunteer, but that are, in principle, readily changed (for individuals), or adapted to (by organizations). Time availability is often cited as an inhibiting "circumstance" in UK research, but many other barriers are also often mentioned (e.g., Low, Butt, Ellis Paine, & Davis Smith, 2007, as discussed in Rochester et al., 2010, pp. 193-200). Some barriers may be easy to recognize and communicate (for instance, childcare responsibilities, a lack of resources for transport, or a lack of knowledge about relevant opportunities). Some circumstances, however, may constitute barriers through a more indirect route (for instance, gender might inhibit participation if an organization uses gender-biased recruitment). Enabling "circumstances" may also facilitate volunteering-for instance, surplus wealth to purchase relevant equipment; a privileged job that provides autonomy over working hours; a skill or capacity that enables a prospective contribution. Whatever the focus, we conceptualize circumstances as scenarios in which individual volunteering is responsive to specific lifestyle arrangements. Crucially, the relevant arrangements could be changed and/or adapted to: this may be through a personal change, such as young dependent children reaching school age; or through a change in organizational support, such as a transport subsidy.
We conceptualize "habits" by contrast as individuals' capacities, orientations, or propensities towards or against volunteering, which are maintained more or less irrespective of immediate lifestyle arrangements. Longitudinal evidence of persistent volunteering might suggest the influence of "habits." For instance, Dawson et al. (2019) decompose empirical persistence in volunteering and suggest that around two-thirds of it might be explained by "… unobserved heterogeneity or underlying behaviours of individuals," propensities akin to what we label as "habits" (see also Lancee & Radl, 2014). The development of volunteering as a habit is also supported by Taylor-Collins et al. (2019), who show that habits of social action formed by adolescents are predictive of their participation.
"Habits" themselves may be substantially shaped by social background-for instance, values ingrained in childhood. Dean (2016) argues that the Bourdieusian concept of "habitus" (often understood as a person's structurally driven, learned social orientations) influences voluntary participation (although we prefer the term "habits" to avoid implying a specifically Bourdieusian approach). "Habits" could also include quite personalized or psychological factors-for instance, Rochester et al. (2010) argue that some individuals don't volunteer because they lack the self-confidence and self-esteem to believe that they could make a useful contribution.
If the difference between circumstances and habits as influences upon voluntary participation is worth investigating, our key argument is that detailed occupational data can make a useful contribution to this end. The social sciences in general, and the UK in particular, benefit from access to many larger scale secondary survey and administrative microdata resources that include data on organized voluntary participation (e.g., Li, 2015), and, amongst other information, detailed occupational records (that is, relatively fine-grained information on occupational titles, such as the 400 or so categories of a standardized occupational taxonomy). As far as we are aware, detailed occupational data hasn't previously been extensively analysed in terms of voluntary participation, but it might shed insight on volunteering in two ways.
First, detailed measures might often be the best available empirical proxy for other important influences.
Occupations link to numerous important inequalities of experience that may influence volunteering. These include, but are not necessarily limited to, inequalities in socio-economic advantage; educational and training background; flexibility and autonomy in working activities and time use; circumstances of security or precarity; social networks and social capital; volume of and control over leisure time; demographic circumstances; and even systematic patterns in health and well-being (e.g., Oesch, 2013;Rose & Harrison, 2010). In principle, such factors could be directly measured in a social survey instrument, and their influence disentangled from others.
Nevertheless, in practice, many such concepts may not be measured directly at all, or even if they are, they may be measured or analysed imperfectly; accordingly rich occupational data might often capture their indirect influence.
