Determinants of participation in voluntary work: a systematic review and meta-analysis of longitudinal cohort studies

A search was conducted in MEDLINE, PsychINFO, SocINDEX, Business Source Premier and EconLit, on August 12th, 2015. The search strategy included a combination of terms related to (a) participation in voluntary work (e.g. voluntary work, volunteers, unpaid work) and (b) determinants (e.g. determinant, factor, association, relation, reason) (see Appendix 1).

Articles were selected if they are (a) peer-reviewed full text publications reporting an association between at least one individual factor (contextual factors are beyond the scope of this study) and participation in formal voluntary work (i.e. voluntary work carried out for organizations [12]) (yes/no) in a quantitative way using a longitudinal prospective cohort study design (i.e. studies in which the determinant is measured at a moment in time before the outcome was measured), and (b) making use of a study sample consisting of adults aged 18 and over from a general population from a developed country (i.e. Japan and countries in Europe, North America and Oceania). Moreover, (c) the article has to be published in English, French, German or Dutch within the time period 2010–2015. Given the large number of publications on the topic, we decided to focus on recent publications from 2010 onwards. Articles exclusively including informal volunteering as the outcome were excluded. In case it is unclear whether volunteering was formal or informal, articles were included and labelled as ‘mixed type of voluntary work’. Finally, articles focusing on very specific cases of volunteering such as disaster volunteering, corporate volunteering and volunteer-tourism were excluded as well, because of their limited comparability with volunteering in the general population, but also because the motives to participate in these kinds of voluntary work may differ from situation to situation. We focus on longitudinal rather than on cross-sectional studies, as the former offer better opportunities for temporal ordering of factors.

The titles and abstracts of all identified records were screened for eligibility by two reviewers (J.N. and L.v.L.) independently. Subsequently, the same two reviewers independently screened the full-text of all potentially eligible articles. Finally, all references of included articles were screened by one reviewer (J.N.) for potentially eligible articles.

Two reviewers (J.N. and L.v.L.) independently extracted the data regarding the characteristics of the study sample (country, mean age, % female, inclusion criteria), the year of baseline measurement, study duration, determinant measurement, outcome measurement, sample size, volunteering at baseline (%), volunteering at follow-up (%), and the results (association between the determinant(s) and the outcome). The same two reviewers independently assessed the risk of bias of the included articles by using the QUIPS (Quality In Prognosis Studies) tool [13]. The following domains were assessed as potential sources for risk of bias: study participation, study attrition, measurement of the determinants and the outcome, study confounding and statistical analysis and reporting (see Appendix 2). Overall disagreement was evaluated and expressed as percentage of agreement and kappa statistics [14]. In a consensus meeting disagreements were discussed and resolved. If consensus could not be reached, a third reviewer (N.Sm.) made the final decision.

In case the results of at least two studies are available, meta-analyses were conducted, using the statistical program Comprehensive Meta-Analysis (3rd version). If studies present several models, estimates from the most complete (fully adjusted) model were used. Odds Ratios (ORs) with 95% Confidence Intervals (CIs) were used, or if needed calculated using the supplemental material of Kuiper et al. [15], to conduct meta-analyses. When insufficient information was available for transforming effect sizes to ORs with 95% CIs, study authors were contacted to obtain the missing information.

In case articles used the same study sample, a-priori defined criteria were used to select the study for the meta-analysis. In order of importance and for each determinant separately, articles were selected based on (a) outcome used in the study (‘formal voluntary work’ was preferred above ‘mixed type of voluntary work’), (b) measurement of the determinant (the determinant measurement was most comparable to other included studies), (c) study sample (the study sample that was the most comparable to the study samples of included studies in the meta-analysis, in terms of the proportion of volunteers at baseline, the age range of participants at baseline, and inclusion criteria for the baseline study sample), (d) sample size (the study with the largest sample size was preferred over smaller studies), and (e) number of determinants quantitatively measured in the study. In case articles presented both a static (e.g. being married) as well as a change score (e.g. transition into marriage) for a certain determinant, the score that is most comparable to the scores used in other included studies for this determinant was used. A random effect method was applied to calculate pooled effect sizes [10].

