Female Fertility and Longevity

TL;DR

This paper critically examines the Caerphilly cohort study's claim that childbearing shortens women's lifespan, revealing statistical flaws and showing that the effect disappears when outliers are removed, suggesting no real link.

Contribution

It identifies limitations of Poisson regression in analyzing fertility and longevity data, and demonstrates that the purported effect is not statistically robust.

Findings

Poisson regression is inappropriate for this data due to non-Poisson distribution.

Removing outliers eliminates the significant effect of fertility on lifespan.

Bi-linear regression shows a positive relationship between fertility and longevity.

Abstract

Does bearing children shorten a woman's life expectancy? Several demographic studies, historic and current, have found no such effect. But the Caerphilly cohort study is far the most prominent and frequently-cited, and it answers in the affirmative. Why has this study found an effect that others fail to see? Their analysis is based on Poisson regression, a statistical technique that is accurate only if the underlying data are Poisson distributed. But the distribution of the number of children born to women in the Caerphilly database departs strongly from Poisson at the high end. This makes the result overly sensitive to a handful of women with 15 children or more who lived before 1700. When these 5 women are removed from a database of more than 2,900, the Poisson regression no longer shows a significant result. Bi-linear regression relating life span to fertility and date of birth results in a positive coefficient for fertility.