Chapter 2: Adult Subjective Well-being and Childhood Bullying Victimisation
2.5 Results
2.5.1 Random Effects Ordered Probit Model
Table 2.6(a) presents the coefficients from estimating equations (2.6) and (2.7) using a random effects ordered probit model. Life satisfaction is the dependent variable. Columns (1) and (3) illustrate the results from estimating equation (2.6) using Sample 1 (the life satisfaction balanced panel) and Sample 2 (the life satisfaction unbalanced panel), respectively. Columns (2) and (4) show the findings from estimating equation (2.7) using Sample 1 and Sample 2, respectively.
All four specifications demonstrate negative, statistically significant effects of bullying victimisation at age 11 on adult life satisfaction. The estimates using Sample 1 and Sample 2 are of a similar magnitude, supporting the robustness of the findings. The coefficient of the measure of bullying victimisation at age 11 is approximately 30% of the magnitude of the unemployment coefficient. Unemployment is a heavily researched determinant of individual life satisfaction, see Kassenboehmer and Haisken-DeNew (2009), for example. Additionally, the magnitude of the effects of being bullied at age 11 on life satisfaction is approximately a quarter of the absolute magnitude of the effect of being married or cohabiting (relative to being single). After conditioning on the childhood characteristics, the coefficient of the bullying variable diminishes in magnitude; this is the case for both Sample 1 and Sample 2.
38 For comparison, Table 2.6(b) replicates the analysis of Table 2.6(a) but excludes the Mundlak averages. After removing the Mundlak averages, the coefficient of the measure of bullying victimisation becomes approximately 20% larger. This suggests that unobserved characteristics that positively affect life satisfaction are negatively associated with the probability of being bullied at age 11.
There are two explanations for this change. Firstly, cohort members with certain early childhood characteristics may be more likely to be bullied at age 11 because of these characteristics. These characteristics may also affect the Mundlak averages. Secondly, bullying victimisation at age 11 may affect the Mundlak averages. Whatever the explanation, the large reduction in the effect of bullying victimisation when accounting for the Mundlak averages is of note, indicating the importance of conditioning upon the individual effects.
Previous literature indicates that children who are bullied have lower educational attainment, lower income, and are more likely to be unemployed as adults, (see, for example, Eriksen et al., 2014; Brown and Taylor, 2008; Varhama and Bjorkqvist, 2005) . Thus, one pathway via which bullying victimisation in childhood may affect SWB as an adult may be via the indirect effect of being bullied on these outcomes. To explore this possibility, Table 2.6(c) presents simplified specifications excluding the household income, labour force status, homeownership status, and highest educational attainment variables. The findings indicate that the effect of being bullied at age 11 on life satisfaction is robust to the exclusion of these control variables. What is more, note that the coefficients become larger in magnitude relative to the estimates presented in Table 2.6(b). This supports the notion that one mechanism via which being bullied at age 11 may affect life satisfaction as an adult is via the indirect effect on adult labour market outcomes.
Table 2.7(a) is presented in an identical fashion to Table 2.6(a) though the dependent variable is total malaise, rather than life satisfaction. Note that the total malaise index is reversed and hence higher scores represent higher, rather than lower, SWB. Columns (1) to (2) and (3) to (4) illustrate the findings for Sample 3 (the total malaise balanced panel) and Sample 4 (the total malaise unbalanced panel), respectively. In common with the life satisfaction results, in each of the four specifications the measure of whether the child was bullied at age 11 has a negative coefficient. In addition, note that if birth-weight and mental health problems are included as control variables, the coefficient of the bullying variable reduces in magnitude.
Table 2.7(a) contains some findings that do not accord with the existing literature. For example, there is no statistically significant effect of unemployment on total malaise. This is surprising; a substantial literature has demonstrated the detrimental effect of unemployment on SWB, see Clark and Oswald (1994). The table also suggests that individuals who are not in the labour force report lower total malaise (lower SWB) relative to employed individuals. This effect is statistically significant at the 1% level.
39 One explanation for this somewhat surprising result is that there is not a clear distinction between being unemployed and being out of the labour force, as argued by Blackaby et al. (2007) and Brown et al. (2010). However, this result is inconsistent with the findings for life satisfaction that demonstrate a detrimental effect of unemployment, and a smaller, negative effect of being out of the labour force.
A second explanation for this difference is a selection effect: individuals in the life satisfaction samples may have a different experience of unemployment, relative to individuals in the total malaise samples. To investigate this explanation, the regressions presented in Tables 2.6(a) and 2.7(a) were re-estimated using a common estimation sample.42 However, this re-estimation leads to little substantive change in the coefficients of the labour force status variables in the models utilising total malaise and life satisfaction as dependent variables. This indicates that the difference in the effects of labour market status on life satisfaction and total malaise is due to the different measures of SWB, rather than a selection effect, thus highlighting the importance of exploring a range of measures of SWB.
Table 2.7(b) shows the equivalent results to Table 2.7(a), without including the Mundlak averages. Relative to Table 2.7(a), the sizes of the coefficients of the measure of whether an individual was bullied at age 11 on total malaise are approximately 20% larger. In summary, the findings presented in this section indicate an adverse effect of bullying victimisation at age 11 on both measures of SWB.43
As previously argued, childhood bullying may affect SWB via its indirect effect on adult labour market outcomes. To explore this, we re-estimate the models presented in Tables 2.7(b) after excluding the household income, labour force status, homeownership status, and highest educational attainment variables. The findings, presented in Table 2.7(c), indicate that the effect of bullying victimisation remains statistically significant and negative after removing these. In addition, relative to the estimates presented in Table 2.7(b), the magnitude of the coefficients of the bullying variable is larger. This indicates that childhood bullying victimisation may affect total malaise via its indirect effect on these adult labour market outcomes.