The Relationship Between Past and Present Volunteering Activity

People tend to consistently engage in charitable activity over time, and indeed, past volunteering is one of the strongest predictors of future volunteering (Mutchler et al. Reference Mutchler, Burr and Caro2003; Thoits and Hewitt Reference Thoits and Hewitt2001). The few studies that have examined changes in volunteering activity for the same individuals over time tend to support this notion of persistent activity. For instance, using two waves of data from a nationally representative longitudinal survey—The American’ Changing Lives Study— Wilson and Musick (Reference Wilson and Musick1997) and Thoits and Hewitt (Reference Thoits and Hewitt2001) found a high level of stability in volunteer activity participation. Furthermore, Butrica et al. (Reference Butrica, Johnson and Zedlewski2009) and Oesterle et al. (Reference Oesterle, Johnson and Mortimer2004) using longitudinal data from the Health and Retirement Study and Youth Development Study, respectively, found considerable stability among volunteers and non-volunteers.Footnote ³ However, what these studies fail to address, and what forms the basis of our investigation, is that persistent volunteer activity participation is consistent with two mutually inclusive possibilities: ‘true’ and ‘spurious’ state dependence.

Firstly, stable and persistent volunteering behavior may reflect a causal influence of past behavior on current behavior. A consequence of experiencing an event (engagement in volunteering) is that preferences or any constraints relevant to future behavior may be altered (Heckman Reference Heckman1981). Individuals who have engaged in volunteering behavior in the past are therefore more likely to engage in that same behavior in the future. There are several possible mechanisms through which this causal effect may operate. Here, we focus on two. The first is linked to theories of social capital. Participation in volunteering activities enhances social networks, social ties (including friendship networks and organizational memberships) and, importantly, connections to others in social institutions. These ties make subsequent volunteer activity participation more likely by fostering trust and norms of generalized reciprocity, and by creating obligations and providing support (Putnam Reference Putnam1995). Moreover, enhanced connections to others in social institutions, as well as increasing the likelihood that one will be invited to engage in civic activities (Brady et al. Reference Brady, Schlozman and Verba1999), also increase information and access for further volunteering opportunities (Oesterle et al. Reference Oesterle, Johnson and Mortimer2004). Similarly, connections to others in social institutions, insofar as these institutions foster the development of civic skills and a sense of civic mindedness, are indicative of subsequent volunteering activity (Brady et al. Reference Brady, Schlozman and Verba1999; Mutchler et al. Reference Mutchler, Burr and Caro2003; Oesterle et al. Reference Oesterle, Johnson and Mortimer2004). The second mechanism is linked to the ‘warm glow’ hypothesis, perhaps the most influential model of why people engage in pro-social behavior (Andreoni Reference Andreoni1989, Reference Andreoni1990). Here, pro-social behavior enters the utility function as individuals derive positive emotional experiences, a ‘warm glow,’ from the act of helping others (Andreoni Reference Andreoni1989). Individuals may also have a desire to win prestige, respect and enhance their social image (Andreoni Reference Andreoni1990; Sanghera Reference Sanghera2016). Therefore, pro-social behavior can also be motivated by a desire to receive social acclaim or avoid scorn from others. These factors which influence the decision to act prosocially exhibit selfish elements and are therefore generally considered to be models of ‘impure altruism.’ There now exists a substantial body of research suggesting that pro-social behavior has many physiological and psychological benefits (Borgonovi Reference Borgonovi2008; Mellor et al. Reference Mellor, Hayashi, Stokes, Firth, Lake and Staples2009; Musick and Wilson Reference Musick and Wilson2003; Post Reference Post2005). Therefore, when individuals derive a positive emotional experience from the act of helping other people (Andreoni Reference Andreoni1989, Reference Andreoni1990), this ‘warm glow’ may increase the likelihood of acting in a pro-social way in the future (van der Linden Reference van der Linden2015). After all, ‘once you have a certain new experience, you need to keep on having more of it if you want to sustain your happiness’ (Layard Reference Layard2011). While individuals may directly receive a ‘warm glow’ or ‘helpers high’ from volunteering—when the act of helping others actually makes people feel good—the psychological benefits of volunteering may also be derived indirectly from enhanced connectedness, social contact and a sense of worth and status (Andreoni Reference Andreoni1990; Nichols and Ralston Reference Nichols and Ralston2011; Son and Wilson Reference Son and Wilson2012). This type of temporal persistence, where there is a causal effect of past behavior on current and future behavior, is referred to as ‘true’ state dependence.

