4.1. Main Results
Table 6 reports Standard and Heteroscedastic Ordered Probit regressions for satisfaction and happiness without including religion-religiosity interaction terms. In comport with the literature on the subject, consumption per capita, livestock ownership, general trust, government trust, participation in formal and informal institutions, and religiosity all positively predict life satisfaction. Consumption per capita, livestock ownership, and general trust positively correlate with momentary happiness.
However, as discussed in preceding sections religiosity levels are correlated with religious denominations. Moreover, since some of the denominations, typically Catholicism and Protestantism, are relatively new to the country and a minority, the implication of religiosity on wellbeing can depend on the type of religion. Hence, the inclusion of ‘religion-religiosity’ interactions terms is essential. Hence, our analysis relies on the estimates with the interaction terms.
Table 6
Determinants of SWB: Without Religion-Religiosity Interaction Terms
|
Standard Ordered Probit
|
Heteroscedastic Orderd Probit
|
|
Satisfaction
|
Happiness
|
Satisfaction
|
Happiness
|
|
Coef.
|
St.Er.
|
Coef.
|
St.Er.
|
Coef.
|
St.Er.
|
Coef.
|
St.Er.
|
Constant
|
-1.66***
|
0.36
|
-0.92**
|
0.36
|
-2.09***
|
0.75
|
-0.62**
|
0.26
|
Welfare Metrics
|
LRCONSUMPTION PC
|
.25***
|
0.06
|
.16**
|
0.06
|
0.36***
|
0.12
|
0.11**
|
0.04
|
LLIVESTOCK
|
.51***
|
0.07
|
.48***
|
0.07
|
0.60***
|
0.19
|
0.36***
|
0.07
|
Institutions
|
TRUST
|
.11***
|
0.02
|
.05**
|
0.02
|
0.11***
|
0.04
|
0.03**
|
0.02
|
POLITICAL TRUST
|
.07***
|
0.02
|
0.04
|
0.02
|
0.08**
|
0.04
|
0.02
|
0.02
|
PARTICIPATION
|
.21**
|
0.09
|
0.01
|
0.09
|
0.27**
|
0.13
|
0.02
|
0.06
|
Religion and Religiosity
|
RELIGIOSITY
|
.014***
|
0.005
|
0.01
|
0.01
|
0.015*
|
0.008
|
0.005
|
0.003
|
CATHOLIC
|
-0.16
|
0.18
|
0.16
|
0.19
|
-0.06
|
0.21
|
0.11
|
0.10
|
MUSLIM
|
-0.21
|
0.15
|
-0.10
|
0.15
|
-0.25
|
0.20
|
-0.09
|
0.11
|
PROTESTANT
|
0.09
|
0.16
|
0.22
|
0.15
|
0.17
|
0.18
|
0.15
|
0.10
|
Others
|
VILLAGE CONTROLS
|
Yes
|
|
Yes
|
|
Yes
|
|
Yes
|
|
CONTROLS
|
Yes
|
|
Yes
|
|
Yes
|
|
Yes
|
|
Mu (1)
|
1.48***
|
0.06
|
1.81***
|
0.06
|
1.82**
|
0.52
|
1.16***
|
0.20
|
OBSERVATIONS
|
1114
|
|
1114
|
|
1114
|
|
1114
|
|
Variance Function
|
LRCONSP
|
|
|
|
|
0.11**
|
0.06
|
|
|
LLIVESTOCK
|
|
|
|
|
|
|
-0.19***
|
0.06
|
VILLAGE CONTROLS
|
|
|
|
|
Yes
|
|
Yes
|
|
Note |
Control variables not reported include land holding size, education, illness, gender, age, and number of younger and older children.
MU (1) refers to the first cut-off/threshold;
***, **, * denotes statistical significance at the 0.01, 0.05 and 0.10 level, respectively using two-tailed tests.
Regression results of the standard ordered probit models with religion-religiosity interactions are reported in Table 7. As shown in the table, the null hypothesis of homoscedasticity is rejected for both the life satisfaction happiness models at conventional significance level. Hence, we incorporate a variance function based on the logarithm of real consumption per capita, and the district controls for the satisfaction model. Similarly, a variance function based on the logarithm of livestock holding and the district controls is adopted to address the heteroscedasticity for the happiness model. Hence, our main analysis relies on the heteroscedastic model reported in Table 8.
