Sibling Gender and Academic Achievement: Evidence from Austrian HBSC Data
Submitted for EC338 (Econometrics 2: Microeconometrics) Group Assignment, 1000 words
Abstract
We study whether the gender of a child’s only co-resident sibling affects academic achievement among 11, 13 and 15 year olds in Austria. Using three waves (2006, 2010, 2014) of the Health Behaviour in School-Aged Children survey, we restrict attention to two-child households and estimate OLS models with school-clustered standard errors and rich demographic controls. Having a brother rather than a sister has an estimated effect of −0.002 standard deviations, with tight confidence intervals around zero, and we find no differences by gender or for same-sex versus opposite-sex pairs. Overall, sibling gender appears irrelevant for academic outcomes in this setting.
Introduction
Family characteristics have long been linked to children’s development (Sharpe and Curwen, 2024). Sibling spillover effects have become a prominent research topic only recently (Ferreira, 2023). Siblings may assist with schoolwork or affect motivation and emotional well-being, influencing academic outcomes (Kipp, 2015). One might expect sibling gender to matter, shaping role models, parental investment, or gendered expectations, yet evidence is mixed: some studies find effects that vary by race or birth order (Eirich, 2010; Zang et al., 2023), and Troesch et al. (2018) report stronger spillovers in same-gender pairs.
Using Austrian HBSC data, we estimate the causal effect of having a brother rather than a sister on perceived academic achievement. Focusing on two-child families, we exploit the quasi-random “biological lottery” of sibling sex and estimate OLS models with controls for demographics, family structure, socioeconomic status, and survey-year fixed effects.
Data
Dataset
We use the 2006, 2010 and 2014 Austrian waves of HBSC, a school-based survey of pupils aged 11, 13 and 15. Questionnaires cover health, family background and school experiences. We pool the three waves to increase precision.
Sample Selection
We retain respondents with non-missing data on sex, age, number of siblings in the main home, perceived academic achievement and perceived family affluence. To define a clear treatment, we restrict to pupils living with exactly one co-resident sibling and construct an indicator Brother = 1 if that sibling is male. We also record how many biological parents (0,1 or 2) live at home. Perceived achievement is reported on a four-point scale; we reverse it so higher values mean better performance and standardise it. Our final sample contains 5,215 pupils.
Descriptive Statistics
Table 1 shows that 51.2% of respondents are girls, mean age is 13.3, and pupils live with 1.8 biological parents on average. Perceived family affluence averages 2.1 on a 1-5 scale, where higher values indicate lower affluence. Among those with one sibling, 50.9% have a brother (Table 2). By construction, the standardised achievement measure has mean zero and unit variance, with values ranging from about-2.25 to 1.50 (Table 3). Overall, the sample is balanced by gender and displays sufficient variation in sibling gender and family background.
Empirical Analysis
Methodology
Estimation Equation
We study the effect of sibling gender on the standardised perceived academic achievement 𝑌𝑖𝑠 for child 𝑖 in school 𝑠. Our main specification is:
where Brother𝑖𝑠 = 1 if the child’s only co-resident sibling is a brother; 𝑋𝑖𝑠 is a vector of individual and family controls, including age, gender and dummies for the number of co-resident parents and perceived family affluence; and 𝛿𝑡(𝑖) are survey-year fixed effects (2006, 2010, 2014). We estimate OLS with standard errors clustered at the school level. We also estimate (i) an otherwise identical model with SameSex𝑖𝑠 (an indicator variable for one’s co-resident sibling being the same sex) as the treatment, and (ii) separate regressions for boys and girls.
Identification Strategy
Our estimand is the average causal effect of sibling gender on perceived academic achievement among children in two-child families. Restricting to pupils with exactly one co-resident sibling avoids confounding gender composition with sibship size and birth-order structure. Under the “biological lottery” argument, sibling sex is approximately random: conditional on covariates and survey-year effects,
so 𝛽1 captures the causal effect of having a brother rather than a sister. The same reasoning applies when the treatment is SameSex𝑖𝑠.
Balance Table
We assess this quasi-randomness with a balance check; predetermined characteristics should be balanced across the treatment and control group.
Table 4 shows that children with brothers and sisters have similar means for age, gender, family structure and perceived family affluence; the same holds for same-sex versus opposite sex sibling pairs. Regressing each characteristic on the treatment indicator and survey-year dummy yields statistically insignificant differences, consistent with approximate random assignment.
Limitations
Exogeneity may not hold perfectly. Fertility stopping decisions (e.g., a preference for “one boy, one girl”), unobserved attitudes to education, and the absence of parental education in our data could generate subtle selection in the two-child sample. We therefore interpret our estimates as approximately causal, applying specifically to Austrian two-child families observed in HBSC.
