Gender and the time-squeeze: childcare versus work-hours in South Africa during lockdown
Image credit: Sandy Millar on Unsplash.
How did the sudden shock in the demand for at-home childcare induced by the Covid-19 lockdowns affect labour market hours for men and women? This is the question Jacqueline Mosomi and I set out to answer in an article recently accepted for publication at Feminist Economics. We used NIDS-CRAM, a survey on employment and other outcomes implemented by SALDRU and other partners during the pandemic, to run both regression and sequence analysis. This empirical strategy took advantage of some of the special attributes of this dataset, namely its collection of hours of childcare data and its panel design. In this article I discuss our regression results, including a missing data problem that we faced that could have halted the project completely. Invaluable advice from Martin Wittenberg about how to solve this problem rescued the project. Overall, we found robust evidence that the increased demand for at-home childcare significantly reduced women’s work hours and did not have the same effect on men’s work hours.
Feminist scholars have long argued that the privatisation of social reproductive activities largely shouldered by women subsidises the formal economy. Effective caregiving produces a healthy workforce both via the nurturance of children (i.e. their healthcare and education) and through care for the sick, differently-abled, and elderly. This is particularly true in South Africa, where the state mostly outsources care work to families. Compartmentalising care work from paid labour market work is thus an unjustified separation, but one that is highly lucrative for those in a position to profit. Few shocks have exposed this false segmentation more than Covid-19, and a wealth of literature has been published in the wake of the pandemic using data from around the world to empirically support this feminist argument. Much of this literature has focused on the gendered imbalance of paid and unpaid work and utilised the concept of the ‘time-squeeze’.
The term ‘time-squeeze’ is used to describe the impossibility of both working a full day doing your paid job as well as being hands-on and available to care for your child at the same time. The Covid-19 lockdowns imposed by the South African government on the 26th of March 2020 placed many working parents in precisely this position by closing schools and ECD centres and sending childminders home. To investigate how the time-squeeze affected paid work hours in South Africa, we used data from NIDS-CRAM and specifically limited our sample to men and women working full-time before the pandemic (in February 2020). There are five waves in NIDS-CRAM, each associated with a lockdown level that varied in the intensity of the demand for at-home childcare. A pre-pandemic baseline was constructed using retrospective questions asked about February 2020 in wave 1.
We then set up the following Tobit regression for daily hours of work (the ‘y’ variable) which we run separately for men and women:
T is a set of six time period dummies (the five survey waves plus the constructed February baseline) which we interpret as lockdown level. C is one of two measures for childcare, being either a dummy for co-residence with a child under the age of 18 or a point estimate for hours of childcare. X is a vector of other controls. By interpreting ß3 – the coefficient on the interaction of childcare and lockdown level – we can learn about how the intensity of a given lockdown level correlated with the change in work hours for those responsible for childcare compared to those who were not, by gender.
A prerequisite for running this regression was having a February or pre-pandemic point estimate of hours of childcare. NIDS-CRAM did not collect a point estimate for February and although we could use other questions to back out an initial estimate, we faced a severe data quality problem related to bracket design and outliers for the top third of the February childcare hours distribution. Because this problem affected the full top third of the distribution, it was no ordinary missing data problem. We brought this problem to a working group session held by Martin Wittenberg in late 2023, and he suggested childcare hours were probably Pareto-distributed. He suggested we model the upper segment of the distribution using auxiliary data sources on childcare hours to parameterise draws on a Pareto distribution.
We combined information from the 2010/11 Time Use Survey and the NIDS-CRAM March 2021 wave to parameterise a Pareto distribution from which to take draws for February 2020. We use NIDS-CRAM March 2021 even though it is a post-shock wave to account for instrument differences between NIDS-CRAM and the Time Use Survey and known problems of underestimation of childcare in the latter. (As far as we are aware, NIDS-CRAM is the first example for South Africa of a one-shot retrospective question asked about daily hours of childcare.) Figure 1 describes the CDF of hours of childcare for each wave and how our Pareto draws fit in. And Figure 2 shows the evolution of hours of childcare over the course of the lockdown for our sample using our imputation for February.
Figure 1: CDF of daily hours of childcare in NIDS-CRAM using 100 draws from a Pareto distribution for Feb L0
Source: Own calculation using NIDS-CRAM. Sample restricted to the regression sample with non-zero hours of childcare. For the draws from the Pareto distribution, the scale factor is 6 hours and Pareto parameter is 2.4 for women and 2.8 for men. All waves plus baseline Feb L0 are in year 2020 except Jan Adj. L3 and March Adj. L1 which are in year 2021.
Figure 2: Hours of daily childcare by sex over time for people co-residing with children
Source: Own calculation from NIDS-CRAM. Sample restricted to those working full-time in Feb L0. Adjusted using cross-sectional sampling weights with a 95% confidence interval. Red vertical line = start of lockdown.
Using our imputed February baseline, we ran our regression and found that an additional hour of childcare in April 2020 reduced women’s hours of work by about 2 hours and men’s by a statistically insignificant 20 minutes. It is encouraging for the quality of our imputation that running the regression using co-residence with a child produces highly comparable results. Living with a child was associated with reducing women’s hours of work by 1-2 hours compared to women not living with children, but there was no significant effect for men.
A revealing result was that sometimes the interaction effect was larger when lockdown was weaker. When the first lockdowns were imposed, they were so comprehensive that the lockdown itself was the main reason people’s work hours reduced. But later on, when schools and offices started opening up in a patchy way, this was really when households needed to decide who sacrificed work hours to put in childcare. We found large and significant coefficients for women in June 2020, October 2020 and January 2021 when schools were often using rotating class systems where children only attended school on alternate days. Overall, interaction effects for men were often a fraction of the size of those for women and only one coefficient was significant for men out of ten possible cases, compared to eight for women.
Our work contributes to a body of evidence that the pandemic worked to further entrench existing gender inequalities in South Africa and illustrates how the home and workplace cannot be disentangled.
Note: the manuscript has been accepted by Feminist Economics but still needs to undergo editing and stylistic updates. Anyone wanting to see the full manuscript can contact Amy for sharing at amy.thornton@uct.ac.za.