A second way that occupations might link to voluntary participation is if occupations define important social entities characterized by a shared cultural milieu. Previous studies in the microclass tradition have shown that occupational institutions that are otherwise very similar in terms of objective work arrangements still show residual differences in their empirical patterns, a finding that has been attributed to the shared cultural orientations and lifestyles of those in that occupation (e.g., Weeden & Grusky, 2012). For our purposes, therefore, if an occupation has a distinctive pattern of voluntary participation that does not seem to be the proxy function of another factor, it is plausible that the pattern reflects the influence of the shared cultural milieu of the incumbents of that occupation. To give an example, our results suggest that several occupations linked to the finance sector in the UK (bank managers, financial managers, marketing managers, sales representatives) have, net of controls, average or above average voluntary participation, but below average levels of civic participation, and below average patterns in sustaining voluntary participation through time (Table 6). This interesting pattern might suggest the influence in these domains of an individualistic work culture-perhaps people in these occupations will often make voluntary inputs when it suits them, but will steer clear of sustained commitments, memberships, or obligations. As illustrated in the example above, detailed occupational data may provide enough information to support reasonable "post hoc" interpretations of the sociological mechanisms driving an empirical pattern. That is, given a distinctive association between participation and a detailed occupational code (net of other controls), it might be reasonable to speculate about the underlying mechanism that accounts for this relationship, given other knowledge about the occupation in question. Confident adjudication may require further analysis, perhaps with new controls for other direct measures, and there may be a good case for undertaking supplementary qualitative analysis to investigate further (e.g., Friedman & Laurison, 2019;Harding & Seefeldt, 2013). Critically, however, measures of detailed occupations may provide an efficient tool to identify (perhaps unanticipated) statistical patterns in volunteering.
Whether occupational data is helpful in this domain will hinge upon two empirical questions that frame the research analysis that follows. First is whether detailed occupational data does provide significant improvements in statistical fit in predicting outcomes related to organized voluntary participation, net of other controls. Second is whether differences between specific occupations in their relationships to organized voluntary participation can help us to suggest and adjudicate between plausible mechanisms that influence voluntary participation. As we will illustrate in the analyses below, we believe that there is a compelling argument that occupational data can readily be used to provide new ideas about the relative influence of "habits" and "circumstances" upon voluntary participation. Table 1 summarizes the four secondary datasets from which we present results. Each study is designed to support analysis that is nationally representative (subject to adjustments for sample design and response). It is useful to use multiple datasets because they span several different measurements of voluntary participation, as it is recognized that the way in which voluntary participation is recorded can be consequential to results (e.g., Rooney, Steinberg, & Schervish, 2004 Mohan & Bulloch, 2012) and civic association membership (e.g., Li, 2015;Li, Savage, & Pickles, 2003;Li, Savage, Tampubolon, Warde, & Tomlinson, 2002;Sturgis, Patulny, Allum, & Buscha, 2015). for the BHPS-responses are analysed as an ordinal outcome in statistical models. As we use multiple waves of the BHPS, we can use more than one answer from the same individual over different years-the first column in mentary questions about these activities that we did not exploit). "Civic participation" (for instance, membership of a club or society) is an important category of data on voluntary participation, and the BHPS, UKHLS, and SMI surveys feature similar (but not identical) questions covering civic participation (see, e.g., Figure 3). Forms of voluntary participation are many and varied, and our measures do not necessarily assess all nuances of volunteering patterns. Like most literature, our analysis is weighted towards evidence on formal volunteering, which is also traditionally the focus of policy makers and volunteer-involving organizations (Williams, 2003).

| DATA AND ME THODS
However, while the measures used by HOCS specifically distinguish formal and informal volunteering (Table 2), and the survey questions on civic participation will by definition involve an organization, the questions on active volunteering that we use from the BHPS and UKHLS (e.g., Table 2) do not unambiguously distinguish whether an external organization is involved. for association membership, "some volunteering" and "weekly volunteering", respectively), and there are generally as many or more respondents who contribute occasional volunteering that is not as persistent through time. This implies that measures based on longitudinal trajectories of volunteering have the potential to reveal different participation inequalities than those based only upon current behaviors.
Since all adults resident in a responding household are interviewed in the BHPS and UKHLS, for these data-  zations, but we excluded these from our derived measure (following Sturgis et al., 2015, on the grounds that these organizations represent "isolated" forms of civic participation which differ from other domains of participation).
We considered several alternative categorizations of organizations, but variations in the results in terms of which operationalization was used seemed minimal. Although the components of association membership are similar across the surveys, the fluctuations in Figure 3 suggest that there could be artefactual differences between the three surveys in the way association membership patterns are reported.
