Reason to be sceptical about trends in adult height in India

A couple of years ago, I read this 2021 article by Krishna Kumar Choudhary, Sayan Das, and Prachinkumar Ghodajkar (Jawaharlal Nehru University), published in the journal PLoS ONE (open access). I've been holding off blogging about it, in the hopes that I could get one of my past PhD students interested in exploring this data and testing the claims further, but no one seems too interested (or, at least, they're too busy doing other exciting things). So, here we go.

Choudhary et al. use data across multiple waves of the Indian National Family Health Survey, and track trends in adult height in Indian provinces over the period from 1998-99 (NFHS-II) to 2015-16 (NFHS-IV). They found that:

Between NFHS-III and NFHS-IV, the average height of women in the age group of 15–25 showed a decline by 0.12 cm [95% CI, -0.24 to 0.00, p-0.051] while in the 26–50 years age strata it demonstrated significant improvement in the mean height by 0.13 cm [95% CI, 0.02 to 0.023, p-0.015]. However, Between NFHS III and IV, the average height of women in the poorest wealth index category registered a significant decline [-0.57cm, 95% CI, -0.76 to -0.37, p-0.000]. Between NFHS III and IV, the average height of Scheduled Tribe (ST) women in the age group of 15–25 years also exhibited a significant decline by 0.42 cm, [95% CI, -0.73 to -0.12, p-0.007]. Among men, between the two surveys, both the age groups of 15–25 years and 26–50 years showed significant decline in average height: 1.10 cm [95% CI, -1.31 to -.099 cm, p-0.00] and 0.86 cm [95% CI, -1.03 to -0.69, p-0.000], respectively.

You read that right. According to Choudhary et al., people in India are shorter in 2015-16 than they were in 2005-06 (NFHS-III). The distribution of mean height by age for those two surveys is given in Figure 4 in the paper:

Notice that, within every age group, the mean height is lower in 2015-16 than in 2005-06. However, here is where I have severe doubts about this analysis. The sample of Indian men in 2015-16 is (for the most part) the same as the sample of men ten years younger in 2005-06. So, if you compare a given age group's mean height in 2015-16, it shouldn't be too much different from the mean height of the age-group ten years younger in 2005-06. And yet, that doesn't appear to be true for almost any comparison in Figure 4. Look at the mean height for any age on the bold line in the figure, move to the right by ten years, and you will never intersect with the dashed line.

So, one of three things is going on here. Either, Indian men are shrinking, there are measurement errors that are changing over time, or there are compositional changes in the sample that explain the differences. It seems unlikely that people are genuinely shrinking. So, that leaves the other two explanations.

Although the NFHS is a 'nationally representative survey', there are serious issues with the survey (as documented by Sylvia Karpagam here). That suggests that measurement error might be at play. However, it would have to be measurement error that occurs in a way that heights were either systematically under-reported in NFHS-IV, systematically over-reported in NFHS-III, or both. That does seem a little unlikely.

What about compositional changes? There may be differences in survey coverage (see here), especially between women in NFHS-II (which only included ever-married women) and NFHS-III (which included both ever-married women and never-married women). However, it is less clear that the changes affected men in the sample. On the other hand, this bit from the Choudhary et al. caught my attention:

The samples drawn for analysis of women’s height were 83876 out of 90303 from NFHS-II, 121728 out of 138592 from NFHS-III, and 700602 out of 749344 from NFHS-IV. For men’s height, sample of 66468 out of 74396 from NFHS- III and 105783 out of 126543 from NFHS-IV were drawn.

Notice that the sample for women increases nearly six-fold between NFHS-III and NFHS-IV, but the sample for men increases only by about 60 percent. That might be accurate, but it strikes me as odd, unless men are only surveyed in a subset of households, and the proportional subset that were selected was different (and much smaller) in NFHS-IV than in NFHS-III. That could cause a change in the composition of the survey sample, and might explain the results for men (less so for women). Anyway, there is reason to doubt these results, and it might be an interesting project for a suitably motivated Honours or Masters student to follow up on.