What Determines the Child Penalty in the Netherlands? The Role of Policy and Norms

Abstract

Having children can result in large earnings penalties for mothers. Using extensive administrative data from the Netherlands, we assess the magnitude and drivers of the effects of first childbirth on parents’ earnings trajectories in the Netherlands. We show that mothers’ earnings are 46% lower compared to their pre-birth earnings trajectory, whereas fathers’ earnings are unaffected by child birth. We examine the role of two potential determinants of the unequal distribution of parents’ labour market costs by gender: childcare policies and gender norms. We find that while child care availability is correlated with lower child penalty, the immediate short-term causal effect of increasing child care availability on the earnings penalty of becoming a mother is small. By taking advantage of variation in gender norms in different population groups, we show that gender norms are strongly correlated with child penalty for mothers.

Appendices

Additional Tables and Figures

1.1 Additional Tables

See Tables 2, 3.

Table 2 Descriptive statistics by subgroups

1.2 Additional Figures

See Figs. 12, 13, 14, 15, 16, 17.

Fig. 12

Impact of children on mothers’ and fathers’ careers. This figures present the counterfactual and predicted earnings used for the computation of the child penalty, for mothers and fathers and for the different labour market outcomes we consider. Predicted earnings is the predicted value of the outcome based on the estimation of Eq.(1). Counterfactual outcome is the average predicted labour market outcome at event time q, without the event dummies: $\tilde{y}_{it}^q= \frac{1}{N}\sum _i\big [\sum _k{\hat{\beta }}_k \mathbbm{1}[age_{it}=k]+ \sum _t{\hat{\gamma }}_t\mathbbm{1}[time_{it}=t]\big ]$

Fig. 13

Distribution of the years of large childcare expansion. This figure present the distribution of the years for which we observe a childcare expansion (see Sect. 4 for details about the definition of the event)

Fig. 14

Mothers’ child penalty and religiosity by city. This figure presents, for each municipality, the earnings child penalty for mothers (a) and the average religiosity (b). Religiosity is defined as the share of individual going to a religious place at least once a month. See text of Sect. 5 for details

Fig. 15

Child penalty for high and low religious group, with and without matching. This figure presents the estimated earnings child penalty for mothers in high and low religiosity, without weights (A) and with weights (B). High (resp. low) religiosity group is composed of municipalities in which more than 25% (resp. less than 15%) of the population visit a religious place at least once a month. Weights are computed using a coarsened exact matching (CEM) procedure based on education and employment trajectories in the 5 years preceding childbirth (see text in Sect. 5 for details)

Data Construction

1.1 Construction of the Hours Worked Variable

In the baan datasets that we use there is no direct information available on the hours work. We do have access to a variable on the full time equivalent associated to a given job spell. For example, a person working 20 hours when the full-time reference is 40 hours, the % of FTE worked is equal to $\frac{20}{40}=0.5$. The problem is that we do not observe the full-time reference hours, which would make it possible to retrieve the hours worked from the FTE. We therefore adopt the following approach: we use another data source, the polis datasets (see Sect. C above) which contains information on hours actually worked, to impute hours worked in our main estimation sample.²⁴ More precisely, we compute for each each the distribution the yearly hours worked in each sector using the polis data, and take the mode of the distribution as the reference work duration in the sector. This approach generates some measurement errors since the reference duration is determined at a more disaggregated level (collective agreement or even firm) than the one we are using (sectors, split in 70 categories). Another potential source of measurement errors comes from the fact that the polis dataset starts in 2006, and that we use the distributions observed at that year for the computation of the reference durations for years 2001–2005.

Figure 16 compares the observed and imputed distribution, based on year 2010, for which we can observe the hours worked in the polis dataset. Panel (a) compares the observed and imputed distribution of yearly hours worked, and panel (b) presents the distribution of the difference between the observed and imputed values. These figures confirm that we have measurement error in our measure of hours worked, but that the imputed values are on average very close to the observed one for the comparison sample.

1.2 Construction of the Childcare Index

For the construction of our childcare exposure index (Eq. 4 in Sect. 4), we need to compute the number of childcare jobs existing in each municipality. We do not have direct information on that, for example from municipalities’ expenditures as in Andresen and Nix (2019) or Kleven et al. (2020). We therefore use firm-level information on job location combined with sector information to compute our index. More precisely, we compute for each year and city the number of jobs in two different 5-digit sectors: 88911 (day nurseries for pupils) and 88912 (Kindergartens). We then divide this number of jobs by the number of below five years in old by year and city to compute our index.

Panel (a) of Fig. 17 presents the evolution of the index at the country level. We observe a sharp increase in the number of childcare jobs from year 2006. The overall childcare exposure more than doubles between 2006 and 2012, and then slightly decreases. This evolution is consistent with the timing of implementation of the childcare reform described in Sect. 4.1, as illustrated in panel (b) presenting the evolution of the overall public childcare expenditure. Reassuringly, our index closely follow the evolution of aggregate public expenditure, which suggests that we are capturing the effect of the childcare reform through it.