To analyze the survey data, we fit statistical models where the outcome is a measure of voluntary participation. 3 Most models include an array of commonly used "control variables" covering socio-demographic and socioeconomic characteristics, as described in the outputs below. 4 Descriptive statistics on all variables used in each analysis, and "Log files" giving supplementary details, including background information on the operationalization of measures, full details of the occupational codes, and additional details on statistical models, are posted online. 5 In all models we consider whether parameters that capture occupation-to-occupation variations contribute to an improvement in model fit, compared to comparable models that don't allow for occupational patterns. The occupational influence is usually measured by allowing "random effects" for occupations in the detailed unit groups associated with the dataset (see Table 1, Column 3). Random effects models provide a convenient tool for assessing variations between occupations as well as for retrieving statistics on individual occupations in terms of their model-based residuals. 6 We are often interested in "random intercepts" associated with occupations: evidence that, net of other things in the model, a specific occupation stands out as being above or below average on the outcome. In some scenarios it is also of interest to test for "random slopes" with occupational variations: evidence that, net of other controls, specific occupations are linked to an above-or below-average impact of another explanatory variable in its influence on participation.

| RE SULTS
Figures 4, 5 and Table 3 summarize patterns of association between occupational unit groups and data about voluntary participation. Without taking account of any other measured factors, we see a modest but statistically significant association across a range of datasets and a variety of measures related to voluntary participation. number of respondents in the occupation, and the shading is a function of the CAMSIS score for the occupation (see Lambert & Griffiths, 2018; darker shading implies a higher CAMSIS score, indicating an occupation whose incumbents enjoy more advantaged circumstances in the structure of social stratification). The patterns shown correspond to 8.5% and 7.2%, respectively, of the estimated proportions of variance in volunteering patterns that is associated with systematic differences between occupations (i.e., intra-cluster correlations of 0.085 and 0.072)-see Table 3. Indeed, the patterns suggested in Figures 4 and 5  Note: In all models, occupational-level random effects significantly improve model fit. a BHPS and UKHLS figures may be shown either for "most recent" (most recent record with valid measure-UKHLS, N = 31,246; BHPS, N approx = 15k), or for all valid records in dataset, including multiple records per respondent (approx. 150k records from 20k respondents, using weights to deflate influence of repeated measures).
The description above concerns the bivariate relationship between occupations and volunteering, but some of that association seems to reflect an occupation's social stratification position. Arguably, detailed occupations are only important if some of their associations with volunteering persist net of controls for easily-measured heterogeneities that occupations might proxy. the ICC value (compared to Column 1), which indicates that much of the bivariate association is indeed about the social advantage of the occupation. Moreover, the ICCs in Column 2 remain modest but statistically significant.
Indeed, evidence of the distinctive influence of occupations is reinforced when more elaborate statistical models are specified. Tables 4 and 5 show results from selected models that use controls for several other factors that might be linked to voluntary participation and occupations. Most models use "random effects" parameters to allow for occupation-to-occupation variation in the outcomes. 7 The key observation at this point is that, even net of a wide range of controls, the random effects parameters for occupational variation are consistently associated with a significant improvement in model fit. 8 Some of these models also test occupational "random slopes," which evaluate the substantively interesting hypothesis that the impact of another explanatory factor upon voluntary participation varies significantly from occupation to occupation-in Models 1c, 1d, and 3c, this is seen to be the case for the effect of weekly working hours on participation level; in Models 2b and 5a, for the influence of educational score.
We have shown that detailed occupations have some empirical connection to voluntary participation, but can this relationship be leveraged to gain more useful insights? By adding parameters for detailed occupational variations, one possibility is that the other coefficients in statistical models predicting volunteering might change in a consequential way. Some such examples are evident in Tables 4 and 5, 9 where we can point to modest changes in parameters that emerge when we control for occupational variations with random effects-for instance, a lessened direct influence of gender and of occupational and educational scores. Such changes mean that the appropriate control for occupations could modify our substantive story: it is plausible that the larger estimated influences of gender, education, and occupational advantage were spurious, but are now more appropriately estimated, once we have controlled for the more nuanced relationships with fine-grained occupations.
