We now discuss a series of robustness checks to the baseline specifications discussed in Section 5. First, as explained earlier in the paper, we are not able to use family fixed effect models, as we do not observe multiple members of the family of origin for the parents’ generation, namely siblings completed years of schooling. For this reason, we look at the pooled sample. As a robustness check, we run separate regressions for specific family size, following Black et al. (2018) and Bertoni and Brunello (2016). Of course, this implies to estimate the model for smaller samples, but a comparison is needed to check if the birth order effects vary when fixing parental family size.
Results for the pooled sample are reported in Table 7.
In the top panel, we show the results for parents’ years of schooling. Being a firstborn parent is associated with higher years of schooling and such effect monotonically declines with family size. However, the magnitude of the effect is comparable to the one obtained in Table 2. In the bottom panel, we report the estimates for children’s generation, separately for the dyad mother-children and father-children. In line with our baseline estimates, we find that (i) children of a firstborn mother—independently from the family size—complete about 0.2 years of schooling compared to children of laterborn mothers; (ii) the spillover effects for firstborn father’s in terms of children’s education attainment are not statistically significant.
Second, in the baseline model for the children, we include the firstborn dummy separately for each parent. As a robustness check, we estimate a joint model where we include both the birth order and family size of the mother and the father, and additional control variables. This model is estimated for a subsample of SHARE where both the respondent and spouse are interviewed.Footnote 9 The estimates are in line with the findings from the baseline model, namely that children of firstborn mothers have on average 0.188 additional years of schooling compared to children of laterborn mothers, whereas we do not find any statistically significant effect for firstborn fathers. The magnitude of the estimated coefficients is also in line with our baseline model.
Third, we re-estimate the model by excluding parents who are single children. The results are not affected by this choice. We find that being a firstborn parent increases own education attainment by 0.345 years, and children’s education by 0.120 years. The comparison with the baseline estimates holds even when we distinguish by parent and child gender.Footnote 10
Fourth, since we estimated a model with country fixed effects, it could be that results are driven by a specific country in the sample. Since we do not have a sufficiently large sample to estimate the model separately by country, we run the regressions leaving one country out at each time and results are comparable.Footnote 11
Finally, we also re-run our analysis using different samples of parent and children’s birth cohort. In the baseline analysis, we look at parents whose children are older than 25 years old at the time of the interview, to keep a larger sample size and to avoid truncating our data further. Results are robust even when we select parents and children that are not too old. In any case in all specifications, we account for birth year fixed effects for both parents and children, and this accounts for the issue of cohort selection.
All in all, we are confident that our estimates are robust to the specification used and to the sample choice.
Birth order and fertility choices
As discussed in Section 2, different theories and mechanisms can explain the negative correlation between birth order and education attainment. Recent studies have tried to prove the existence of other potential mechanism such as fertility patterns.
A recent paper by Lin et al. (2020) studies the role of unwanted fertility in the observed birth order patterns. The authors rely on the longitudinal micro-data from the US Panel Study of Income Dynamics (PSID) survey which contains a specific module questioning respondents about their behavior during each pregnancy. They find that laterborn children are more likely to be unwanted, simply because it is likely that they were conceived when the family was not planning to have more children. The authors also find that being an unwanted child can lead to negative outcomes later in life, especially on education, as parents could disrupt their investments on unwanted children. This has implications for the laterborn children who are the more affected ones. Thanks to the richness of the data, the authors document that these birth order effects disappear if families can plan optimally and have proper control of their fertility behavior.
Differently from PSID, the SHARE survey does not include questions on perceptions or assessments of pregnancy history. However, we can check if there exists a relationship between parental birth order (being firstborn or not) and parental fertility, namely the probability to have a child and the total number of children. Results are reported in Table 8.
In columns 1–3, we show the results for the probability that a SHARE respondent has a child (we use the full sample of SHARE respondents that do or do not have kids), whereas in columns 4–6 we use as outcomes the number of conceived children. We report results separately by gender: columns 1 and 4 the pooled sample, columns 2 and 5 female respondents, and columns 3 and 6 male respondents. Our results do not point to any significant correlation between parent birth order and fertility preferences. We account for a large number of characteristics as we do in the baseline models, so with the data at hand we can exclude this existence of this channel as a prevailing one.