Alternatively, persistence in behavior may be driven by personality traits and other stable characteristic behaviors or values that are not always readily observable to researchers (Brown et al. Reference Brown, Srivastava and Taylor2015; Janoski and Wilson Reference Janoski and Wilson1995). For instance, a higher tendency to volunteer has been found to be associated with altruistic values (Hodgkinson Reference Hodgkinson2003), higher openness to experiences (Binder and Freytag Reference Binder and Freytag2013) and higher agreeableness (Carlo et al. Reference Carlo, Okun, Knight and de Guzman2005). It is generally assumed that these characteristic behaviors are established in pre-adult life through intergenerational transmission mechanisms and other early socialization experiences and remain largely unaltered during adulthood (Janoski et al. Reference Janoski, Musick and Wilson1998). To the extent that these often unobservable factors are persistent over time, they will induce persistence in volunteering behavior. Past volunteering behavior may therefore appear to have a causal influence on future volunteering behavior by simply picking up the effect of permanent unobserved individual heterogeneity. This mechanism is commonly referred to as ‘spurious’ state dependence. An understanding of the relative magnitude of ‘true’ and ‘spurious’ state dependence has important consequences for policy design.

If volunteering behavior exhibits (at least in part) ‘true’ state dependence, then public policy designed to influence volunteering today will simultaneously be an investment in future volunteering behavior. For example, if volunteering is truly state dependent, a government policy that is able to turn a non-volunteer into a volunteer at a given point in time will induce a permanent change in this individual’s future volunteering behavior. Policies such as the UK government’s introduction of the National Citizen Service for all young people between 16 and 17 years of age will therefore not only be an investment in current volunteering, but also by generating the foundation necessary to induce ongoing pro-social behavior, a strong investment in future volunteering (McCulloch Reference McCulloch2014). Conversely, if permanent individual heterogeneity is the main source of persistent behavior, then even government support programmes which are successful in turning a non-volunteer into a volunteer at a given point in time are unlikely to durably influence the individual’s future volunteering participation. Unless, of course, these support programmes focus on measures which have the potential to improve volunteering-relevant permanent individual-specific characteristic behaviors.

Data Source and Descriptive Statistics

The data used for analysis are taken from the British Household Panel Survey (BHPS), a nationally representative survey of more than 5000 households which contains approximately 10,000 individuals aged 16 and over. The survey instrument is a questionnaire involving a household section, and individual sections, covering a range of topics including household composition, housing characteristics, education and training, health, labor market status and values and opinions on social and political matters. The questionnaire is administered to all adult household members (including new household members at each wave). Repeat interviews take place annually, with 18 annual waves available to researchers between 1991 and 2008. The data used in the subsequent analysis are restricted to the original BHPS sample covering Great Britain. The dependent variables in all analyses that follow are responses from this survey instrument:

The question was administered in the 1996, 1998, 2000, 2002, 2004, 2006 and 2008 waves of the BHPS. From this question, we create two key variables. First, we create a binary variable of engagement in voluntary work (VE):

• • Voluntary Work Engagement = 0 if reported ‘never/almost never.’

• • Voluntary Work Engagement = 1 if reported ‘once a year or less’ or ‘several times a year’ or ‘at least once a month’ or ‘at least once a week.’

Secondly, we create an ordinal variable to measure voluntary work frequency (VF) that is consistent with the ONS (2017) definition of frequent volunteers:

• • Voluntary Work Frequency = 0 if reported ‘never/almost never.’

• • Voluntary Work Frequency = 1 if reported ‘once a year or less’ or ‘several times a year.’

• • Voluntary Work Frequency = 2 if reported ‘at least once a month’ or ‘at least once a week.’