Another possibility is that individual occupations may emerge that are associated with a distinctive pattern of above-or below-average voluntarism. Such evidence could be both descriptively and analytically useful in understanding voluntarism: descriptively, such as by identifying pockets of society with unusual patterns of volunteering; analytically, if the occupational pattern could be given a useful theoretical interpretation. In Figure 5, for example, one of the largest occupations with a statistically significant positive residual (the large dark circle in the top right of image) are CODOT category 193 ("Primary and secondary school teachers"). This suggests that among men in 1972 (the sample survey was given only to males), teaching was unusually often associated with civic association memberships. This pattern might be consistent with a few hypotheses about voluntary engagement: perhaps it reflects a "habit" associated with teachers, such as pro-social cultural orientation; or perhaps a "circumstance" such as an enabling factor (say, if male teachers in 1972 had more spare time and flexibility than most other occupations). Figure 5, however, summarizes bivariate patterns, and more compelling evidence might emerge from examining the occupation-level residuals in random effects models that have ample controls for other factors. Table 6 shows some of the occupations that stand out, in different scenarios, as having distinctively higher or lower levels of voluntarism compared to average and net of controls. A first notable point-already implicit in previous results-is that these occupations exist at all. Even though the overall empirical association between volunteering and occupations is modest rather than strong, we have found that standard secondary survey datasets are sufficiently powerful as to allow us to identify important social groups (occupations) that have distinctively above-or below-average patterns in voluntary engagement, net of a range of standard individual controls (age, gender, marital and family status, and broad characteristics of educational attainment and economic activities). A more powerful result, however, might arise if we believe that some of the occupations that emerge in Table 6 (or indeed, that emerge in other analyses aside from those that we have presented here), lead us to new and meaningful insights about the social determinants of volunteering. We believe this is the case-one example was highlighted in the preceding discussion, in the dual pattern of above-average one-off volunteering, but below average association membership and sustained volunteering, linked to several jobs in finance.
One interesting pattern in the "random slopes" results in Table 6 (i.e., the occupations with heightened or lessened effects of working hours or education) seems apparent in the patterns associated with educational effects.
Across a handful of examples, there seems to be some pattern that occupations that tend to be more educationally homogenous (such as traditional professions) are characterized by a less positive individual level educational effect upon volunteering; in some occupations whose incumbents do not necessarily have high levels of education but that can be more heterogeneous in terms of formal education, individual educational background matters more. Further analysis might be necessary to substantiate this pattern, but the apparent suggestion is that in some occupations, only those individuals with higher levels of education will tend to volunteer, and in other occupations, the impact of individual education matters much less. Pragmatically, such findings might provide valuable market intelligence for stakeholders in the third sector.
What seems to us the most interesting pattern, in Table 6 and further results, concerns a post-hoc rationalization related to how "pro-social" occupations are. In Table 6, it is perhaps surprising to notice several examples TA B L E 6 Selected occupations with extreme residuals (net of standard controls) in random effects models of voluntary participation when "pro-social" occupations seem to have lower rather than higher levels of volunteering, net of controls (for instance, doctors, teachers and nurses). We should reiterate that these are the patterns net of controls-there are generally positive effects of education, and of working in the public sector, on voluntary participation, and the occupational residuals show merely that these occupations do not tend to show quite as high a level of participation as might otherwise be predicted. Nevertheless, there is a plausible post-hoc rationalization that seems to be consistent with these patterns and suggests the force of "habits" rather than "circumstances": perhaps many people in "pro-social" jobs already feel that they make a sufficient positive social contribution through their occupation, and don't feel too much duty or pressure to do more through volunteering; by contrast, perhaps the same trade-off influences some people in much less "pro-social" jobs to be a little more inclined than average to volunteer-for instance, the above average volunteering of financial institution managers in Table 6 (Models 1b and 1c). 10 This rationalization is speculative, as other mechanisms might also account for the empirical pattern-but it seems to us to provide a plausible hypothesis about a complex pattern.
