The Value of Postponing Pregnancy: California’s Paid Family Leave and the Timing of Pregnancies

Tax policies can affect the timing of marriages (Alm and Whittington 1997) or divorces (Dickert-Conlin 1999) as well as the decision to enter a marriage agreement (Alm and Whittington 1999). When social insurance eligibility depends on marriage, this has been shown to affect the composition of marriages, entry and exit into the marriage contract, as well as the assortative nature of marriages (Persson 2014).

The maximum weekly benefit increased each year since the law went into effect in 2004, based on an inflation factor. In 2012, the maximum weekly benefit was $987.

The estimate of 22% who were aware of the PFL program, discussed in Appelbaum and Milkman (2004), is from a survey covering the entire adult population over 18 years old in California. Thus, this estimate is likely under-reporting the awareness among adults considering to have a child within the next year, as for many adults in the survey, knowledge of PFL benefits may not be relevant. For this reason, this estimate is not used to adjust the estimates in the forthcoming analysis and obtain the treatment on the treated effect.

Until January 1, 2011, Australia also did not provide any paid maternity leave.

Source: http://www.dol.gov/whd/fmla/chapter3.htm

For pregnancy, the start of disability insurance eligibility is 4 weeks before the expected due date.

Father’s education level is no longer provided in the natality data beginning 1995.

Information about smoking during and before pregnancy is available for all states, except California.

Note that PFL benefits for newborn bonding can be utilized up to 1 year after the birth of a child. Thus, theoretically, it may be worthwhile to postpone a birth by a few months to August or September 2003, so that PFL benefits can be utilized beginning July 1, 2004.

Studies have shown that the probability of a clinical pregnancy during the first two cycles attempted among healthy women is roughly 50%. In a study tracking over 500 healthy women attempting to conceive, roughly 1.5% completed the 1-year study without a clinical pregnancy (Wang et al. 2003).

Early pregnancy loss (or miscarriage) is defined as the spontaneous loss of pregnancy before 20 weeks’ gestation. About 15% of known pregnancies result in a miscarriage (source: The American College of Obstetricians and Gynecologists – https://www.acog.org//media/For%20Patients/faq090.pdf?dmc=1&ts=20140703T2234081180).

Additional analysis suggests that any biases resulting from the use of a linear probability model are likely to be small. Specifically, all predictions from the regressions in the analysis fall in the [0,1] range (even for the monthly regressions which have relatively low probabilities). Furthermore, probit regressions of eq. [1] without the interaction term resulted in nearly identical estimated marginal effects.

County-level characteristics could only be matched to the natality data when the mother’s county of residence was identified, and this is only the case for counties with a population exceeding 100,000. The mother’s county of residence has a population less than 100,000 for ~25% of the births in the data. In order to not omit these births from the analysis and bias the sample to births only occurring in large counties, annual state-level characteristics, rather than the county-level characteristics, were used as covariates for observations for which the county was not identified.

Percent employed in the public sector may have an effect on the fraction of employees in California eligible for the PFL benefits, as most public-sector employees are excluded from the PFL benefits.

For computational feasibility, the set of control counties for which positive weights are assigned to in the synthetic control method is limited to counties with an average population for the period 2001, 2002, and 2004 ranging between 0.5 and 1.5 of the average population of the treated county for that same period, or to counties with populations exceeding 500,000 if the treated county’s population exceeded 1 million. The set of control counties is then further narrowed based on similar demographic characteristics at the county level (available at http://www.census.gov/support/USACdataDownloads.html) – in particular from within the set of counties with similar population levels, the counties were selected as potential control counties if their median household income, percent in poverty, percent white, percent black, percent Hispanic, or births per 1,000 women were sufficiently similar to that of the treated county (to emphasize – not all characteristics had to be similar, just one had to be similar for the county to be added to the pool of potential control counties). This resulted in a pool of potential controls ranging from 57 to 242 for each treated California county, out of a total of 489 identifiable counties.

The population segments are the following characteristics for the mother (in order of their appearance in Table 2): child born is not the first child, 12 years of schooling or less, 13–15 years of schooling, 16 years of schooling, more than 16 years of schooling, black; white (non-Hispanic), Hispanic; not married, 25 years old or less; 26–34 years old, and 35 years old or more.

A county-level analysis can be used to look into differential effects based on county characteristics. An analysis was conducted which evaluated the effect of PFL on births during July through December based on female labor force participation rates in 2000 or the percent employed in the public sector in each California county and produced no statistically significant results. The underlying assumption was that the effect of a change in the birth distribution should be greater for counties with higher female labor force participation rates and of a lesser extent for counties with greater public-sector employment, as public-sector employees are mostly not included in California’s State Disability Insurance program, which includes the PFL.

In California, 36 counties have populations exceeding 100,000 during all 3 years in the analysis. In total, California has 58 counties. In the United States, only 573 counties have a population exceeding 100,000 during all three sample years in the analysis, out of 3,147 counties in the United States.

Regressions at the individual birth level (as presented in Table 2) were also run while limiting the sample to other TDI and the next three largest states after California, and the results were very similar to those in Table 2, with the exception of no statistical significance for the positive differential effect of the PFL on mothers for whom this was not their first child and unmarried mothers. All other differential effects based on mothers’ characteristics were very similar in terms of their magnitude and statistical significance to those reported in Table 2.

The exact percentage changes for May/August is not displayed in Figure 2. The proportion of births in May/August for California in 2001–2002 is shown in Figure 1.

Or alternatively, some couples may have chosen to postpone a pregnancy to several months after PFL’s introduction due to personal reasons leading to not wanting a birth specifically in July or August.

For 2001, 2002, and 2004, this resulted in omitting a little less than 1% of the sample of all births, due to also having some missing values for this variable in these years.