Microfoundations of the weakening educational gradient in fertility

Figures & data

Figure 1 Association between the Human Development Index and the fertility gap, by world region, 1976–2005

Notes: The HDI is a summary measure of average achievement in three dimensions of human development: length of life, educational attainment, and material living standards. It is calculated as the geometric mean of normalized indices for each of the three dimensions. The fertility gap is measured by subtracting total fertility from the average desired number of children. Observations correspond to different countries and to different points in time for the same country. A fertility gap above zero indicates excess fertility; a fertility gap below zero indicates a fertility deficit. The solid line represent the LOESS regression estimate.

Source: For European countries HDI and TFR data come from Myrskylä et al. (2009) and information on desired family size comes from the Demographic and Health Surveys and from Sobotka and Beaujouan (2014). For the rest of the countries information comes from the Demographic and Health Surveys.

Figure 2 Hazard ratios of transitioning to a second child by cohort and educational level, as reported in Kravdal and Rindfuss (2008): women, Norway

Note: The lowest educational level (primary) is the reference category.

Source: Kravdal and Rindfuss (2008).

Figure 3 Observed and forecasted proportions of women by (a) education level and (b) labour force participation by education: cohorts, Norway

Notes: As the data clearly show, the proportion of women who have attained tertiary education and the proportion of women participating in the labour market have increased significantly over time. Although our simulation model classifies the simulated women as inactive/active in the labour market, here, for the sake of clarity, only the proportion of women actively participating in the labour market is displayed in panel (b). See the Data subsection for more information on these data and indicators.

Source: Population and administrative registers, Norway; National Education Database, Norway.

Table

Obtain an initial sample of X of size n0, θ1, … , θn0

Compute Δθ1,  … , Δθn

Set n = n0

While n ≤ N do

Map the relationship θ → Δθ with a Gaussian process emulator G(·)

Obtain a new sample of X of size n^∗

Obtain predictions for G(θk) k = 1,→, n^∗

Compute acquisition function A(·)

Obtain the new locations to be explored θj

Compute Δθj

Increment n

end While

Figure 4 Observed vs simulated age-specific fertility rates: Norway, selected years

Notes: Observed age-specific fertility rates are plotted as solid diamonds. Dashed lines represent two separate simulations with the set of parameter values that best fit the observed data. The transparent diamonds represent the mean of these two simulations. The shaded area shows 95 per cent prediction intervals (range of values that contains the central 95 per cent of the posterior predictive distribution).

Source: Authors’ analysis; observed age-specific fertility rates were obtained from the Human Fertility Database.

Figure 5 Model fit and 95 per cent prediction intervals: selected fertility indicators, Norway

Notes: Observed age-specific fertility rates are plotted as solid diamonds. Dashed lines represent two separate simulations with the set of parameter values that best fit the observed data. The transparent diamonds represent the mean of these two simulations. The shaded area shows 95 per cent prediction intervals (range of values that contains the central 95 per cent of the posterior predictive distribution).

Source: As for Figure 4.

Figure 6 Hazard ratios for the transitions to second and third births by cohort and educational level for women in Norway: simulated vs observed results

Notes: The lowest education level (primary) is the reference category. Given that the focus of the analysis is the changes over time, the simulation model does not distinguish between different levels of secondary or tertiary education. Simulated results are reported with 95 per cent confidence intervals (shaded areas). The uncertainty of the estimates computed on the observed data is not reported in Kravdal and Rindfuss (2008). See Data subsection for additional details on the educational attainment data. More information on the hazard regression models, including the coefficients presented here, can be found in the supplementary material.

Source: As for Figure 4.

Figure 7 Hazard ratios for the transition to third birth by cohort and educational level for women in Norway: comparison of model specifications

Note: The lowest education level (primary) is the reference category. Simulated results are reported with 95 per cent confidence intervals (shaded areas).

Source: As for Figure 4.

Table

Table A1 Model parameters and best-fitting values: Norway

Name	Domain	Eq.	Description	Value	Source
δ	Desired fam. size	1	Effect on Di of working	0.009	Estimated
β	Intention	2	Baseline intention	0.8	Fixed
λ	Intention	2	Rate of reduction penalty after pregnancy	2.5e−8 s^−1	Fixed
η	Intention	3	Penalty on intention of working	0.57	Estimated
τ	Intention	3	Years of education after which η is reduced	25.1 yr	Estimated
ε	Intention	3	Speed at which η is reduced	0.53 yr^−1	Fixed
ϕ	Fecundability	4	Maximum fecundability	0.22 mo^−1	Fixed
γ	Fecundability	4	Inflection point of decline (age)	38 yr	Fixed
κ	Fecundability	4	Speed of decline with age	0.25 yr^−1	Fixed
A	Contraception	5	Additional effect due to achieved D	0.07	Fixed
ρ	Contraception	6	Governs minimum effect on conception risk	0.037	Estimated
υ	Contraception	6	Maximum effect on conception risk	0.39	Estimated
ψ	Contraception	6	Inflection year in the diffusion process	1977	Estimated
r	Contraception	6	Speed of the diffusion	0.23 yr^−1	Estimated
α	Contraception	6	Differential access/use by education	0.07	Estimated
ξ	Age at union formation	7	Time to union after schooling	4.3 yr	Estimated

Note: The t⁻¹ notation refers to ‘per unit of time’ (where yr = year and mo = month).

Myrskylä, Mikko, Hans-Peter Kohler, and Francesco C. Billari. 2009. Advances in development reverse fertility declines, Nature 460(7256): 741–743. https://doi.org/10.1038/nature08230

 PubMed Web of Science ®Google Scholar

Sobotka, Tomáš and Éva Beaujouan. 2014. Two is best? The persistence of a two-child family ideal in Europe, Population and Development Review 40(3): 391–419. https://doi.org/10.1111/j.1728-4457.2014.00691.x

 Web of Science ®Google Scholar

Kravdal, Øystein and Ronald R. Rindfuss. 2008. Changing relationships between education and fertility: A study of women and men born 1940 to 1964, American Sociological Review 73(5): 854–873. https://doi.org/10.1177/000312240807300508

 Web of Science ®Google Scholar