Hitherto our results have concerned how occupations are linked to the propensity to engage in voluntary activities, but it is a natural extension to ask if occupations have an impact upon the personal benefits or consequences of volunteering (cf. Wilson et al., 2019). This can be achieved by modelling other outcomes-for instance, personal well-being, or employment attainment-as a function of voluntary engagement (and other controls), and then, allowing for "random slopes" with occupational units on the effect of voluntary participation. These models would estimate a parameter for occupational-level variance in the impact of participation, and would allow us to calculate residuals that might indicate specific occupations that are associated with a more positive (or negative) impact of volunteering (for examples of the use of "random slopes" by occupations, see Lambert & Griffiths, 2018, c9). Using the datasets described above, we have tried out a range of models of this character (results not presented). Hitherto we have not found strong patterns that hold consistently across the different datasets. In the BHPS data, however, a similar pattern to that of Table 6 seemed to emerge in terms of a few occupations that stood out for being associated with a considerably more favorable well-being benefit from volunteering (i.e., more than the average premium, which is positive), and likewise a number of occupations that stand out at the other end of this distribution-occupations where the benefit of volunteering for well-being is more muted, or even negative. In the BHPS, we observed a pattern of lessened benefits from volunteering in occupations that might be seen as more "pro-social" (examples included nurses, restaurateurs, and university teachers), and heightened benefits for those that are rarely portrayed as pro-social (examples include marketing managers, accounts clerks, and bank managers). Nevertheless, although we find this a plausible social pattern, we were not able to replicate similar results when modeling well-being outcomes in a similar way using the UKHLS dataset (where we found no significant occupational-level random slope) nor using the European Social Survey (where we found a small occupational-level random slope, but the outlying occupations did not seem to have a similar pattern). It is not particularly clear, therefore, that the benefits of volunteering vary substantially by detailed occupations net of other controls, but it is possible that future research may find convincing, consistent patterns in such relationships. Note: Based on analysis of datasets described in Table 1. Occupations are categories of SOC2000 (UKHLS, HOCS) or SOC90 (BHPS). Not all extreme occupations are listed. Occupations shown are represented by at least 100 cases in the dataset (unless indicated *). Standard font: occupational residual exceeds its standard error by at least 2; italic font: occupational residual exceeds its standard error by less than 2 but more than 1.
TA B L E 6 (Continued)

| CON CLUS IONS
We have argued that the interpretation of detailed occupational patterns can sometimes be used to distinguish "habits" and "circumstances" as influences on voluntary participation. A distinction is worth making as it could inform relevant policies: for example, if habits matter more than circumstances, policies on volunteering might be re-directed to focus upon wider issues of socio-economic inequalities and social origins. Pragmatically, if an organization wanted to expand its voluntary participation and emphasized a model of circumstances, it might prioritize means of accommodating, say, flexible and part-time contributions; if it were persuaded that habits were relatively more important, it might direct its efforts towards managing its reputation.
Our results demonstrated that occupations were indeed associated with distinctive empirical links to volunteering net of controls for other factors, consistently across surveys, time periods, measures of volunteering, and across longitudinal and household contexts as well as at an individual level. That is, the empirical link applies at the individual level (an individual's job is associated with their current voluntary participation), but it also transmits through households and over time (data on an individual's job could help predict volunteering of their household sharers, and their patterns of volunteering through time). These patterns were consistently statistically significant, albeit often of only a modest effect size. Moreover, we stressed that some of the empirical association between occupations and volunteering is likely to reflect other measurable social inequalities that occupations can proxy.
Nevertheless we also argued that the net relationships (controlling for many relevant direct measures) revealed specific patterns of participation that, we argued, could usefully be rationalized, drawing upon other knowledge about the occupation, to make plausible claims about the relative influence of "habits" and/or "circumstances." A complex aspect of our argument is that the rich empirical insights that can come from analyzing detailed occupational positions in relation to volunteering do not unequivocally identify the underlying mechanism behind the associations that we find. We believe that many of our examples reveal some occupational patterns that reflect what can usefully be interpreted as the effects of "habits," and others that can helpfully be thought of as reflecting "circumstances." However, these interpretations are premised on post-hoc rationalizations that might readily be debated; a more conclusive statement about the underlying mechanisms could require further supplementary analysis. The exciting opportunity presented by analyzing detailed occupational data reflects that measured occupations provide rich data about circumstances in an unusually efficient way-knowledge of a specific occupation gives us a lot of insight into a person's circumstances, and so the depth of data that is captured in an occupational title has the capacity to suggest precise, sometimes unexpected, social processes. By contrast, most previous studies of social inequalities in volunteering have used relatively broad-brush occupation-based measures that provide little clue as to the precise mechanisms related to a pattern of association. Wilson et al.'s (2019) compelling evidence of social class differences in the benefits of volunteering, for instance, might reveal patterns that result from any number of social inequalities that are clustered in class categories (such as employment relationships, economic advantages, skills, and social environments). A focus on micro-occupational differences will not eliminate such ambiguities, but we believe it increases the chances of us identifying the most plausible specific mechanism.