Model Specification
Achievement is reported on a four-category scale; we reverse and standardise it and estimate OLS, which is appropriate for approximately continuous ordered outcomes. Ordered probit gives similar results (see §3.4).
Controls include only predetermined or slow-moving factors: child’s age and gender, number of co-resident parents, perceived affluence, and survey-year dummies. These capture demographic and socioeconomic differences without conditioning on post-treatment outcomes.
Results Discussion
In the main OLS specification (Table 5), the estimated effect of having a brother rather than a sister is essentially zero (≈ −0.00SD) and far from significant: the standard error is about 0.03 and p-values exceed conventional thresholds. Replacing the treatment with a same-sex-sibling indicator (Table 6) yields similarly small, insignificant coefficients. Separate models for boys (Table 7) and girls (Table 8) also show no meaningful differences: point estimates are small, opposite in sign, and have overlapping confidence intervals.
These magnitudes are negligible. A one unit change in the standardised outcome equals a full standard deviation, and the 95% confidence interval around the main coefficient is roughly 0.05−0.06 SD. Even in the most favourable direction, having a brother instead of a sister would shift achievement by at most about five percent of a standard deviation, far less than one category on the original scale.
A concern with null results is low power. Given our standard error of about 0.03, the minimum detectable effect at the 95% level is approximately 0.05 standard deviations. The point estimate lies well within this bound, indicating the absence of an effect reflects a genuinely small or zero relationship rather than imprecision.
Extensions
We consider two extensions. First, allowing heterogeneity by child gender and same- versus mixed sex sibling pairs does not change conclusions: coefficients remain small and statistically insignificant, implying neither within-pair peer mechanisms nor gender-specific parental responses generate detectable differences in performance. Second, ordered probit models with the unstandardised reversed scale (Table 9) yield qualitatively identical results, with marginal effects near zero and high p-values. Together, the null effect of sibling gender on academic achievement is robust across specifications, heterogeneity splits and functional-form assumptions.
Conclusion
Using three Austrian HBSC waves for two-child households, we find no evidence that sibling gender affects perceived academic achievement: effects are essentially zero and even modest impacts can be ruled out. Thus, in two-child families, sibling gender composition is unlikely to be a useful policy lever; resources should target more powerful determinants such as socioeconomic disadvantage. Future research could examine other outcomes (time use, aspirations, well-being), replicate in other countries and family sizes, and study larger families using the share of male versus female siblings
Statement of Contribution
All members contributed equally to the paper.
Bibliography
[1] Sharpe, G. and Curwen, T. (2024). An Examination of the Importance of Gender and Sibling Characteristics on Academic Perceptions. Journal of Education and Human Development, 13(2), 78–90. https://doi.org/10.15640/jehd.v13n2a9
[2] Ferreira, J.P.R. (2023). Sibling Spillovers in Educational Achievement: Evidence from Tanzania. PhD thesis. https://www.novasbe.unl.pt/Portals/0/KnowledgeCenters/ Economics%20of%20Education/Teses/2023/TM_Joao_Ferreira_2023.pdf
[3] Kipp,A.J.(2015).The influence of sibling presence on grade point average. Theses, Dissertations and Capstones, Paper 924. https://mds.marshall.edu/etd/924
[4] Eirich, G. (2010). Educational Inequality Between Brothers and Sisters in the United States. International Journal of Education. https://doi.org/10.5296/ije.v2i2.521
[5] Zang, E., Tan, P. L. and Cook, P. J. (2023). Sibling Spillovers: Having an Academically Successful Older Sibling May Be More Important for Children in Disadvantaged Families. The American Journal of Sociology, 128(5), 1529–1571. https://doi.org/10.1086/724723
[6] Troesch, L. M., Ledermann, T., Jones, J. W. and Grob, A. (2018). School Engagement and Achievement in Sibling Pairs: Gender and Birth Order Matter. Journal of Relationships Research, 9, e19, 1–8. https://doi.org/10.1017/jrr.2018.18
[7] University of Bergen (2025). HBSC Data Management Centre. https://www.uib.no/en/ hbscdata/113290/open-access (Accessed 9 November 2025)
Result
Mark: 74% (Lower First)
Feedback: The paper is very well written, concise, and technically polished. The introduction motivates the question effectively, and the modelling choices are clearly justified. By restricting to exactly one co-resident sibling, the paper correctly exploits the quasi-random assignment of sibling gender within fixed sibship size. The identification assumption is explicit and supported with a balance table and regression-based balance checks, which is unusually thorough. Interpretation of results is particularly strong. The authors express effects in standard-deviation units and show what effect sizes the 95% confidence intervals can rule out. Mechanisms are examined via sex-specific estimates, and robustness is demonstrated using ordered probit models. . Overall, this is a very strong empirical exercise, though the presentation would be improved by avoiding pasted raw Stata output and instead formatting results into clean, readable tables.