Decomposing the increase in men’s time on childcare during the great recession

Abstract

In this paper, I examine the time men spent on childcare during the recession of 2007–2009. The recession provides a sudden change in the employment opportunities of men relative to women in the United States. Using the American Time Use Survey and the linked Current Population Survey, I show that this lopsided shock to employment opportunities was accompanied by an increase in the average amount of time men spent on childcare. In particular, men’s average time on physical care for children increased during the recession; this is an element of childcare that men perform less than women. I decompose the total change in average time on childcare into behavioral, compositional, and between group change. A behavioral change among employed men accounted for the majority of the total increase in the average time spent on childcare; among men who are out of the labor force, the increase is entirely due to compositional changes.

Appendices

Appendix 1: Regression results

Figure 3 showed that men who lived in a household with children spent 6.9 more minutes on average on childcare during the recession. To see whether men’s raw increase persists after accounting for other characteristics, I employ the following regression:

$$y_{i} = \varvec{X}_{\varvec{i}}\varvec{\beta}+ year_{i} \xi + I\left( {DURING_{i} } \right)\delta + I\left( {AFTER_{i} } \right)\gamma + \varepsilon_{i}$$

In this framework, the amount of time a person spends on childcare (y _i ) is a function of their demographic characteristics (X), a linear year trend, and an indicator variable for if the time diary occurred during the recession. The coefficient on the indicator variable for being in the recession, $\hat{\delta }$, will show if there was a change in the amount of childcare provided by men as a whole during the recession. The demographic variables include education, race, ethnicity, marital status, number of children in the household, age of the youngest child, age, and age squared. All regressions use the complex sampling weights provided by the ATUS-X. All regressions use robust standard errors and the observations are clustered by state.

Table 8 demonstrates that the average time men in households with children spent on childcare increased during the recession. This finding is consistent across specifications; the increase remains regardless of whether childcare is measured in total or restricted to only when a man’s spouse or partner is not present. Likewise, the increase is consistent whether measured for all men in households with children or just those who spent a positive amount of time on childcare.

Table 8 Selected coefficients from regression of men’s time on childcare on an indicator variable for during the recession, and indicator variable for after the recession, and control variables

In contrast, Table 9 shows that the average time women in households with children spent on childcare did not change during the recession.

Table 9 Selected coefficients from regression of women’s time on childcare on an indicator variable for during the recession, and indicator variable for after the recession, and control variables

To check whether the result shown in Table 8 is solely due to a change in the composition of men living in households with children, I ran the same regressions on all men. The results are consistent with those found above (Table 10).

Table 10 Selected coefficients from regression of men’s time on childcare on an indicator variable for during the recession, and indicator variable for after the recession, and control variables

As shown in Table 11, the proportion of men engaging in any childcare did not change during the recession.

Table 11 Selected coefficients from regression of men’s participation in childcare on an indicator variable for during the recession, and indicator variable for after the recession, and control variables

Table 12 shows that the average time spent on physical care by men with children in their household increased during the recession.

Table 12 Selected coefficients from regression of men’s time on childcare on an indicator variable for during the recession, and indicator variable for after the recession, and control variables

Appendix 2: Decomposition method

To decompose the total change in average time spent on childcare, I start with the basic difference between the average minutes of childcare during the recession ($M^{D}$) and before the recession ($M^{B}$).

$$\Delta = M^{D} - M^{B}$$

Each average can be re-written as the weighted average of the childcare performed by men in each employment status. $P_{i}^{t}$ is the proportion of men who are in employment status i in period t; $M_{i}^{t}$ refers to the average minutes of childcare performed by men in employment status i in period t.

$$\Delta = P_{E}^{D} M_{E}^{D} + P_{U}^{D} M_{U}^{D} + P_{N}^{D} M_{N}^{D} - [P_{E}^{B} M_{E}^{B} + P_{U}^{B} M_{U}^{B} + P_{N}^{B} M_{N}^{B} ]$$

$$\Delta = [P_{E}^{D} M_{E}^{D} - P_{E}^{B} M_{E}^{B} ] + [P_{U}^{D} M_{U}^{D} - P_{U}^{B} M_{U}^{B} ] + [P_{N}^{D} M_{N}^{D} - P_{N}^{B} M_{N}^{B} ]$$

Now, add and subtract $P_{i}^{D} M_{i}^{B}$ for each employment status.

$$\Delta = P_{E}^{D} \left[ {M_{E}^{D} - M_{E}^{B} } \right] + M_{E}^{B} \left[ {P_{E}^{D} - P_{E}^{B} } \right] + P_{U}^{D} [M_{U}^{D} - M_{U}^{B} ] + M_{U}^{B} \left[ {P_{U}^{D} - P_{U}^{B} } \right] + P_{N}^{D} [M_{N}^{D} - M_{N}^{B} ] + M_{N}^{B} [P_{N}^{D} - P_{N}^{B} ]$$

The minutes spent on childcare during the recession by each employment status $(M_{i}^{D} )$ can be separated out by a lagged employment measure. That is, $M_{i}^{D} = P_{E,i}^{D} M_{E,i}^{D} + P_{U,i}^{D} M_{U,i}^{D} + P_{N,i}^{D} M_{N,i}^{D}$. Where $P_{j,i}^{D}$ is the proportion of men during the recession who were in employment status j in the lagged CPS survey conditional on being in employment status i in the current ATUS survey (that is, $P_{N,i}^{D}$ + $P_{E,i}^{D}$ + $P_{U,i}^{D}$ = 1), and $M_{j,i}^{D}$ is the average time spent on childcare by men during the recession who were in employment status j in the lagged CPS survey and are in employment status i in the current ATUS survey.