The variance functions based on the log of consumption per capita and the log of livestock holdings are significant at conventional levels for the satisfaction and happiness models respectively. Likewise, district dummies are also significant in both models. As expected households at the high-end of consumption per capita report higher variability of responses in satisfaction to changes in the covariates as indicated by a significant positive coefficient in the variance function. In the happiness model, the log of livestock holding has been found to be the significant source of variation. The significant negative coefficient of log livestock in the variance function means that livestock-rich households exhibit a less varied response to happiness than their poorer counterparts. This might be taken to reflect the importance of livestock as an insulator against shocks for livestock-rich households (see, for example, Rosenzweig and Wolpin (1993) for the role of bullocks for consumption smoothing in India, and Gilligan and Hoddinott (2007) in Ethiopia). However, it is puzzling that this variance effect is not detected in the satisfaction equation itself. The more standard variables such as 'illness' and 'age' that are expected to be sources of variances in health-related wellbeing as suggested by Greene et al. (2014), do not yield significant variance effects in our models.
Table 7
Determinants of Subjective Wellbeing: Standard Ordered Probit
DEP.VAR.
|
SATISFACTION
|
HAPPINESS
|
|
Coef.
|
St.Er.
|
Coef.
|
St.Er.
|
CONSTANT
|
-1.5***
|
0.36
|
-0.86***
|
0.37
|
|
Welfare Metrics
|
|
|
LRCONSUMPTION PC
|
.24***
|
0.06
|
.17**
|
0.06
|
LLIVESTOCK
|
.51***
|
0.07
|
.48***
|
0.07
|
|
Institutions
|
|
|
|
TRUST
|
.11***
|
0.02
|
.05**
|
0.02
|
POLITICAL TRUST
|
.07***
|
0.02
|
0.04
|
0.02
|
PARTICIPATION
|
.21**
|
0.09
|
0.01
|
0.09
|
|
Religion and Religiosity
|
|
|
RELIGIOSITY
|
-0.001
|
0.01
|
0.003
|
0.01
|
CATHOLIC
|
-0.12
|
0.24
|
0.3
|
0.24
|
MUSLIM
|
− .33**
|
0.16
|
-0.13
|
0.17
|
PROTESTANT
|
-0.13
|
0.19
|
0.02
|
0.2
|
CATHOLIC*RELIGIOSITY
|
-0.02
|
0.03
|
-0.02
|
0.03
|
MUSLIM*RELIGIOSITY
|
.02*
|
0.01
|
0.002
|
0.01
|
PROTESTANT*RELIGIOSITY
|
.04**
|
0.02
|
.03*
|
0.02
|
|
Others
|
|
|
|
VILLAGE CONTROLS
|
Yes
|
-
|
Yes
|
-
|
CONTROLS
|
Yes
|
-
|
Yes
|
-
|
Mu (1)
|
1.48*
|
0.06
|
1.81*
|
0.06
|
Diagnostics [P-Values in Parenthesis]
|
OBSERVATIONS
|
1114
|
|
1114
|
|
LOG-LIKELIHOOD VALUE
|
-1017.59
|
|
-930.5
|
|
HETEROSCEDASTICITY
|
82.5*** (0.00)
|
|
75.3*** (0.00)
|
|
Note |
Control variables not reported include land holding size, education, illness, gender, age, and number of younger and older children.
MU (1) refers to the first cut-off/threshold;
***, **, * denotes statistical significance at the 0.01, 0.05 and 0.10 level, respectively using two-tailed tests.
Table 8
Determinants of Subjective Wellbeing: Heteroscedastic Ordered Probit
DEP.VAR.
|
SATISFACTION
|
HAPPINESS
|
|
Coef.
|
St.Er.
|
Coef.
|
St.Er.