Heterogeneity between studies was assessed by using the Index of Inconsistency (I²) [16]. In case of substantial heterogeneity (I² > 50%), sources of heterogeneity between studies were explored by conducting either subgroup analysis (in case < 10 studies are available) or univariable random-effects meta-regression [10] (in case ≥10 studies are available) with regard to the following a-priori defined criteria: (a) outcome measurement (formal voluntary work versus mixed measure); (b) determinant measurement, based on (b1) measurement scale (continuous versus dichotomous scores), (b2) type of measurement (static versus change scores, because the presence of a certain event, (e.g. being married), may have a different association with the outcome than the transition into a certain event (e.g. transition into marriage)), and (b3) conceptual differences in the measurement of the determinant; (c) proportion of volunteers in the baseline study sample; (d) mean age at baseline, because some determinants may be important to a different extent for study samples for which participation in paid work is more or less common; (e) continent in which the study was performed (United States of America (USA), Europe, other), because differences in government regimes and culture may influence the association between a certain determinant and the outcome; (f) year of baseline measurement, because although the included studies were published between 2010 and 2015, the baseline measurement year varies substantially and determinants of participation in voluntary work may differ for different birth cohorts; (g) duration of follow-up (for time-variant variables only); and (h) the risk of bias for each methodological quality domain separately (low risk of bias versus high/unclear risk of bias).

The likelihood of publication bias was assessed graphically by constructing funnel plots for each determinant (in case at least ten studies were available) using the statistical program Comprehensive Meta-Analysis (3rd version). Asymmetry of the funnel plots was tested using Egger’s method. Publication bias is likely if p < 0.10 [17].

The results of the risk of bias assessment of included studies are presented in Table 2.

The risk of bias varied substantially. Most methodological flaws (i.e. high risk of bias) were found for (2a) adequate follow-up rate (62.5% high risk of bias), and (1b) adequate participation rate (29.2% high risk of bias). The inter-rater agreement was good (agreement 91.7% (484/528); kappa statistic: 0.78) [14].

Meta-analyses were conducted for a total of 20 determinants (see Appendix 4). For each determinant, all studies reporting an association between the determinant and the outcome are listed in the appendix, as well as the studies selected for inclusion in the meta-analysis.

Publication bias was assessed for the following determinants: age, gender, marital status and educational attainment. Eggers’ test and visual inspection of the funnel plots indicate that publication bias is likely for the determinants age (Egger’s test: age (p = 0.007) and marital status (p = 0.074)). The funnel plots are available upon request.

This review was conducted according to the latest standards for conducting systematic reviews and meta-analyses. This is the first review for which findings can be generalized to the general adult population in developed countries and for which associations between identified factors and participation in voluntary work were quantified by conducting meta-analyses. We transformed all estimates into ORs in order to compare the results of included studies. A thorough overview of all determinants of volunteering studied in recent publications is provided (i.e. demographic determinants, as well as determinants related to socioeconomic status, participation in other productive activities, health status, religion and social relationships), instead of focusing on a single determinant only (e.g. health status or socioeconomic status). Updating the current state of knowledge on factors related to volunteering was important, as the research on volunteering has taken a giant leap recently. Results from our search strategy showed that compared to a decade ago, publications on factors related to volunteering have more than doubled (our search resulted in 1620 hits for the period 2000–2005 compared to 3774 hits for the period 2010–2015).

Some limitations must be mentioned as well. We limited the inclusion of studies to those published in the period 2010–2015. The choice for including this quite narrow time period was made for two main reasons. Firstly, because participation in voluntary work is related not only to individual characteristics but also to macro factors such as the demographic composition of populations, economic circumstances and government regimes, we argue that taking into account the most recent time period is the most relevant period to study in order to increase our knowledge on contemporary determinants of volunteering and provide insight in the characteristics of potential volunteers nowadays. Secondly, in trying to find a good balance between recency and efficiency, we chose to develop a rather broad search strategy without specifying any determinants beforehand, in order to provide the most comprehensive overview of all determinants studied in relation to volunteering. We cannot be sure whether our results would have been different if all studies irrespective of the date of publication were to be included. Probably the results for the factors studied in this review would be more heterogeneous due to cohort effects and probably some additional factors studied in earlier publications may have been identified. Moreover, although the studies included in this review were all published recently, the vast majority (79%) of the included studies used data with baseline measurements before the year 2005. Potentially, the results would have been different if we would have limited the inclusion of studies to those using recent data. Our search was conducted in August 2015 and several relevant articles have been published afterwards [42‐46]. The results of this articles are in line with the results included in this systematic review and meta-analysis, showing for example that individuals in worse mental health are less likely to participate in voluntary work [42], religiosity in adulthood is positively associated to volunteering [43], and that previous volunteer experiences, good health and higher education are positively associated to volunteering [44]. Moreover, two studies investigated the association between providing care to grandchildren and volunteering [45, 46] but with opposite results. Therefore, it seems unlikely that including studies published after August 2015 would alter the conclusions drawn in the current study. Moreover, the inclusion of studies was limited to studies written in English, Dutch, French or German. The inclusion of only English, Dutch, French and German language studies may have led to missing some studies, however there is little evidence that exclusion of non-English-language studies leads to systematic bias in systematic reviews [47‐50].