Birth order and mortality
Empirical studies that test the relationship between birth order and mortality lead to mixed results. Modin (2002) finds that laterborn children have greater mortality compared to laterborn children, whereas other studies do not find a clear pattern, nor significant results. More recently, Barclay and Kolk (2015) using Swedish population register data look into the relationship between individual birth order and mortality later in life.
In our analysis, we look at the effect of parental birth order on children’s educational attainment, so somehow we are interested in selective mortality of both generations. To test for the existence of this relationship, we rely on data from the Human Mortality Database (HMD)Footnote 12 which allows to construct long annual time-series death rates by age, gender, and country for 10 of the countries included in our sample of age and gender-specific death rates for 10 of the countries included in our sample. In order to check for the presence of birth order effects in mortality of two linked generations, and given that for younger cohorts of children we have shorter series, we look at mortality at birth.Footnote 13 We then merge this database by country, calendar year (corresponding to the birth year of parents and children), and gender for both generations. Results are reported in Table 9, where in the top panel we report the results for parents and in the bottom those for the children.
We do not observe any significant relationship between being a firstborn parent and mortality at birth for both cohorts.
Birth order and inheritance
Another plausible channel through which parental birth order could affect children’s education could be through their degree of wealth. For instance, there could be a relationship between birth order and inheritance. In SHARE, we do not have specific data on individual wealth, but only on household wealth at the time of the interview. However, we do have information on whether the SHARE respondents (parents) have received inheritances from their family.
In Table 10, we show estimates from a linear probability model where the outcome variable is a binary indicator which takes value 1 if the parents have ever received a house as bequest (top panel) or have bought/built it with some help from the family (middle panel).
For each outcome, we use two specifications: the full sample in columns 1 and 3 and CSL sample in columns 2 and 4. We notice that being a firstborn parent seems not to have a significant effect on the probability of inheriting a house or building it with some help from the family. Thus, this channel does not seem to matter for parent education.
Birth order and risk preferences
Another potential channel through which parental birth order may affect children education could be related to risk-taking behavior or other types of attitudes. A common discussion in the sociological literature is that laterborn children could make more risky choices compared to firstborn children, to make up for the disadvantage of being a laterborn. Although it is not possible to fully test such assumption, we provide some evidence on the relationship between parental birth order and risk aversion.Footnote 14 We use a subjective measure of risk aversion based on the following question asked to parents:Footnote 15 “Which of the following statements on the card comes closest to the amount of financial risk that you are willing to take when you save or make investments? i) Take substantial financial risks expecting to earn substantial returns; ii) Take above average financial risks expecting to earn above average returns; iii) Take average financial risks expecting to earn average returns; iv) Not willing to take any financial risks.” We create a binary indicator that takes value 1 if the parent is willing to take either substantial financial risks or above average financial risks and 0 otherwise. Results are reported in the bottom panel of Table 10.Footnote 16 Interestingly, birth order seems not to have an effect on risk preferences. The results are in line with most of the research on the topic. A recent paper by Lejarraga et al. (2019) uses a comprehensive approach to study the birth-order effects on risk taking. They draw data from the German Socio-Economic Panel, the Basel-Berlin Risk Study, and additional data sources allowing to measure behavioral traits. The authors find that all these data sources and different analytical methods point toward no birth order effect on risk-taking behavior.
Are firstborns so different from middle and laterborns?
Due to data limitations, we cannot use the complete birth order of parents and clearly distinguish a firstborn from a second born, third born, and so on. If we expect firstborns to be special compared to laterborns (coherent with the idea of a role model), then we should also expect that second borns, third borns, and so on have similar education. This hypothesis cannot be tested using data for parents but we fill this gap using data for children, as we know precisely their birth order and family size. In Fig. 3, we show the residuals from a regression of children’s education on country fixed effects, birth year fixed effects, and the age of the parent (mother or father) at birth, separately by birth order and family size.
The descriptive evidence rejects the hypothesis that firstborns are special. It is clear from the figure that educational attainment decreases monotonically with birth order, keeping a fixed family size.