«IZA DP No. 8764 PAPER Voluntary Activities and Daily Happiness in the US J. Ignacio Gimenez-Nadal DISCUSSION Jose Alberto Molina January 2015 ...»
5. Religion, working days, and daily happiness Various studies have highlighted the link between religion/religiosity and happiness. For instance, Helliwell (2003) finds greater life satisfaction associated with church attendance of once, or more often, a week. In Eastern Europe, Hayo (2004) finds similar results, and using ESS data Clark and Lelkes (2005) report that church attendance of at least once a month is sufficient for an effect on life satisfaction. Furthermore, as shown in Figure 1, a disproportionate share of voluntary activities take place on Sunday, which happens to be a day of high church attendance for religious individuals in the United States. If we consider the distribution of individuals in our sample according to whether they went to church or places or worship during the diary day, the greater proportion of individuals going to church is obtained on Sundays (see Figure 2). We find a positive correlation between the probability of participation in voluntary activities and the probability of going to church or a place of worship during the day (R=0.394, p0.01). Thus, it could be that people who do voluntary activities during the day also go to church or places of worship, and thus the greater daily happiness experienced by those who volunteer can be explained to some extent by the fact that they went to church.
First, we have created a dummy variable to indicate whether the diary corresponds to week days (1), or to weekend (0). The reason is that those who report going to the church on the day of the survey, 13.62%, did so on Saturday, and 69.65% did so on Sunday; thus, those who went to the church did so on the weekend in 83.27% of cases. If religious activities drive the difference in daily happiness between volunteers and non-volunteers, we should expect to find a closer association between participation in voluntary activities and daily happiness on the weekend. Thus, we include the interaction between participation in voluntary activities and the weekday dummy, to see if the difference between volunteers and non-volunteers varies at the weekend. Panel A of Table 5 shows the results of estimating Equations (1) and (2) when we include voluntary activities in the regressions (Columns (1) to (4)), and when we exclude voluntary activities from the regressions (Columns (5) to (8)). 9 We find that individuals obtain lower daily happiness (a lower net-affect and a higher u-index) during the week, consistent with the fact that individuals work normally during these days, while they have more leisure time on weekends. However, we find no differential association between voluntary activities and daily happiness according to the weekend/week day distinction, and thus no evidence of a weekend effect.
Second, we have modified our dependent variable, and we now include religious activities as part of voluntary activities. Thus, we have created a dummy variable to indicate participation in voluntary activities and/or religious activities. If the association between voluntary activities and daily happiness is driven by religious activities, the inclusion of the latter should lead to almost identical results, compared to the results considering voluntary activities only (Table 3).
Results are shown in Panel B of Table 5. We obtain statistically significant coefficients for all of our regressions. It is important to note that the coefficients are larger compared to the coefficients shown in Table 3, where we only include participation in voluntary activities, indicating that when we include religious activities in our definition of voluntary activities, we obtain greater differences in the daily happiness of individuals, consistent with religious activities providing a high level of happiness (see Table 2).
Third, we have controlled for whether the individual went to church at any time during the day (1) or not (0), including this variable in the regressions. We have also included an interaction term between participation in voluntary activities and church attendance. If this interaction term is statistically significant, it would mean that church attendance, and the religiosity of individuals, is important in the relationship between daily happiness and voluntary activities. Results are shown in Panel C of Table 5, and we still find a statistically significant relationship between voluntary activities and daily happiness. The interaction term is not statistically significant at standard levels, indicating that religion/religiosity does not drive an association between voluntary activities and daily happiness.
Another posible channel through which participation in voluntary activities and daily happiness could be related is market work. It could be that those who volunteer on the day of the survey do not have to work, or work fewer hours on that day, compared to non-volunteers.
In fact, the analysis of the diaries considering whether the diarist devoted some time to market work, or not, on the day of the survey, shows that 65% of those who devoted time to voluntary activities did so on non-working days. On a day off, they can enjoy other activities, like personal care or leisure, much more than on a workday, and the fact that individuals do not have market work on that day indicates that individuals have the freedom to choose any activity they Given that we control for whether the diary corresponds to the weekend, or to weekdays, we do not include daydummy variables.
want, and thus some may choose to volunteer while others may choose to have more leisure.
Thus, we propose to analyze differences between volunteers and non-volunteers on nonworking days (i.e. no time on market work and commuting), to see if there are differences in the daily happiness according to participation in voluntary activities. Panel D of Table 5 shows the results of estimating Equations (1) and (2) on daily happiness, considering whether the individual participated in voluntary activities during the non-working day. Our main results are maintained, that those who volunteer on their days off report a higher experienced utility. We interpret these results as that, despite volunteers devoting less time to market work during the day, the difference does not depend on the fact that people who volunteer had to work less, or did not work at all, on the particular day, and thus we disregard this possible channel.
6. Accounting for reverse causality between volunteering and daily happiness One problem when analysing the effects of voluntary activities on daily happiness is that of reverse causality (people volunteer more when they are happy) and simultaneity biases (some third factor leads to more volunteering and more daily happines). Many prior studies are correlational as they do not take into account these problems, and those studies that have done so have found that the effect of volunteering on happiness decreases. In this Section, we develop an Instrumental Variable (IV) estimation to deal with the issue of reverse causality, to see if those who are happier are also more likely to participate in voluntary activities.
