Are Children Barriers to the Gender Revolution? International Comparisons

Abstract

Children seem to present a barrier to the gender revolution in that parents are more likely to divide paid and domestic work along traditional gender lines than childless couples are. However, the extent to which this is so varies between countries and over time. We used data on 35 countries from the 2012 International Social Survey Programme to identify the contexts in which parents and non-parents differ the most in their division of labour. In Central/South America, Eastern Europe, Southern Europe, Asia, and South Africa, labour sharing configurations did not vary as much with the presence of children as in Australia, Western Europe, North America, and Northern Europe. Our multilevel models helped explain this pattern by showing that children seem to present a greater barrier to the gender revolution in richer and, surprisingly, more gender equal countries. However, the relationship between children and couples’ division of labour can be thought of as curvilinear, first increasing as societies progress, but then weakening if societies respond with policies that promote men’s involvement at home. In particular, having a portion of parental leave reserved for fathers reduces the extent to which children are associated with traditional labour sharing in the domestic sphere.

Appendix: Implementation of Two-Stage Least Squares

We attempt to estimate the following model:

$${\text{modern}}_{i} = { \propto } + \beta \;{\text{haschildren}}_{i} + \gamma X_{i} + \mu_{i}$$

where ${\text{modern}}_{i}$ is an indicator of whether respondent i has a modern configuration of labour, ${\text{haschildren}}_{i}$ is an indicator of whether respondent i has children, and $X_{i}$ represents a vector of controls of respondent i. This equation has a causal interpretation and is known as the structural equation.

The problem, however, is that ${\text{haschildren}}_{i}$ is most likely an endogenous regressor. That is, it may be correlated with the error term $\mu_{i}$ due to the fact that couples’ labour configurations can affect fertility decisions. A solution to this is to find an exogenous regressor, which in our case is ${\text{wantchildren}}_{i}$, that is uncorrelated with $\mu_{i}$ but correlated with the endogenous variable ${\text{haschildren}}_{i}$. More specifically, for this instrument to be valid it must be correlated with the endogenous variable and is exogenous in the structural equation, i.e. it does not influence labour configurations except through fertility decisions given preferences.

If this variable satisfies those requirements, then it can be used to “cleanse” the endogeneity by using predicted values based on the exogenous variables only. In other words, we are decomposing the variation of ${\text{haschildren}}_{i}$ into an exogenous and endogenous part. This is known as the first stage regression, which can be represented by the following equation:

$${\text{haschildren}}_{i} = \pi_{01} + \pi_{11} \;{\text{haschildren}}_{i} + \varphi X_{i} + \omega_{i}$$

The next step would be to estimate the structural equation using the predicted values of the first stage regression ($\widehat{\text{haschildren}}$). This is known as the second stage regression, which can be represented by the following equation:

$${\text{modern}}_{i} = \alpha + \beta\;\widehat{\text{haschildren}}_{i} + \gamma X_{i} + \mu_{i}$$