|
CONSTANT
|
-1.83***
|
0.68
|
-0.55**
|
0.26
|
|
Welfare Metrics
|
|
|
LRCONSUMPTION PC
|
0.34***
|
0.11
|
0.11**
|
0.04
|
LLIVESTOCK
|
0.57***
|
0.18
|
0.36***
|
0.07
|
|
Institutions
|
|
|
|
TRUST
|
0.11***
|
0.04
|
0.03*
|
0.02
|
POLITICAL TRUST
|
0.08**
|
0.04
|
0.02
|
0.02
|
PARTICIPATION
|
0.26**
|
0.13
|
0.02
|
0.06
|
|
Religion and Religiosity
|
|
|
RELIGIOSITY
|
-0.004
|
0.01
|
0.001
|
0.01
|
CATHOLIC
|
-0.09
|
0.25
|
0.22
|
0.14
|
MUSLIM
|
-0.40*
|
0.22
|
-0.13
|
0.12
|
PROTESTANT
|
-0.07
|
0.2
|
0.05
|
0.12
|
CATHOLIC*RELIGIOSITY
|
0.01
|
0.03
|
-0.02
|
0.02
|
MUSLIM*RELIGIOSITY
|
0.03*
|
0.02
|
0.004
|
0.01
|
PROTESTANT*RELIGIOSITY
|
0.04**
|
0.02
|
0.02
|
0.01
|
|
Others
|
|
|
|
VILLAGE CONTROLS
|
Yes
|
-
|
Yes
|
-
|
CONTROLS
|
Yes
|
-
|
Yes
|
-
|
Mu (1)
|
1.75***
|
0.5
|
1.17***
|
0.2
|
Variance Function
|
LRCONSUMPTION PC
|
0.10*
|
0.06
|
|
|
LLIVESTOCK
|
-
|
-
|
-0.18*
|
0.06
|
VILLAGE CONTROLS
|
Yes
|
-
|
Yes
|
-
|
Diagnostics [P-Values in Parenthesis}
|
OBSERVATIONS
|
1114
|
|
1114
|
|
LOG-LIKELIHOOD VALUE
|
-998.3
|
|
-913.7
|
|
HETEROSCEDASTICITY
|
-
|
|
-
|
|
Note |
Control variables not reported include land holding size, education, illness, gender, age, and number of younger and older children.
MU (1) refers to the first cut-off/threshold;
***, **, * denotes statistical significance at the 0.01, 0.05 and 0.10 level, respectively using two-tailed tests.
Consumption per capita and livestock holding emerge as strong predictors of both life satisfaction and happiness. An analysis of the marginal effects from the Heteroscedastic model reported in Table 9 reveals each additional consumption (in logarithmic form) makes an average individual nine percentage points less likely to report ‘dissatisfied' and eight percentage points more likely to report `satisfied. This implies that a 5% increase in consumption decreases the probability of reporting `dissatisfied' by 0.45 of a percentage point, and increase the probability of reporting `satisfied' by 0.4 of a percentage point. The positive role of these welfare metrics is evident in all specifications.
In addition to own income, relative income (relative consumption) compared to neighbours or peers can potentially affect SWB. However, due to lack appropriate information on neighbours or smaller geographic unit, we have not included a measure of relative income in our main models. Most studies in developing countries have reported little or no effect of relative income on SWB. For example, Ravallion and Lokshin (2010) find little support for positional concern (measured by relative deprivation) for most poor households in Malawi with the exception of the relatively well-off. Similarly, Using surveys of two separate villages in Northern Ethiopia Akay and Martinsson (2011) and Akay et al. (2012) studied the role of ‘positional concern' using experimental methods and find no evidence for the existence of positional concern as defined by the income of others in the community. One potential measure of relative income could be per capita consumption quartiles by village and include an indicator of which quartiles an individual belongs to. Regression results of our satisfaction and happiness models with this measure are provided in Table A2 in the Appendix A.
The results reveal that religiosity has a differential impact on SWB based on religious denomination. Muslims report significantly lower satisfaction levels than Orthodox Christians. Catholics and Protestants also exhibit lower satisfaction levels albeit the difference is not statistically different at a conventional level of significance. On the other hand, religious Muslims and Protestants report significantly higher satisfaction levels than Orthodox Christians revealing a differentiated role for religiosity on wellbeing. Muslims and Protestants report significantly higher religiosity levels than their Orthodox counterparts in the survey. The marginal effects reported in Table 9 reveal that being a religious Muslim makes an individual 0.8 percentage points less likely to report dissatisfied and 0.7 percentage points more likely to report satisfied. Similarly, being a religious protestant makes an individual one percentage points less likely to report being dissatisfied and one percentage points more likely to report being satisfied. The positive role of religiosity in newly introduced religions in Ethiopia such as Protestantism can be indicative that religion can create a platform for the development of social capital for minority groups. Azzi and Ehrenberg (1975) find a similar result for racial minorities for the US.
