Maternal longevity and the sex of offspring in pre-industrial Sweden

Abstract

Background: Helle et al. (2000. Sons reduced maternal longevity in preindustrial humans. Science, 296, 1085) argued that giving birth to sons reduced maternal longevity in pre-industrial societies due to higher physiological costs of bearing sons and the elevated testosterone levels observed in mothers carrying male foetuses.

Aim: The present study examined this hypothesis using a more comprehensive dataset and evaluated the merits of the statistical approach used in previous studies to identify the cost of giving birth to sons in terms of maternal old-age longevity.

Subjects and methods: The analysis in Helle et al. (2002. Sons reduced maternal longevity in preindustrial humans. Science 296, 1085) was extended by using a considerably larger dataset of pre-industrial Swedish women, and with careful consideration paid to methodological problems of sample selection and omitted variable bias. We argue that the previous literature has underestimated the difficulties in quantifying the trade-off between parity and longevity due to unobserved heterogeneity in health. However, under less restrictive assumptions, one can estimate the marginal impact of a son for a fixed family size.

Results: No evidence was found of a negative relative impact of sons. Neither was any evidence found in favour of the male-biased intra-household resource competition hypothesis proposed elsewhere in the literature, despite the poverty of the study population. These results are robust to a wide range of specifications tested.

Conclusion: The failure to reproduce earlier findings and the fact that studies in this area of research seem to continue to yield conflicting results warrant much caution in discussing and evaluating results. It is likely that the negative effect of sons, if it existed, only manifested itself under conditions that are not yet fully understood. We also argue that the previous literature on this topic has not fully acknowledged the inference problems associated with omitted variable bias and sample selection.

• Maternal longevity
• sons
• demography

Notes

Notes

1.  Even though we frame much of our methodological discussion using Helle et al. (2002) as our point of departure, we want to emphasize that most of the literature, in our view, is plagued by similar problems.

2.  In the restricted sample, the potential sample selection problem still persists, however.

3.  However, it is possible that a selection bias  – if it exists  – differs in severity across samples, implying that our results are not fully comparable. We ask the reader to bear this caveat in mind in interpreting our results.

 4. We test this by restricting the sample to workers and running a logistic regression with sex of offspring on a skilled worker indicator variable. Standard errors are clustered at the family level. The skilled worker indicator variable is not statistically significant, despite a sample size of over 10 000 children. Repeating the same analysis on a sample of farmers, tenants and crofters, we find that a freehold farmer dummy variable is statistically insignificant as well.

 5. The estimates we here cite from other studies were obtained simply by calculating from Equation 1. However, since neither Beise and Voland (2002) nor Helle et al. (2002) report the covariance between β₂ and β₃, we cannot calculate the standard error of , and so we are unable to comment on statistical significance in these studies.

 6. We have also run regressions (1) to (3) in Table I with father's longevity as the dependent variable. The estimated coefficient varies between −0.129 and −0.148 and is never statistically significant.

 7. Henceforth, any results cited in the text which are not reported in the tables are available from the authors on request.

 8. The results in Figure 2 refer to the simplest specification with only the number of children and the husband's age at death as control variables. We have also run the regressions with the full set of control variables, obtaining very similar results (not reported here but available from the authors upon request). We choose not to report the results for families with 17 or more children since, due to the very small sample size (five families with 17 children and two with more), the results are very imprecise.

 9. To avoid sample selection on the dependent variable, we here restricted the sample to women born prior to 1790. Separate dummies of birth were recorded for women born 1658–1689, 1690–1714, 1715–1739, 1740–1764 and 1765–1789.

10.  We also ran the same regressions comparing freehold farmers to tenants and crofters, with similar results. Even though the coefficient of the freehold farmer dummy variable had the anticipated sign, it was fairly small and not statistically significant. Since this suggests that the socio-economic classification is uninformative, we opt not to report these results.