Fig. 16

Hours worked variable. These figures compare the observed and imputed hours for workers observed in 2010 in both the baan and polis datasets. a shows the overlapping distributions of imputed and observed yearly hours. b shows the distribution of the gap between observed and imputed hours, computed for each individual

Fig. 17

Evolution of childcare exposure index and childcare expenditure. The red line presents the yearly evolution of the aggregated childcare exposure index, measured as the ratio between the number of childcare jobs and the number of children below five years old. The green line presents the evolution of the total public spending for childcare (source: Ministry of social affairs). Both time series are shown relative to the 2005 value

Description of the Databases

In this appendix we describe the different data-sources used in the paper. Table 3 lists all the sources we use, and we then provide more detailed information about each of the data we use.

Table 3 Description of the datasets used in the analyses

1.1 Dataset Provided by Statistics Netherlands

Information about the administrative micro data sets can be found at https://www.cbs.nl/nl-nl/onze-diensten/maatwerk-en-microdata/microdata-zelf-onderzoek-doen/catalogus-microdata (available in Dutch only)

1.1.1 gbapersoontab²⁵

It contains demographic background data (e.g. gender, year of birth, migration background) for the universe of the Dutch population, that is all persons who appear in the registered in the population register (Basic Register of Persons, BRP) since 1 October 1994.

1.1.2 gbaoverlijdentab²⁶

Contains the date of death of all persons who have died since 1 October 1994 and were registered in the population register (Basic Register of Persons, BRP) at the time of death. It also contains the date of death of persons who are not residents but were once residents of the Netherlands since 1 October 1994 and whose information about the death is received in the Register of Non-Residents (RNI). The main source of information for this dataset is the municipal registries (Gemeentelijke Basisadministratie Persoonsgegevens, GBA).

1.1.3 gbamigratiebus²⁷

It contains all migration spells for the full universe of the Dutch population (as defined in the gbapersoontab). For each immigration (resp. emigration) spell, a date of beginning and end is registered, as well as the country of origin (resp. destination). For each individual, we have as many spells as migration events occurring since 1994. The main source of information for this dataset is the municipal registries (Gemeentelijke Basisadministratie Persoonsgegevens, GBA).

1.1.4 gbahuishoudensbus²⁸

For the full universe of the Dutch population (as defined in the gbapersoontab), it contains information about the household composition: their place in the household, and the details of the household they belong to (e.g couple or not, married or not, with or without children, etc.). Retrospective information is available, as the data is presented as spells (one additional line when one characteristic of the household changes). The main source of information for this dataset is the municipal registries (Gemeentelijke Basisadministratie Persoonsgegevens, GBA).

1.1.5 GBAADRESOBJECTBUS²⁹

For the full universe of the Dutch population (as defined in the gbapersoontab), it contains information about the individual address, with a unique object identifiers at the level of the building. Retrospective information is available, as the data is presented as spells (one additional line when one characteristic of the address changes). The main source of information for this dataset is the municipal registries (Gemeentelijke Basisadministratie Persoonsgegevens, GBA).

1.1.6 baankenmerkenbus³⁰

A spell database with information on job characteristics (contract, sector) for the full universe of the jobs. It is available for years 1999-2016.

1.1.7 baansommentab³¹

A yearly database with information on wages (gross wage, fiscal wage, full time equivalent) for the full universe of the jobs. It is available for years 1999-2016.

1.1.8 polisbus ³² and spolisbus ³³

It contains information on the full universe of job in the Netherlands, available from year 2006. There is one line by employment spells, with information on both the individual (wage, hours worked, contributions, etc) and the firm (sector, collective agreement, etc).

1.1.9 gemstpl

A yearly database containing information on job location at the municipality level for the full universe of job in the Netherlands existing at the end of the year (31/12). The address is based on the firm address. When there are several establishment for a given firm, the location is imputed by Statistic Netherlands as follows. The location attributed to a given job is the one which is the closest to the home address of the worker.

There are different version of this dataset, depending on the years considered: GEMSTPLAATSBUS for 1999-2005, GEMSTPLBUS for 2006-2014, GEMSTPSBUS for 2015-2019.

1.2 Other Datasets

LISS panel The LISS (Longitudinal Internet Studies for the Social sciences) panel administered by CentERdata (Tilburg University, The Netherlands) is a representative sample of Dutch individuals who participate in monthly Internet survey on household background information. The panel is based on a true probability sample of households drawn from the population register. Households that could not otherwise participate are provided with a computer and Internet connection. A longitudinal survey is fielded in the panel every year. The information we use in this paper is retrieved from the religion and ethnicity; and the political views and value module.

1.3 Survey on Religiosity

We use a survey on religious practices conduct by CBS to construct our religiosity index.³⁴ See Schmeets (2016) for a description of the data and a more complete analyses of religious practices in the Netherlands.