Unlike with the general life satisfaction model, religion and religiosity do not feature as significant determinants in the happiness model. This is indicative of a subtle characteristic, often overlooked in wellbeing studies, which is that respondents can differentiate between general satisfaction and momentary happiness. This is especially so in religious communities that consider future gains from religion (such as going to heaven) in their general satisfaction function but not so in happiness function as it tends to be temporary in nature. For example, a religious person who fasts and avoids feasts may report low responses for momentary happiness, but higher in the overall response to life satisfaction. Moreover, in the event of adverse circumstances, while temporary happiness can drop for religious and non-religious people alike, the overall satisfaction of religious people may not drop significantly as they are likely to attribute the bad events to the will of God (see, for example, Pollner (1989), Ellison (1991), Frey and Stutzer (2002), Inglehart and Norris (2004) regarding the soothing role of religion in times adversary.
As in many studies in developing and developed economies general trust among the people and trust in local political officials emerge as strong correlates of satisfaction and happiness in our data. In developing countries, people are dependent on each other and their community in everyday life. Labour sharing in farming and sharing of costs during important social events such as weddings and mourning are hallmarks of life in rural areas. Hence, social capital in the form of trust is an important element. Moreover, in those regions where formal political and administrative institutions are not fully developed, informal institutions based on trust play an important role in society by creating peace and stability in the community and the management of communal resources. A measure of general trust emerges as a robust determinant of SWB in our satisfaction model as reflected in its large magnitude of 0.11 of a standard deviation and its statistical significance at the 1% level. The marginal effects from the Heteroscedastic model reported in Table 9 reveal each additional level of trust makes an average individual three percentage points less likely to report `dissatisfied' and three percentage points more likely to report `satisfied'. Other studies that find as strong a direct effect of trust on SWB include Bjørnskov (2003) for more affluent countries, Helliwell (2003) for individual-level data across many countries controlling for national trust levels; Helliwell and Putnam (1995) using cross-country and national surveys for US and Canada; and Asadullah and Chaudhury (2012) in rural Bangladesh.
Confidence in political institutions (government trust) is a statistically robust predictor of SWB. The role of political trust on life satisfaction is not statistically different from that of the effect of general trust. However, while the general trust positively affects both life satisfaction and happiness, confidence in local political institutions affects life satisfaction, but not happiness. The role of local political institutions in rural households is paramount. In the context of rural Ethiopia, the role of local administrative units includes land and agricultural input allocation, arbitration during conflict, food aid distribution, and safety net participation. Hence, the confidence that households have in such institutions that play a significant role in their lives affects them directly through a sense of security and indirectly through effects on resource allocation. Studies that find a positive role of quality of governance and political freedom on SWB include Helliwell (2006) and Diener and Diener (2009), both using cross-country data. On the other hand, Veenhoven (2000) finds that a stronger role of political freedom on SWB in richer countries, while economic freedom has a stronger effect in poorer countries. Our finding implies that effective and transparent political institutions at a local level are highly beneficial for household self-reported satisfaction.
It is interesting to note that general trust significantly predicts life satisfaction and momentary happiness while political trust affects life satisfaction but not momentary happiness. This is a sensible result as an individual's trust in political institutions affects their overall welfare but has little significance in momentary emotions. On the other hand general trust that is crucial in the day-to-day interactions with people surrounding the individual is important for both an overall wellbeing and a momentary happiness.
Participation in local institutions in the form of having an official position in community organisations, local administrative committees, or religious institutions, is positively associated with life satisfaction while not so with happiness. This implies that such positions can be fulfilling regarding the overall objective and purpose of life and the sense of contribution to society. However, shouldering the responsibility and ensuring the smooth operation of institutions in poor areas may involve stress and also use up time which otherwise would have been spent on family or other domestic activities. This implies that participatory political and socio-economic institutions can boost wellbeing. Using survey data from Switzerland, Frey and Stutzer (2000) find people in regions with more developed institutions of direct democracy report significantly higher levels of self-reported wellbeing.
Similarly, the distinctive effective of religiosity on life satisfaction and happiness is another interesting finding. In line with Azzi and Ehrenberg (1975) individuals take the after-life utility into account when evaluating their overall wellbeing as reflected in religiosity being a predictor of life satisfaction but not of momentary happiness.
