Height, health, and development

Results

Preliminary inspection of the data shows that South Asian women are markedly shorter than other women; the population-weighted average of the three countries (Bangladesh, India, and Nepal) is 151.2 cm compared with 155.0 cm for Latin America and the Caribbean (14 surveys from eight countries: Bolivia, Brazil, Colombia, Dominican Republic, Guatemala, Haiti, Nicaragua, and Peru), 156.9 cm for five Central Asian countries (Armenia, Kazakhstan, Kyrgyz Republic, Turkey, and Uzbekistan), and 157.8 cm for Africa (29 surveys from 27 countries). Although Africans are taller overall, there is a good deal of variation across countries. Dispersion of height, as measured by the SD calculated for each country and then population weighted over countries within a region, is also lowest in South Asia (5.80 cm) followed by Central Asia (5.86 cm), Africa (6.58 cm), and Latin America and the Caribbean (6.81 cm). The regional (and indeed country) variation in dispersion is larger than the regional and country variation in heights, so that the lower SD in South Asia is not simply “explained” by its lower mean; the cross-survey regression of the logarithm of the SD of heights on the mean height has a coefficient of >1.5. Within each region, the dispersion in heights mostly reflects within-country dispersion; for example, in Africa, the overall variance in heights is 43.9 cm², of which fully 40.4 cm² is the within-country component. This dominance of within- over between-country variation has previously been noted for children's heights in ref. 12.

The lack of any obvious pattern in these means and variances presents a challenge to the view that population mean heights are predominately determined by economic and sanitary environments. Africa is the poorest of the regions and has the highest disease burden yet is the tallest of the regions.

Fig. 1 displays the height data by region and by date of birth within region. It also includes average heights for women for a group of European countries and the U.S. taken from ref. 11. In Fig. 1, I have calculated average heights by date of birth for each country and then averaged the averages over country within regions without weighting by country population or sample size. Here I am interested in the experience of each country as an example of all possible histories, not in trying to estimate the average heights by date of birth of all women in a region. For each country grouping, Fig. 1 also shows its per-capita gross domestic product (GDP) in 1970 in 1996 international dollars, taken from the Penn World Table. When country years (e.g., Armenia before 1992) are not available in the Penn World Table, they are dropped, and means are computed over all available countries.

With the exception of Africa, where women are tall relative to their national incomes, there is at least a gross interregional correspondence between height and income. Over time, as real incomes have grown, heights have grown, too. In the richer group of northern countries, heights grew until about the birth cohort of 1970, but there has been little growth since. In the poorer northern group, growth in heights has been steady from 1950 through to 1980. In the other regions, the birth cohort of 1980 was taller than the birth cohort of 1950, with the exception of Africa. African heights have been falling since the mid-1960s and appear to have continued to fall in (fully mature) cohorts that have been born since 1980 (not shown in Fig. 1). The brief turndown in heights in South Asia up to 1980 does not indicate any change in trend but rather that women in the region do not attain their full adult height until their early 20s, so that the birth cohorts immediately before 1980 show artificially low average heights. Real incomes in the three South Asian countries have been growing rapidly relative to those in the African countries included in the DHS. Real per-capita incomes in Africa have been falling since 1980, and by 2000, the simple average of per-capita incomes in Bangladesh, India, and Pakistan was 10% higher than the simple average over the African countries.

Figs. 2 and 3 (which exclude the rich countries) show the difficulty of linking population heights to population differences in child mortality rates and to income. These figures plot for each country a circle with diameter proportional to the size of the country's population. In both cases, the y axis shows the average height of women aged 25–50 from the DHS. In Fig. 2, the x axis shows the logarithm of real per-capita GDP in the nearest year to the average date of birth of the women in the survey, typically a year in the 1960s. In Fig. 3, the x axis shows child mortality rate per thousand, again in the average year of birth. (The average year of birth is also used to calculate the population that determines the circle size.) Because child mortality rates are available only for 1960, 1970, and 1980, I have interpolated between them to match the average year of birth. Each region is shown in a different color and, because of data limitations for the “new” countries of Central Asia, Turkey is the only representative of the region in the figures.

If height is determined in childhood by the availability of food, and if the availability of food is well proxied by income per capita, Fig. 2 should show a positive slope, as is the case across Europe and the U.S. (see ref. 11). Instead, the slope is negative; the (unweighted) regression corresponding to Fig. 2 has a slope of −1.63 with a t value of −2.1, so that a doubling of income per capita is associated with a reduction of 1.63 cm in average women's adult height. If height is determined not by food availability but by the disease environment in childhood, Fig. 3 should show a negative slope, and once again, this prediction was verified for Europe and the U.S. in ref. 11. But the slope in Fig. 3 is small and positive; the regression slope is 0.012 with a t value of 1.7. Fig. 3 can be redrawn with infant, in place of child, mortality, and although the graph is somewhat different, the slope is virtually unchanged at 0.013 with a t value of 1.04.