The sample used for our analysis is limited to only those individuals who were observed in all seven waves and who had valid responses to the dependent and independent variables used. This yields a final balanced panel of 4323 individuals with 30,261 individual-year observations. Table 1 reports the descriptive statistics and includes the control variables used in the subsequent multivariate analysis. Approximately 77 percent of the balanced sample reports ‘never/almost never’ engaging in voluntary work, while 7.3 percent reports engaging at least once a week. The mean age of the balanced sample is approximately 49 years. Just over 56 percent of the sample is female, 14 percent report holding a university degree with 19.8 percent reporting leaving compulsory schooling with no formal qualifications. Lastly, 63.2 percent of the sample is currently in some form of employment.

Table 1

Descriptive statistics

Variables	Mean/frequencies	SD
Dependent variable
Voluntary work
Never/almost never	0.768
Once a year or less	0.053
Several times a year	0.054
At least once a month	0.052
At least once a week	0.073
Control variables
Female	0.562
Age	49.120	15.750
White	0.971
Self-employed	0.075
Employee	0.557
Unemployed	0.019
Retired	0.219
Family care	0.005
Economically inactive	0.125
Hours worked	21.670	19.120
Married	0.751
Widowed/divorced/separated	0.138
Never married	0.111
Spouse/partner employed	0.504
Number of dependent children in the household	0.553
Household size	2.763	1.273
University degree	0.139
Vocational college qualification	0.314
A-level	0.098
O-level/GCSE’s	0.168
Other qualifications	0.083
No qualifications	0.198
Number of cigarettes smoked	3.330	7.420
General health	0.082
Log household income	7.695	0.755
Own house outright	0.316
Own house with mortgage	0.494
Rents house, private sector	0.058
Rents house, social sector	0.132
Observations	30,261
Number of individuals	4323

To illustrate the persistence of volunteering revealed in the seven waves of data, we first analyze two Markov chains for our binary and ordinal indicators of voluntary work. For our binary measure, with the two possible states of volunteering engagement, $VE=\left(0,1\right)$ , the transition matrix is given by:

$P=\begin{array}{cccc}& 0& 1& \\ 0& 87.39& 12.61& n=25,938\\ 1& 39.95& 60.05& \end{array}$

Here, the rows indicate the previous volunteering engagement behavior, while the columns indicate current volunteering engagement behavior. This is illustrative of considerable persistence in volunteering behavior. For instance, the probability of volunteering conditional on volunteering in the previous period, $\text{Prob}(V{E}_t=1|V{E}_{t-1}=1)$ , is 60.05 percent. Moreover, the probability of not volunteering conditional on not volunteering in the previous period, $\text{Prob}(V{E}_t=0|V{E}_{t-1}=0)$ , is 87.39 percent. Prior volunteering also increases the likelihood of current volunteering from 12.61 percent to 60.05 percent, or alternatively, by 47.44 percentage points.

For our ordinal measure of frequency, with the three possible states of volunteering frequency, $VF=\left(0,1,2\right)$ , the transition matrix is given by:

$P=\begin{array}{ccccc}& 0& 1& 2& \\ 0& 87.39& 7.61& 5.00& n=25,938\\ 1& 55.09& 29.15& 15.75& \\ 2& 26.51& 15.68& 57.81& \end{array}$

The rows continue to indicate the previous volunteering frequency behavior, while the columns indicate current volunteering frequency behavior. Again persistence is observable. For the most extreme cases, $VF=\left(0,2\right)$ , it is clear that the probabilities of transitions to the highest volunteering frequency from the lowest volunteering frequency, or the reverse, are very small. Consequently, individuals are more likely to remain close to their prior volunteering frequency state than adjust significantly from it. For the moderate frequency case, $\left(VF=1\right)$ , although the highest probability is the transition to not volunteering, important persistence is still observed: $\text{Prob}(V{F}_t=1or2|V{F}_{t-1}=1)$ = 44.91 percent.