In summary, the differential impact of institutions on life satisfaction and momentary happiness is in comport with Deaton’s (2008) and Stevenson and Wolfers’s (2008) proposition that life satisfaction and happiness are not synonymous.
Table 9
Marginal Effects of Selected Variables
|
LIFE SATISFACTION
|
HAPPINESS
|
|
0
|
1
|
2
|
0
|
1
|
2
|
LRCONSUMPTION PC
|
-0.09
|
0.01
|
0.08
|
-0.06
|
0.04
|
0.02
|
LLIVESTOCK
|
-0.16
|
0.03
|
0.13
|
-0.21
|
0.12
|
0.08
|
RELIGIOSITY
|
0.001
|
-0.0002
|
-0.0009
|
-0.0005
|
0.003
|
0.0004
|
CATHOLIC*RELIGIOSITY
|
-0.003
|
0.0004
|
0.002
|
0.01
|
-0.005
|
-0.004
|
MUSLIM*RELIGIOSITY
|
-0.008
|
0.001
|
0.007
|
-0.003
|
0.002
|
0.001
|
PROTESTANT*RELIGIOSITY
|
-0.01
|
0.002
|
0.01
|
-0.01
|
0.007
|
0.004
|
TRUST
|
-0.03
|
0.005
|
0.03
|
-0.02
|
0.01
|
0.01
|
POLITICAL TRUST
|
-0.03
|
0.003
|
0.02
|
-0.01
|
0.01
|
0.004
|
PARTICIPATION
|
-0.07
|
0.01
|
0.06
|
-0.012
|
0.007
|
0.005
|
Note |
The Marginal Effects are based the Heteroscedastic Ordered Probit estimates reported in Table 8.
To get a sense of the relative importance of the determinants of SWB, we can construct ‘indifference curves’ between any two continuous covariates whose slopes represent the ‘marginal rate of substitution’ between them. The indifference curves represent various combinations of two covariates that yield the same level of satisfaction. In the current application, the slopes of the indifference curves are given by the minus of the ratio of their β-coefficients (see, for example, Stewart et al. (2004) and Litchfield et al. (2012)). Table 10 reports indifference curves for an average individual for selected covariates based on the estimates of Table 10.
Focusing on the heteroscedastic models, the slopes of the indifference curves reveal that individuals are willing to sacrifice a 5% of consumption for an additional visit to a church/mosque to stay at the same level of satisfaction. This implies that the value of a visit to a church/mosque is equivalent to 5% of their per capita consumption. Similarly, one extra visit to a church/mosque per month offsets the loss in satisfaction due to a reduction of 2.6% of livestock holdings. A one point increase, which is large relative to the mean, in general trust or government trust, can compensate for 32% and 23% reduction in consumption per capita respectively. With a mean of 4.37, a one-point increase in general trust corresponds to 23% in percentage terms. Similarly, with an average of 4.19, a one-point increase in government trust corresponds to 24% in percentage terms. Therefore, a 23% increase in general trust compensates a reduction in consumption of 32%. Moreover, a 24% increase in government trust compensates for a decrease in consumption of 23%. The sizes of the effects of trust (general and government) are surprisingly large. It can be due to the absence of formal institutions in rural areas. Since we have not adequately controlled for potential endogeneity of these factors, the results should be taken only as indicative.
Table 10
Trade-offs between Selected Covariates: Slopes of Indifference Curves
Slope for a given Satisfaction level
|
Standard Order Probit
|
Heteroscedastic Order Probit
|
Change in per capita consumption required to compensate for extra day in church/mosque
|
-0.056**
(-0.03)
|
-0.04**
(-0.02)
|
Change in livestock holdings required to compensate for extra day in church/mosque
|
-0.027**
(-0.01)
|
-0.025**
(-0.01)
|
Change in per capita consumption required to compensate for an extra trust level
|
-0.43***
(-0.13)
|
-0.32***
(-0.1)
|
Change in per capita consumption required to compensate for an extra government trust level
|
-0.29***
(-0.11)
|
-0.23***
(-0.09)
|
Note |
Based on Table 6 for a more straightforward treatment of religiosity as we don’t include interactions
***, **, * denotes statistical significance at the 0.01, 0.05 and 0.10 level, respectively using two-tailed tests.