So, $M_{i}^{D} - M_{i}^{B} = P_{E,i}^{D} \left( {M_{E,i}^{D} - M_{i}^{B} } \right) + P_{U,i}^{D} (M_{U,i}^{D} - M_{i}^{B} ) + P_{N,i}^{D} (M_{N,i}^{D} - M_{i}^{B} )$

Substituting this in for each $M_{i}^{D}$ and $M_{i}^{B}$ gives

$$\Delta = \mathop \sum \limits_{i = N,E,U} \left[ {P_{i}^{D} \left[ {P_{N,i}^{D} \left( {M_{N,i}^{D} - M_{i}^{B} } \right) + P_{E,i}^{D} \left( {M_{E,i}^{D} - M_{i}^{B} } \right) + P_{U,i}^{D} \left( {M_{U,i}^{D} - M_{i}^{B} } \right)} \right] + (M_{i}^{B} )\left( {P_{i}^{D} - P_{i}^{B} } \right)} \right]$$

Finally, because $\left( {P_{E}^{D} - P_{E}^{B} } \right) + \left( {P_{U}^{D} - P_{U}^{B} } \right) + \left( {P_{N}^{D} - P_{N}^{B} } \right) = 0$, add $- M^{B} \left[ {\left( {P_{E}^{D} - P_{E}^{B} } \right) + \left( {P_{U}^{D} - P_{U}^{B} } \right) + \left( {P_{N}^{D} - P_{N}^{B} } \right)} \right]$. This gives Eq. 1:

$$\Delta = \mathop \sum \limits_{i = N,E,U} \left[ {P_{i}^{D} \left[ {P_{N,i}^{D} \left( {M_{N,i}^{D} - M_{i}^{B} } \right) + P_{E,i}^{D} \left( {M_{E,i}^{D} - M_{i}^{B} } \right) + P_{U,i}^{D} \left( {M_{U,i}^{D} - M_{i}^{B} } \right)} \right] + (M_{i}^{B} - M^{B} )\left( {P_{i}^{D} - P_{i}^{B} } \right)} \right]$$

Appendix 3: Analysis of panel data

The ATUS contains rich data on time use, but it does not track the same person over time. The Panel Study of Income Dynamics (PSID) contains only very coarse measures of time use, but has multiple observations on the same respondent over time. I use the PSID to show that the time male heads of household report spending on housework increased during the 1990–1991 recession.² This echoes Casper and O’Connell (1998), who analyzed cross-sectional data and found an increase fathers who were care providers for preschool children during the 1990–1991 recession. The consistency between cross-sectional and panel data during the 1990–1991 recession suggests that the analyses of the ATUS for the 2007–2009 recession in the main text are credible.

The PSID began in 1968 with a nationally representative sample of 5,000 families; these individuals and their descendants have been followed over time (Panel Study of Income Dynamics 2014). The PSID asks respondents how much time they spend on housework in the average week. Stylized questions about time use tends to be biased upwards (Hofferth 2000) and inconsistent between a self-report and a proxy-report: husbands report spending more time on housework than their wives say they do (Achen and Stafford 2005). A respondent fixed effect accounts for time-invariant reporting bias.

To examine if male household heads increased their time on housework during the 1990–1991 recession, I employ the following regression: $y_{i,t} = \alpha + year_{i,t} \xi + I\left( {DURING_{i,t} } \right)\delta + \mathop \sum \limits_{j = 1}^{n - 1} I(i = j)\beta_{j} + \varepsilon_{i,t}$. In this framework, y _i,t is the amount of time a person i spends on housework per week in year t. The time the respondent spent on housework is a function of a linear year trend, a respondent fixed effect, and an indicator variable for if the survey occurred during the recession (measured as 1990 or 1991, to conform to Casper and O’Connell). The coefficient on the indicator variable for being in the recession, $\hat{\delta }$, will show if there was a change in the amount of housework reported by men during the recession relative to their own average. In the following table a “male household head” is a man who was the head of household for all ten survey years (1986–1995).

As shown in Table 13, male household heads reported spending more time on housework during the 1990–1991 recession relative to their own average. This pattern is consistent when restricted to men who were employed during all ten surveys. Thus, both cross-sectional data (Casper and O’Connell 1998) and panel data show an increase in male unpaid labor during the 1990–1991 recession. This is supportive, although not conclusive, evidence that the analysis in the main text is credible.

Table 13 Selected coefficients from regression of men’s self-reported time on housework per week on an indicator variable for 1990 or 1991 sample years, a time trend, and respondent fixed effects