Stature and sibship: historical evidence

Abstract

This paper examines historical evidence for a quality–quantity trade-off between sibship size and height as an indicator of health. The existing literature has focused more on education than on health and it has produced mixed results. Historical evidence is limited by the lack of household-level data with which to link an individual’s height with his or her childhood circumstances. Nevertheless a few recent studies have shed light on this issue. Evidence for children in interwar Britain and for soldiers born in the 1890s who enlisted in the British army at the time of World War I is reviewed in detail. Both studies support the idea of a significant trade-off, partly due to income dilution and partly because, in these settings, large families were a conduit for infection. Evidence from country-level time series is consistent with this view. The fertility decline that began in the late nineteenth century made a modest but nevertheless significant contribution to the overall increase in heights during the following half-century.

Keywords:

• Health and height
• family size
• quality–quantity trade-off

Acknowledgements

I am grateful for comments from participants at the Third CEPR Economic History Symposium, Oslo and for detailed comments from Stefan Öberg and two anonymous referees.

Funding

I am grateful for funding from the Australian Research Council Discovery Project [DP140103603].

Notes

1. Two qualifications are in order. The first is that the greater the number of births that a family chooses, the greater the likelihood of having a multiple birth; therefore the probability of a twin birth depends on sibship size. One solution is to utilise twin births occurring only at a particular parity such as first birth. The other point is that twin births are rare (about 1 in 80) and they may have a limited impact on final sibship size as families adjust subsequent fertility. So the overall leverage exerted by twins on sibship size may be slight.

2. Although not considered in detail here, the issue of timing may also be important. At what point in the child’s lifetime is sibship size observed? Do younger siblings matter more than older siblings? And at what point do we observe the measure of quality? While it might be possible to look at what happened over time as family size increases, few studies have done this.

3. Although widely cited, this result presents something of a puzzle: if the underlying model is one of resource constraints, then family size should matter whereas birth order may or may not matter depending on whether or not scarce resources are unequally allocated between siblings. But it is hard to understand why the allocation between siblings should matter while the overall resource constraint does not.

4. At first sight this contrasts with the well-known finding of Clark (2005) that, in pre-industrial times, family size was positively correlated with wealth and human capital. However this can be interpreted as a comparison between points A and B’ in Figure 1, where the shift of the budget line from Z₁ to Z₂ is interpreted as an increase in lifetime income. Hence it may not be inconsistent with a quality–quantity trade-off if differences in income and the endogeneity of tastes are taken into account.

5. Earlier studies of height at age seven in the 1946 cohort include Goldstein (1971) and Fogelman (1975), both of which found some evidence of a negative effect for the number of siblings.

6. These children range in age from 2 to 14; they are fairly evenly distributed across the age range, with 33% aged 2–5, 37% aged 6–9 and 30% aged 10–14.

7. It is worth noting that the R² in the regressions is low even though the heights are family averages. According to a Finnish study, genetic factors account for about 80% of the variance in height across individuals (Silventoinen, Kaprio, Lahelma, & Koskenvuo, 2000). See also McEvoy and Visscher (2009) for a survey of genetic influences on height.

8. This is despite the fact that, as income per capita is not adjusted by an equivalence scale, this could impart some positive bias to the coefficient on the number of children.

9. It is important to recognise that birth order and sibship size are naturally correlated – one cannot be birth order five in a family of three. Birth order is a ranking and its ‘true’ effect is the hypothetical effect of moving up the birth order by one while holding family size constant. In order to identify the true birth order effect (when looking across families) one has to somehow purge the birth order variable of the sibship size effect. One way of doing this is to use instead the deviation of birth order from the sibship average birth order (which is (N+1)/2) (see Hatton and Martin, 2010a, p. 162). For WWI servicemen this index did not yield a significant coefficient. One reason may be that we have only an estimate of completed sibship size (which is relevant for final heights); alternatively there may simply be too much heterogeneity when estimating across families as compared with estimating within families.

10. The prediction is based on a regression coefficient of the number of children on mother’s age for mothers aged 20–40 of 0.191 (‘t’ = 20.4). To calculate predicted sibship, e.g. for a mother aged 30 we add (40–30)*0.191 = 1.91 additional children to the observed number. For families where the mother is over 40 or there is no mother present, no further children are added. This gives an average predicted sibship of 5.19 as compared with the unadjusted average of 4.16.

11. If the number of rooms is replaced by a dummy variable for the number of persons greater than the number of rooms, this takes a negative coefficient with a ‘t’ value of 2.33. However, if both variables are included the latter becomes insignificant.

12. Studies that directly link reductions in infant mortality to sanitary reforms include Cutler and Miller (2005) for the US, Macassa et al. (2006) for Sweden, and Newell and Gazeley (2012) for the UK.

13. Several studies point to the importance of the supply of protein, especially milk, as proxied by the density of cattle in the locality (see Baten, 2009). Adding to the column (3) regression the share employed as cattlemen as a proxy for access to protein gave a positive but insignificant coefficient with very little effect on the coefficient on sibship size. This probably reflects the fact that by 1901 the railways had provided good access to farm products for urban populations, and milk was far less adulterated than 30 years earlier.

14. In Britain, from 1910 to 1950 the heights of schoolboys at ages 6, 8, 10 and 12 increased by between 1.7 and 2.3 cm per decade (Hatton, 2011, p. 963). By contrast the heights of adult males increase by about 1 cm per decade, see Table 7 below.

15. The clearest evidence is for the age at menarche in girls. In Europe this declined from an average age of 14 in 1900 to 12.8 in 1947 (Wyshak & Frisch, 1982); there is also evidence that earlier menarche is associated with smaller sibship size (Morris, Jones, Schoemaker, Ashworth, & Swerdlow, 2010).

16. The turning point implied by the estimated quadratic is an infant mortality rate of 21.6%, which is within the range observed in the data.