The global failure of either income or the disease environment to predict mean adult height comes almost entirely from Africa, particularly the contrast between African countries on the one hand and the rest of the world on the other. Although most of the countries in the graphs are African, both Figs. 2 and 3 show that the relationships are mostly as expected in Latin America and the Caribbean, especially if Haiti is excluded, as well in the combined data of Latin America, the Caribbean, South Asia, and Turkey. But African women are tall, even in the poorest and most disease-ridden countries (top right of Fig. 3). Although these women may have been well nourished in childhood, they were so despite some of the world's lowest levels of national income per head.

Figs. 2 and 3 use only one observation per country (or survey) and make no use of the variation of heights over dates of birth and the corresponding variation in incomes and childhood disease. Fig. 4 corresponds to Fig. 2 but now shows all individual observations for each country, so that each point shown is the average height of women in a specific birth cohort and country plotted against the logarithm of real national income in the year of birth. I have excluded average heights when the number of women in the average is <10 and, when I run regressions, I weight by the square root of the number of observations in the average to allow for the varying degrees of precision in these estimates.

Fig. 4 shows that, to a remarkable extent, the developing countries of the world are regionally separated in the space of income and mean height. And, although it is possible to see a weak association between height and income across the regions other than Africa, there appears to be no association either within Africa or between Africa and the other regions. Note too the extent to which Haiti looks like an African, not a Caribbean or Latin American, country.

The rich countries are excluded from Fig. 4, although it is clear what would happen if they were included. There would be another “block” of data at the top right of Fig. 4, which, if we forget what we are looking at, would “improve” the all-country relationship between height and income. Yet it would do nothing to establish such a relationship across or within the countries currently shown.

Table 1 uses the developing country data to estimate a set of regressions in which average height is related to income and child mortality rate. When the columns “region” and “country” show a star (*), the corresponding regression contains a set of regional dummies (for Africa, Central Asia, South Asia, and Latin America) or a set of country dummies, one for each country. With regional dummies, the regressions use only variation within countries and regions; with country dummies, the regressions use only variation within countries. In the first row, which corresponds to fitting a regression line through Fig. 4, height is negatively and significantly related to income. This negative effect is smaller, but still significant, if regional dummies are included in the regression, and becomes positive and insignificant when country dummies are included. So the failure for richer countries and regions to be taller than poorer countries and regions also holds within countries over time. When all of these countries are pooled together, and we look only within countries, there is once again no significantly positive relationship between height and economic success. These inconsistent patterns are not much changed when I add the child mortality rate to the regressions. Child mortality is normally expected to decrease height, because of the predominance of the scarring effects of high disease environments in childhood, but the effect is estimated to be positive when all data are pooled and positive when we look only within countries in the bottom row. The effect is insignificant and negative when only regional dummies are included. (Note that, because the child mortality data are interpolated, these regressions exaggerate the true amount of information and will exaggerate the absolute size of the t values. Regressions that retain only the noninterpolated values of child mortality rates for 1960, 1970, 1980, and 1990 show very similar patterns with smaller t values, which are now underestimated, because I am discarding genuine information on incomes.)

If Africa is excluded, the regression of height on income has a positive slope, as is to be expected from Fig. 4. But the estimate changes sign when regional dummies are included, and changes back to a significant positive effect when country dummies are included. In the non-African regions of the world, people get taller with higher incomes, although the cross-country effect within regions does not conform to this within-country effect. For the countries of Africa alone (Table 1 bottom), the cross-country effect of income is positive (richer countries are taller), but the within-country effect is (marginally) significantly negative.

Once again, including child mortality does nothing to clarify the pattern, although, as before, the non-African patterns are slightly more palatable, with child mortality negatively associated with heights in all three regressions. But even in those cases, the income effects are either insignificant or of the wrong sign, or both. In Africa, child mortality is positively associated with heights; conditional on income, which has the expected positive effect, children in countries with the highest disease burden turned into the tallest adult women. These regressions can be run with infant mortality in place of child mortality, and the results are virtually identical and are not shown.

Whatever determines adult height in these countries, it is neither income per head in childhood nor the disease burden in childhood as measured by either infant or child mortality. Such relationships as exist in the data are inconsistent within and between countries, or across regions of the world, so that even if further research were to make sense of one or the other of those regressions, we would still not have a story linking height to income or disease that is consistent within and across countries. One possibility, suggested by the last line of Table 1, is that in Africa, where child mortality is very high, selection predominates over scarring, so that both income and child mortality are positively associated with height. In rich countries, at much lower levels of mortality, scarring predominates. Yet this explanation does nothing to explain the perverse or insignificant effects of income in the non-African countries in Table 1.