Was there, then, anything in our results to adjudicate between the relative influence of "habits" and "circumstances" upon volunteering? Empirical results suggest, first, the coexistence of both mechanisms, as examples could be found that were consistent with either narrative. However, our results also suggest an asymmetry. There were a number of instances when occupational patterns seemed to reveal niche examples of "circumstances" that shape voluntary participation-for instance, outlying patterns for some jobs, such as in higher education, seem to reflect the unusual levels of day-to-day autonomy that might enable them to better accommodate voluntary engagement irrespective of working hours. Nevertheless, the weight of evidence from occupational variations seems to be more often illustrative of the role of "habits." For instance, there are several occupations with aboveor below-average participation patterns, net of other controls, where there is no obvious example of a tangible circumstance that constrains or enables participation, but there is a plausible explanation in terms of "habit." The model of habits might indeed be reinforced by our evidence on the influence of the occupations of a person's household sharers upon their own participation patterns (as it is plausible that the "habits" of one person shape those of the household). Much of the practical literature on supporting volunteering has historically concentrated upon identifying "circumstances" (e.g., Rochester et al., 2010), but our evidence suggests an argument for paying relatively more attention to "habits" such as lifestyles and cultural orientations as factors that link to voluntary participation.
Our results also suggest that other studies of influences upon voluntary participation ought to consider controlling for detailed occupational positions (which has rarely been done in previous analyses). In a statistical modeling analysis, failure to do so might risk "omitted variable bias," which can ultimately lead to inappropriate interpretation of statistical patterns and inappropriate statistical estimates. However, the same point also applies to other social research on voluntary participation (such as in the interpretation of qualitative interview data): attention to specific occupations might be important and should not be overlooked.
It is possible that some of our findings could be contingent on our context. Occupations themselves may change over time in their contents or influence. Our analysis showed evidence of occupational influences over quite a lengthy period in the UK , but detailed occupational data is available in the social sciences across wider periods and different countries in a format that might readily suit extending analyses to wider contexts. Likewise, our assessment of voluntary participation has largely been limited to that of "formal" volunteering, based upon available measures in secondary survey data. Our results did not seem to point to many substantial differences between occupational inequalities in voluntary activities, and association memberships, although these two categories have the potential to reflect different patterns, and might usefully be explored further.
Likewise there may also be scope for interesting explorations of the links between informal forms of volunteering and occupations (which might have different features from those involving formal volunteering-e.g., Bradford, Hills, & Johnston, 2016).
One of the more interesting post-hoc rationalizations that may reflect "habits" was the evidence, across a few applications, of the influence of how "pro-social" an occupation might be. Freeland and Hoey (2018) argue that occupations are differentiated in an important dimension that is often overlooked but can be measured by patterns of deference-Freeland and Hoey describe this as a structure of "status," but it might equally be labeled one of the "goodness" or "pro-social character" of jobs. Our results from the UKHLS and BHPS might have tapped into that dimension and suggested a plausible post-hoc interpretation: perhaps those in "good" occupations already enjoy the "warm glow" of "doing good" and engaging with others that volunteering can sometimes provide, and tend to avoid taking on further commitments of this nature; by contrast, perhaps efforts to recruit volunteers might be most efficiently targeted at those in other occupations.
There are many other ways that increased understanding of the specificities of how volunteering relates to occupations could be useful. They might support insights that could lead to policy initiatives that facilitate more volunteering, and/or challenge existing social inequalities in volunteering. Ethnic and gender biases in volunteering activities might be confronted, for instance, by introducing policies targeted at occupations that have traditionally skewed gender or ethnic distributions. Evidence might also contribute helpfully to recent narratives that recognize the negative as well as positive aspects of some forms of voluntary participation. Some modes of volunteering might be seen as exploiting insecure workers such as interns, or obliging and pressuring contributors ("voluntolding" in the language of Kelemen, Mangan, & Moffat, 2017). Whilst above we have concentrated on evidence for facilitators of volunteering, in further work, the same relationships might also be cross-examined to build evidence and theory on negative aspects of voluntary participation.
One last point worth emphasizing is that voluntary participation remains socially structured. Detailed occupations can account for a moderate proportion of social inequalities in volunteering, but so too can many other measurable social heterogeneities across socio-demographic and socio-economic factors. It is sometimes suggested that traditional structural inequalities (including occupations) are increasingly less influential in shaping