An alternate representation of persistence in volunteering behavior is presented in Fig. 1—illustrating the distribution of the individual variability in volunteering engagement and volunteering frequency by utilizing the sum of the absolute values of movements from one wave to the next. Approximately 46.5 percent of individuals experienced no change in their volunteering engagement across the seven waves of data, with 194 individuals recording engagement in voluntary work in every wave. To put this in perspective, consider a seven-period game where people are assigned to either participate or not participate in voluntary work. Behavior is determined on sequential independent draws from a binomial distribution across the seven periods. Assuming that the probability of engaging in voluntary work is 0.23 (the average probability of engaging in voluntary work in our dataset), for our sample of 4323 individuals, we would expect $\left(0.{23}^7\right)\times 4323=0.15$ individuals to record voluntary work engagement in all seven periods. Similarly, for volunteering frequency, 43.7 percent of individuals experienced no change in their volunteering frequency across the seven waves of data with only 21.1 percent of individuals exhibiting three or more changes. In summary, the raw data provide strong evidence that prior volunteering behavior is an important predictor of current behavior. In extending the previous research, what remains is to decompose the ‘persistence’ observed in the raw data into that which can be explained by unobservable permanent heterogeneity (‘spurious’ state dependence) and ‘true’ state dependence.

Fig. 1

Frequency distribution of the observed variability in volunteering engagement and frequency

Econometric Strategy

To formally model the persistence of volunteering, we use a dynamic random effects probit model to decompose the ‘persistence’ observed in the raw data into that which can be explained by unobservable permanent heterogeneity (‘spurious’ state dependence) and ‘true’ state dependence. The model is then extended to an ordered probit to analyze for the frequency measure of volunteering behavior. This methodology has been used extensively in the economics discourse for accessing health mobility (Contoyannis et al. Reference Contoyannis, Jones and Rice2004), within marketing for testing consumer choice dynamics (Erdem and Sun Reference Erdem and Sun2001) and more recently within sociology for testing the persistence of generalized trust beliefs (Dawson Reference Dawson2017). The general form of the dynamic probit model for volunteering engagement can be written as follows:

(1) $V{E}_{\mathrm{it}}^{\ast }=\delta V{E}_{it-1}+{\beta }^{\prime }{X}_{\mathrm{it}}+{\alpha }_i+{\epsilon }_{\mathrm{it}},\left(i=1,\dots ,N;t=2,\dots ,T_i\right)$

where $V{E}_{\mathrm{it}}^{\ast }$ is the individual’s latent probability of volunteering in each year of the sequence of $T_i$ . $V{E}_{\mathrm{it}}$ is a binary indicator that takes on the value of one in each year $t$ ( $V{E}_{\mathrm{it}}=1$ ) when an individual is observed to have engaged in volunteering, which occurs when his/her propensity to volunteer exceeds a threshold (zero in this case). Correspondingly, $V{E}_{it-1}$ is the indicator for the individual’s volunteering behavior in the previous period, and $X_{\mathrm{it}}$ is a vector of sociodemographic and socioeconomic control variables. The remaining variation in volunteering behavior is represented by ${\alpha }_i+{\epsilon }_{\mathrm{it}}:{\alpha }_i$ is an unobservable individual-specific attribute (random effect), and ${\epsilon }_{\mathrm{it}}$ is an idiosyncratic error term which captures the effect of time-varying unobservable determinants. Both are assumed to be normally distributed with a mean of zero. The variance of ${\epsilon }_{\mathrm{it}}$ is normalized to one, and the variance of ${\alpha }_i$ estimated by the model.

Two issues arise from this standard random effects model. Firstly, it assumes that ${\alpha }_i$ and $X_{\mathrm{it}}$ are uncorrelated with each other. Secondly, because a dynamic model is estimated an ‘initial conditions’ problem arises as $V{E}_{i1}$ is correlated with ${\alpha }_i$ which then induces a correlation between ${\epsilon }_{\mathrm{it}}$ and $V{E}_{it-1}$ and leads to a bias in the estimated parameter. To address the initial conditions problem in estimating dynamic models and to allow ${\alpha }_i$ to be correlated with the regressors, we follow the respective approaches laid out by Wooldridge (Reference Wooldridge2005) and Mundlak (Reference Mundlak1978). Specifically, we specify a model that assumes ${\alpha }_i$ is both correlated with the regressors and the initial endowment of volunteering. This approach is implemented by parameterizing the individual effect as:

(2) ${\alpha }_i={\alpha }_0+{\gamma }^{\prime }{\overline{X}}_i+\phi V{E}_{i1}+u_i$

where ${\overline{X}}_i$ represents the individual time means of all the time-varying control variables, $u_i$ is the individual effect which is assumed to be distributed $N\left(0,{\sigma }_u^2\right)$ , and $V{E}_{i1}$ is the individual’s initial volunteering behavior. Substituting Eq. (2) into (1) provides the full model as shown in Eq. (3). The parameter $\delta$ measures ‘true’ state dependence and therefore, the extent to which past volunteering behavior is passed on to both contemporary and future volunteering behavior. At the two extreme cases, an exogenously determined change in prior volunteering behavior (shock) will either be permanently passed on to future volunteering behavior $\left(\delta =1\right)$ or alternatively, shocks will fully dissipate $\left(\delta =0\right)$ and the individual will revert immediately to his or her baseline volunteering behavior. The usefulness of a distinction between absolute and partial persistence is illustrated in Fig. 2. Here, an individual has a baseline propensity to volunteer in each period, but faces an observable exogenous shock in volunteering behavior between time $t-2$ and $t-1.$ This shock will then be passed on to the individual’s behavior at time $t$ , through either total, partial or nonpersistence. The estimate of $\phi$ in Eq. (3) is also relevant as it provides information about the correlation between the individual effect and the individual’s initial volunteering behavior.

(3) $V{E}_{\mathrm{it}}^{\ast }=\delta V{E}_{it-1}+{\beta }^{\prime }{X}_{\mathrm{it}}+{\gamma }^{\prime }{\overline{X}}_i+\phi V{E}_{i1}+u_i+{\epsilon }_{\mathrm{it}},\left(i=1,\dots ,N;t=2,\dots ,T_i\right)$

Fig. 2

Adjustment path of an exogenous shock to volunteering engagement

In order to account for the heterogeneity in sociodemographic and socioeconomic factors that have been shown to influence volunteering behavior, standard control variables are included in the regression model. Basic demographic variables include: age, gender, education and ethnicity. In general, volunteering increases with younger- and middle-aged cohorts and decreases in older groups as health becomes an issue (Einolf Reference Einolf2009). The young may volunteer to increase employment prospects and other work-related outcomes (Johnson et al. Reference Johnson, Beebe, Mortimer and Snyder1998), while older volunteers tend to be motivated to a greater extent by service or community concerns (Omoto et al. Reference Omoto, Snyder and Martino2000). Gender is included as an important control as women tend to be more altruistic and pro-social than men (Helms and McKenzie Reference Helms and McKenzie2014; Mesch et al. Reference Mesch, Rooney, Steinberg and Denton2006). Education has been found as one of the strongest predictors of engagement in volunteering (Son and Wilson Reference Son and Wilson2012) with more educated individuals having the highest levels of volunteer engagement (Dekker and van den Broek Reference Dekker and Van Den Broek1998; Lancee and Radl Reference Lancee and Radl2014; Mutchler et al. Reference Mutchler, Burr and Caro2003). Household structure is another important determinant of volunteering and is captured by the following variables: log-transformed household income, marital status, housing tenure, the number of dependent children in the household and household size. These household variables affect the resources (time and/or money) available to give to philanthropic organizations.

Health status, an important determinant in volunteering studies, is also included and is captured by a binary variable capturing self-reported general health over the last 12 months. The binary health variable captures individuals with ‘excellent’ or ‘good’ health as compared to ‘fair,’ ‘poor’ or ‘very poor’ health. We also include as a further indicator of health, the number of cigarettes smoked per day. Poor health has been found to be one of the most important factors acting as a barrier to volunteer engagement (Sundeen et al. Reference Sundeen, Raskoff and Garcia2007). Economic activity is captured through a series of dummy variables capturing heterogeneity in employment and economic inactivity as well as a continuous measure of usual weekly working hours. While work commitments limit an important resource related to volunteering behavior (Mutchler et al. Reference Mutchler, Burr and Caro2003; Sundeen et al. Reference Sundeen, Raskoff and Garcia2007), individuals who work longer hours may give more time, in a ‘workaholic’-like commitment (Taniguchi Reference Taniguchi2006). In general, those who are in full-time employment are more likely to volunteer (Martinez and McMullin Reference Martinez and McMullin2004).