Does the birth order affect the cognitive development of a child?

Abstract

This article investigates the link between position in the birth order and early scholastic ability. Using matched mother–child data from the National Longitudinal Survey of Youth (1979 cohort, NLSY79), I find that being the first-born is beneficial even after controlling for (nonlinear) effects of family size and child characteristics. The verbal ability of first-borns is about one-tenth of a SD higher than for children in the middle of the birth order. There is no evidence that last-borns fare better than intermediate children. The first-born advantage is confirmed by estimates from within-family variation models and I argue that the findings are consistent with the resource dilution hypothesis.

Acknowledgements

I thank Warren C. Sanderson, Mark R. Montgomery, Michael Grossman, Shirley Liu, Hugo Benítez-Silva, Chris Swann, Debra Dwyer, seminar participants from the Center of Demography and Population Health at Florida State University, and by participants at the 2005 Meetings of the Population Association of America, the 2003 Meetings of the Population Association, the Southern Economic Association and the Eastern Economic Association. This article also benefited greatly from comments and suggestions by an anonymous referee. This research was funded by a grant from the National Institute of Child Health and Human Development (NICHD), Demographic and Behavioral Sciences Branch (Grant R03 HD046604-01).

Notes

¹ For example, Peabody Picture Vocabulary Test (PPVT) scores, the test of early verbal ability studied in this article, have been found to be highly correlated with subsequent scholastic achievements and general measures of (verbal) intelligence (Dunn and Dunn, 1981; Center for Human Resource Research, 1995, p. 16).

² Economists have provided estimates of the economic returns to schooling. In his classic work Becker (1975), for example, estimates that the rate of return to a college education was 14.8% in 1959. For a recent survey, see Card (1999).

³ For example, parents who depend on their children for support at old-age or to take care of younger siblings may invest more in children early in the birth order (e.g. Kessler, 1991).

⁴ A related literature has included the sex composition of siblings as a determinant of child achievement. Butcher and Case (1994) present evidence that girls raised with brothers attain more education than girls raised with at least one sister. The validity of this evidence has been questioned subsequently (Hauser and Kuo, 1978; Kaestner, 1997).

⁵ Related literatures have studied the role of birth order in other outcomes. Argys et al. (2006) find evidence that being later in the birth order is associated with risky behaviour. Some recent popular books relate birth order to personality differences (Sulloway, 1997; Wallace, 1999; Lehman, 2001). For a critique of this evidence, see Freese et al. (1999).

⁶ Some studies estimate outcome equations that control for the number of older and younger siblings (e.g. Peraita and Sanchez, 1998). As pointed out by Olneck and Bills (1979), the differences between these effects is identical to the birth order effect estimated from a model linear in birth order and family size. The estimates in Peraita and Sanchez (1998) suggest that having an older sibling is worse than having a younger sibling (implying a negative birth order effect in this restrictive specification), but it is not clear if the difference is statistically significant.

⁷ The validity of their evidence has been questioned since it is based on estimates of the child's absolute position in a family rather than the effect of being born early or late relative to one's siblings (Kessler, 1991).

⁸ Some recent studies also questioned whether family size has a negative effect on child outcome (Guo and VanWey, 1999; Black et al., 2005). The inverse nature of this relationship had been established by numerous studies (e.g. Stafford, 1987; Downey, 1995; Haveman and Wolfe, 1995).

⁹ Some studies on developmental outcomes of children have used larger samples from the NLSY79 (e.g. Joyce et al., 2000; Waldfogel et al., 2002). Since their focus is not on birth order, these studies do not systematically account for family size and other confounding variables.

¹⁰ The low-income White over-sample is choice-based and is therefore excluded from the analysis. Interestingly, Blau (1999) found identical results with and without the low-income White over-sample in his study.

¹¹ Since the low-income Whites over-sample is excluded I do not use the sample weights in the analysis following the recommendations by the Center for Human Resource Research (1995).

¹² Since the PPVT is based on receptive hearing of standard American English vocabulary, its cultural fairness has been debated (e.g. Washington and Craig, 1999).

¹³ Since the 1988 survey, Hispanic mothers in the NLSY79 had the option to have the Peabody administered to their children in Spanish.

¹⁴ A comparison of the PPVT-R and a successor test kit that became available in 1997 suggested that the vocabulary items in the updated version may be less culturally biased (Washington and Craig, 1999). The NLSY79 continued to use the PPVT-R in 1998 for comparability.

¹⁵ The PPVT-R scores adjusts for age at the time of test administration.

¹⁶ If the child is an only child, the dummy variables for being the ‘first-born’, ‘middle-born’, and the ‘last-born’ are all coded zero.

¹⁷ Note that in families with exactly two children, the two children are each coded as ‘first-born’ or ‘last-born’, as there are no ‘middle-born’ child.

¹⁸ The employment spells include periods of vacation, paid sick leave and paid maternity leave.