Visual inspection of the forest plots and Egger’s test have shown the presence of funnel plot asymmetry for the studies investigating the factors age and marital status in relation to the likelihood to volunteer. Therefore, these results should be interpreted with caution, as the reported effect sizes might be an overestimation of the true effect due to publication bias. However, using Egger’s approach could lead to false-positive results in the case of dichotomous outcomes [10]. However, we do not consider publication bias to be very likely. The majority of the included studies took age and marital status into account as control variables and their main interest was often directed towards the association between other factors and volunteering. Therefore, we do not expect publication bias to be a substantial problem for the results presented in this review. Funnel plot asymmetry can be caused not only by publication bias, but low methodological quality could also lead to the inflation of effects in smaller studies [10]. The latter could play a role. The majority of the studies included in this review did not provide information on the characteristics of respondents compared to participants lost to follow-up and differences between these groups could have contributed to funnel plot asymmetry.

The studies included in this review were very heterogeneous in terms of the methodological quality and study population. Results were also heterogeneous and, unfortunately, heterogeneity could often not be properly explained. The risk of bias assessment of the included studies has pointed out the presence of reporting flaws in included studies. Although attrition was in general high in the included studies, the majority of the studies (79.2%) did not report information regarding potential differences between participants and drop-outs, therefore insufficient information is available to assess the likelihood of attrition bias in included studies. Moreover, more than half of the studies (54.2%) did not report information on the participation rate at baseline. These are important reporting flaws, because selectivity in the study sample could have a major influence on the findings. Finally, half of the studies (50.0%) did not report information on how missing data was dealt with. For correct interpretation of the findings, it is important to know whether and how data has been imputed. In future research, more attention should be directed towards the quality of reporting as recommended in the STROBE guidelines (39).

The focus of the current systematic review and meta-analysis was to investigate the individual determinants of volunteering. Other determinants play a role as well in predicting volunteering. Contextual factors, for example, are important determinants of volunteering too [51]. Anheier & Salomon [52] (page 43) described that volunteering is determined by the way how societies are organized, how they allocate social responsibilities, and how much engagement and participation they expect from citizens. The heterogeneity between the results of the included studies could also be attributable to contextual differences between countries or cultures. Therefore, in future cross-national research on individual determinants of volunteering, it would be interesting to take into account cultural and country specific aspects. Moreover, our review showed that most studies are concentrated in the USA and selected European countries. It would be important for future research to diversify studies in terms of geographical spread. Our review has provided evidence for the association between several factors (e.g. socioeconomic status, marital status, parental status, functional health, previous volunteering, social network size and religion) and participation in voluntary work. In future research on determinants of participation in voluntary work, these factors should thus be taken into account as potential confounders in the analyses. This review has identified several gaps in the literature as well. Firstly, (weak) evidence was found for the negative association between age and volunteering from middle-age onwards. Studies including adults below middle-age were substantially underrepresented in this review. Therefore, more research should be done to examine the determinants of participation in voluntary work among younger adults and more specific, on the association between age and volunteering in younger age groups.

Secondly, more research is needed on the association between socioeconomic status and volunteering. Socioeconomic status seems to be related to the likelihood to volunteer; weak evidence for the association between education and volunteering was found and, although the overall result for the association between income and volunteering was inconclusive, for specific groups of individuals (aged 55 and over and those living in the USA) we did find a positive association. Another related factor is employment status. We did not find an association between employment status and volunteering but the pooled estimate was boundary significant and indicates the presence of a possible negative association between employment status and the likelihood to volunteer. Studies assessing the association of participation in voluntary work with employment status, level of income and education were very heterogeneous with respect to the confounders they took into account. Further research should investigate the association between these factors and participation in voluntary work, taking both educational attainment, employment status and income level into account as not only these factors themselves but also the interplay between these factors may be important in predicting participation in voluntary work.

Thirdly, this review has shown the importance of two types of life course transitions in predicting the likelihood to volunteer. For both increases in the degree to which an individual is functionally limited as well as the recent birth of a child in the household, a strong negative association with participation in voluntary work was found. Regarding parenthood, the recent birth of a child is negatively associated to volunteering whereas the presence of children in the household in general seems to be positively associated to volunteering, which shows the importance of disentangling these factors. Despite the evident importance of life course transitions in predicting the likelihood to volunteer, the majority of studies included in this review did not take them into account. In future research, the effect of major life course transition with respect to family life (for example changes in household composition, partnership status and health of family members), work (for example starting a career after graduation, transitions into and out of unemployment, changes in working hours and retirement) and health should be taken into account.

New research in the field of volunteering should aim at filling the gaps mentioned above, because volunteering is an increasingly important activity for developed societies facing aging populations. Therefore, it is important to know for policy makers which characteristics are related to volunteering in order to identify potential volunteers.