The drivers of mothers parental leave secisions

Methodology

In this paper, we draw on data from the Growing Up in New Zealand survey, a longitudinal survey of the families of 6,846 children born in the Auckland, Waikato, and Counties-Manukau regions in 2009-2010. The participating families are roughly ethnically and socioeconomically representative of the overall New Zealand population, though Pacific and Asian mothers are slightly oversampled and Māori mothers are slightly under sampled (Morton et al. 2013; Morton et al. 2014).

We focus on the first five waves of the survey, conducted approximately three months before the children’s birth (the “antenatal” survey) and approximately nine, 24, 45, and 54 months after the children’s birth.

Sample construction
In our analyses, we restrict to a sample of mothers who satisfy three criteria. First, the mother must be present across the five surveys. Imposing this restriction decreases the sample size to 5,605. Second, the mother must be employed (self-employed or an employee) during the antenatal survey, because mothers must be employed to take leave, which is our focus. Antenatal employment is identified using the employment status question in the antenatal survey. This restriction decreases the sample to 3,212. Mothers with previous children are more likely to be dropped in this step than are first-time mothers; the mothers with previous children retained in our sample are those who are more strongly attached to the labour market, as evidenced by their return to work after having their first child. This should be borne in mind in interpreting our results. Finally, we keep only mothers who say they intend to take leave (because only these mothers are asked about their preferred length of leave) and give nonmissing responses to the questions about anticipated and preferred leave (described in the data appendix). These restrictions drop another 421 and 203 mothers respectively. This leaves a final sample of 2,588 mothers, whom we use in our main analyses.

In some sub-analyses, we distinguish between the 1,314 mothers in our sample for whom this is their first child and the 1,274 mothers with previous children.

Some of the mothers in our sample are missing data for covariates we use in our analyses (e.g. income). In our main regressions we retain these observations and include “missing value” dummies. In analyses where we split the sample by whether a mother is above or below median income, we exclude mothers with missing income.

Leave measures
In the antenatal survey, mothers are asked how much leave (paid and unpaid) they would prefer and how much they anticipate. We use their responses as our measures of preferred and anticipated leave. In the 9-month survey mothers who have completed their leave are asked how long it was; this is our actual leave measure, right-censored at 9 months. Our main measure of realised leave is “imputed actual leave”, which is actual leave for mothers who returned to work by the 9-month survey and is imputed based on employment status in subsequent waves for mothers still on leave at 9 months. See the data appendix for a full description of the construction of our leave measures and other variables used in the analysis.

Due to its construction, our “imputed actual leave” variable is continuous up to 39 weeks, but afterwards is clustered at a few discrete points midway between survey waves. Effectively we observe realized leave measured with some nonclassical error. The positive and negative measurement errors are likely to largely balance each other out; to verify measurement error is not driving our findings, we run alternative tobit regressions using actual leave right-censored at 39 weeks.

Empirical strategy
In addition to summarising the outcomes and situations of mothers, we use a range of regressions to explore the conditional relationships between outcomes and mothers’ characteristics.

We first investigate how mothers’ preferences and expectations of leave and leave outcomes vary with their circumstances and characteristics. We regress each of preferred, anticipated, actual, and imputed actual leave on a set of characteristics of the mother and her antenatal situation. The regressions of preferred, anticipated, and imputed actual leave are ordinary least squares regressions; to account for the right-censored nature of actual leave we use a tobit regression. In our conceptual framework, mothers are endowed with preferences and face constraints. Their preferences combine with expected constraints to give anticipated leave and combine with realised constraints to give actual leave. These regressions illuminate the demographic characteristics that predict these variables.

We next explore how the types of leave vary with each other across individuals, which sheds light on the extent to which mothers are constrained in their leave, and their ability to follow their leave plans. We run linear and piecewise linear regressions of one type of leave, such as anticipated leave, on another type, such as preferred leave. We also run versions in which we allow the relationship to differ by several maternal characteristics. These regressions are further explained in the data appendix.

We next use regressions to investigate the types of mothers who are more likely to anticipate less leave than they prefer for each different reason. We include our full sample of mothers and construct binary dependent variables that take the value 1 if the mother reported less anticipated leave than preferred leave and said a particular reason contributed to this. The coefficients thus give the conditional correlation between maternal characteristics and anticipating less leave than preferred for the reason. We use a similar approach to explore the relationship between mother’s characteristics and either returning or not returning to work by 9 months. These analyses of mothers’ stated reasons provide insight into the types of constraint that prevent mothers from taking the leave they prefer or cause them to change their leave plans. Deviations from anticipated leave may be desirable or undesirable depending on whether they’re caused by a tightening of constraints, a loosening of constraints, or a change in preferences. To explore whether deviations from plan are desirable or undesirable, we explore the relationship between deviating from leave plans and wellbeing. We regress measures of stress at 9 and 24 months on three dummy variables that represent (1) not being in work and having anticipated not being in work, (2) not being in work despite anticipating being in work, and (3) being in work despite anticipating not being in work. The omitted category consists of mothers who are in work and correctly anticipated this. We additionally control for whether the mother currently works part-time, whether she is contemporaneously partially or solely self-employed, and a full set of controls for personal characteristics and antenatal situation.

Engagement with policy collaborators
This research project was conceived and conducted in collaboration with policy experts at the Ministry for Women and the Productivity Commission. The policy collaborators were consulted on the research question, analytical approach, several rounds of empirical results, and the interpretation of the results. Their feedback was incorporated into this report. Any errors, omissions, or misinterpretations that remain are the authors’ own.