¹⁹ The income measure is deflated using the Consumer Price Index (1982–84; all urban consumers).

²⁰ Notice that the average reported in Table 3 is lower than 12 grade levels, since I code an educational attainment of 0 if no spouse resides in the household.

²¹ Birth-order effects are identified by the variation that exists between children of the same rank in different families as well as the variation between children of different ranks in the same family.

²² It should be noted that within-estimators are inefficient in the sense that they do not use all variation in the data. In addition, identification relies on variations in sibling outcomes among families with at least two children with test scores only (i.e. the sub-sample of children who are the ‘only-child’ do not contribute to the identification of birth order effects). As a result, the sibling sample may systematically differ from the sample of all families, and the issues that I will address in more detail in the following text. Also, family fixed effects does not correct for child-specific unobserved heterogeneity that may be correlated with inputs in the production process (neither does OLS or RE of course). For an early critique of the siblings approach see Chamberlain and Griliches (1975).

²³ See Becker and Lewis (1973) for their classic work on the trade-off between quantity and quality of children.

²⁴ Since race/ethnicity is defined according to the mother, it does not vary across children born to the same mother and therefore its effect cannot be identified in the FFE and SFD models.

²⁵ Low birth-weight is generally found to be a good predictor of subsequent health. A discussion of the long-term developmental problems of low birth-weight children can be found in Hack et al. (1995). Currie and Gruber (1999) and Corman and Chaikind (1998), among others, provide evidence that low birth-weight children are more likely to display poorer health and scholastic performances compared to their normal birth-weight peers.

²⁶ It should be noted that this is accounted for using controls for family size. Moreover, since most studies on sex preferences in the US find that parents consider one child of each sex as ideal (e.g. Angrist and Evans, 1998), a strong correlation between child gender and birth rank in the NLSY79 is unlikely.

²⁷ The main specifications are comparable to Kessler (1991), among others.

²⁸ I also estimated the models with an indicator for whether the child is a twin (1.7% of the children in the sample are born as twins, see Table 3). The motivation is that twin-births potentially introduce noise in the birth order and family size estimates. However, the estimation results from the models with control for twin-birth are essentially the same as the ones reported here. These additional results are available from the author upon request.

²⁹ While models of the absolute position like Model (1) have been investigated by several authors (e.g. Olneck and Bills, 1979; Behrman and Taubman, 1986; Kantarevic and Mechoulan, 2006), they provide biased estimates of the effect of the relative birth position within a family even after accounting for family size as first pointed out by Kessler (1991, p. 417).

³⁰ The OLS model in Table 6 explains 37.8% of the variation in child test scores compared to 29.5% for the basic specification in Model (4).

³¹ As shown at the bottom of Table 6, the null hypothesis that there is no variation in a family-specific random effect (Var(u(i)) = 0), as assumed in the OLS specification (Section ‘Statistical models’), is rejected at the 1% significance level.

³² Note that the family size effects are identified since the number of siblings can differ across siblings if births occur after an older sibling has taken the test.

³³ As shown at the bottom of Table 6, the Hausman Test statistic of 67.5 rejects the RE specification in favour of FFE at the 1% significance level. This suggests that treating the u(i) as a random effect (RE) uncorrelated with the explanatory variables may be inappropriate.

³⁴ The OLS and RE estimates in Table 6 also indicate that the disadvantage of being the last-born rather than a middle-born child (Models (3) and (4) in Table 3) disappears once parental resources are accounted for. Additional analysis using FFE models, however, suggests that the change in the coefficient for the last-born child may be related to unmeasured family background factors correlated with birth order and ability rather than differences in the availability of resources across siblings. Specifically, the estimated coefficients of Model (4) in Table 5 using FFE are 1.930 (first-born) and 0.253 (last-born).

³⁵ These differences are not the result of the fact that FFE exploits the variation of families with at least two children in the sample. Estimates from the OLS and RE models using the sibling sample (2 or more children with valid observations per family) find similar (or smaller) family size effects in absolute terms. Specifically, the RE coefficients based on the sibling sample are 5.081 (only child), 2.403 (two children), and −2.388 (four children).

³⁶ Notice that this association may be countered by parental preferences for the gender composition of the family. In particular, given that the preferred gender composition of the family in the US is to have one male and one female child (Angrist and Evans, 1998), children later in the birth order are increasingly likely to have (older) siblings that are all of the same sex.

³⁷ These additional results are available from the author upon request.

³⁸ See Edwards and Grossman (1979) for one of the first comprehensive studies of the determinants of child development.

³⁹ The within-family variation is insufficient to identify this effect in the FFE and SFD models.

⁴⁰ Recent studies find that maternal employment or the amount of time mothers spent in the labour market may be detrimental early in the child's life but beneficial later (Baydar and Brooks-Gunn, 1991; Blau and Grossberg, 1992; Ruhm 2000; Han et al., 2001; Waldfogel et al., 2002; Baum, 2003; Heiland, 2003).