To that end, we regress the possible endogenous variable included in
Equations (1) and (2) on a set of excluded (IVi) and included instruments (Xi, and Dayi). The 2equation system can be written as follows:
(3) (4) where represents the happiness measure of individual “i” in episode “j”, is a dummy variable that indicates whether respondent “i” engaged in any voluntary activity (1) or not (0) during the day, represents a vector of socio-demographic characteristics, and represents the error terms. We additionally estimate Equation (4) where we regress participation in voluntary activities on a set of instruments, and the other socio-demographic characteristics included in Equation (3). Regarding the instruments, they must fullfill rank and orthogonality conditions, such that 1) the instruments must be correlated with the variables to be instrumented, and 2) the instruments must be uncorrelated with whatever is not observed that is a determinant of (Greene, 2008; Angrist and Pischke, 2009; Woolridge, 2010). Intuitively, a reliable instrumental variable must be first theoretically justified and statistically correlated, after controlling for all other exogenous regressors, with the endogenous variable of interest, and it must be exogenous to all other important and unobserved factors (i.e. uncorrelated with the disturbance term in the outcome equation).
The literature has identified differences and changes in laws as suitable instruments. Angrist and Krueger (1991) instrumented education using compulsory schooling laws, or state laws on minimum legal drinking age, and the state beer tax to instrument substance use (see French and Popovici, 2011, for a review). Under this framework, we use cross-state differences in laws on deductions for charitable gifts as our main instrument. We consider that cross-state differences in laws on deductions for charitable giving allow us to identify participation in voluntary activities, despite that the direction of the relationship is unknown a priori. In states that have an income tax and a charitable deduction, there is a greater incentive to donate to charity, but the opportunity cost of time is also lower because the income tax reduces the opportunity cost of time, assuming some of the incidence falls on the worker. In states with an income tax but without deduction for charity, the reduction in opportunity cost of time would be present, but not the incentive to make charitable contributions (Gale and Scholz, 1994).
We use the information on which states have a deduction for charitable contributions, and which ones have no such deduction, obtained from the National Bureau for Economic Research (http://users.nber.org/~taxsim/charity-state/). From this web site, we obtain an indicator for charitable deductions that takes value “1” for deduction, value “0” for none, and “-1” for no tax due, witn this information being available at the state level for the year 2010, the year of reference. We propose this indicator as an instrument for voluntary activities.10 Looking at the Spearman’s rank correlation between participation in voluntary activities and this instrument, we observe that Spearman’s rho is 0.0266, and the null hypothesis that the variables are independent is rejected (p0.01). Thus, we can assume that the two variables are correlated. On the contrary, and considering the relationship between the net-affect and u-index, on the one hand, and the instrument, on the other, the Spearman’s rho are 0.0051 and -0.0086 respectively, thus accepting the null hypothesis that the the net-affect (p=0.429) and u-index (p=0.1681) are independent of the instrument.
Table 6 shows the results of estimating equations (3) and (4), for both the net-affect and uindex. 11 We present estimations according to two types. Columns (1) and (2) present results for the net-affect and u-index when we use the Two-Stage Least Squares (2SLS) estimator.
Columns (3) and (4) present results for the net-affect and u-index when we use the Generalized Two-Stage Least Squares random-effects (G2SLS) estimator from Balestra and VaradharajanSee table A3 in the Appendix for a description of the values of the indicator for each state.
We present results only for Equation (4), which is the equation of interest. Results for the first stage regressions are available upon request.
Krishnakumar (1987). The first-stage estimates of the IV in the bottom panel show that the instrument is statistically different from zero at conventional levels in all regressions.
Additionally, if the instrument fails to explain a sufficient amount of the variation in the endogenous regressor, it can generate IV estimates with large standard errors, as well as lead to large asymptotic biases (Bollen et al., 1995; Bound et al., 1995; Staiger and Stock, 1997), and thus we have to look at the “weakness” of the instrument (Stock et al., 2002). An F-statistic above 10 is commonly viewed as the threshold (Staiger and Stock, 1997; Stock et al., 2002), and in our case the F-Statistic in the 2SLS is well above this value (27.05).
We observe that coefficients for participation in voluntary activities are not statistically significant at standard levels in any of the columns in Table 6. Thus, once we take into account the reverse causality, we find that volunteering has no causal effect on the daily happiness of individuals. The interpretation of these results is that voluntary activities do not make individuals happier, but that happier individuals are more likely to participate in voluntary activities, in such a way that there is selection into participation in voluntary activities.
In another set of regressions, we have instrumented participation in voluntary activities using the number of nonprofit institutions per 10,000 inhabitants at the state level, for the year 2010 (information obtained from the National Center for Charitable Statistics http://nccsweb.urban.org/). We consider this instrument is able to capture differences at the state level in the level of altruism, with a larger number of institutions indicating that the level of altruism is higher in the state, and we suppose that a higher level of altruism is related to a greater participation in voluntary activities. Columns (5) and (6) present results for the net-affect and u-index when we use the (2SLS) estimator, and Columns (7) and (8) present results when we use the G2SLS random-effects estimator.
The first-stage estimates of the IV in the bottom panel show that both instruments are statistically different from zero at conventional levels in all the regressions. Additionally, the FStatistic in the 2SLS is 14.94, above the threshold value of 10. Furthermore, when we have two instruments, we can apply Sargan’s test of overidentifying restrictions in the 2SLS estimator.