Accounting for India’s ‘Missing’ Women

It’s been well over twenty years since Amartya Sen first coined the term ‘missing women’ to describe the highly-skewed sex ratios across a number of countries, including India. As of the latest census, India had a sex ratio of 940 women for every thousand men. Assuming an ideal sex ratio of 1057 women to thousand men (UN figures for developed regions), that gives us a ‘shortage’ of around 117 women, or roughly 73 million ‘missing’ women as of 2011.

But how do India’s missing women break down by age group? It was long assumed that the main cause of India’s missing women was either female foeticide or infanticide, or worse care in younger age-groups than male peers, leading to higher mortality rates for girls.

But a new paper published last December in the Economic and Political Weekly by Siwan Anderson and Debraj Ray throws entirely new light on that question. Their methodology is slightly different and they estimate ‘missing women’ per year, but it’s entirely consistent with the approach above. Here’s their conclusion:

We estimate that a total of more than two million women in India are missing in a given year. Our age decomposition of this total yields some striking findings. First, the majority of missing women, in India die in adulthood. Our estimates demonstrate that roughly 12% of missing women are found at birth, 25% die in childhood, 18% at the reproductive ages, and 45% die at older ages.

Further:

Only in Punjab are the majority of missing women to be found at birth. There are just two states in which the majority of missing women are either never born or die in childhood (i e, before age 15), and these are Haryana and Rajasthan. Moreover, the missing women in these three states add up to well under 15% of the total missing women in India.

Here’s a simplified example to show how they calculate the number of ‘missing’ women : Assume that in the west, four of every thousand women in the 30-35 year age group die every year, compared with five of every thousand men. That relationship – four female deaths to every five male deaths, should ideally hold here as well. If instead, India sees six female deaths to every five male deaths in the 30-35 year age group, that difference – two ‘extra’ female deaths a year – is the measure of missing women for that year (again, please don’t get confused between missing women at any point in time – 73 million in 2011 – and missing women per year which is 2 million).

You can perform a similar exercise for all age groups, as well as a separate calculation for the sex-ratio at birth, which is the figure that is most talked about in the ‘missing women’ debate.

The paper calculated the numbers for 2003. I’ve redone the numbers to calculate excess female deaths per thousand across age-groups for India, China, and a number of Indian states for 2010. Each bar shows excess mortality per 1000 women in that age-group only.

I deliberately didn’t go the final step and actually calculate the number of ‘missing women’ in 2010, because the data should ideally be averaged over several years before you do that (the paper does this). So I left it at excess deaths per thousand.

The final point, and I’ll quote from the paper again, is the following. It’s not exactly earth-shattering but I guess it’s worth emphasizing:

what is most clear from our exercise is that the plight of adult women in India is as serious a problem as that of young girls who were either never born or die prematurely in childhood.

Another key point to note is that the total number of missing women per year in absolute terms, is heavily dependent on the structure of the population i.e. how the female and male populations are distributed across different age-groups. I’ll leave you with a question: What does this research mean, in the context of the changing age structure of the Indian population? I’ll probably look at this topic in a future blog post.

Finally, I strongly urge you to read the paper (also linked here), and another, earlier paper by the same authors exploring the same question, but in an international context.

Notes:

My benchmark to calculate missing women is the UN data for what they call ‘more developed regions’. The data for India is from the Sample Registration System on the census website.  UN link here, and SRS link here.

For some reason my web hosting provider doesn’t allow me to upload excel files. If you want worksheets where I calculated the data, please email me at datastories.in@gmail.com

Please note that India, China, UP and Haryana are scaled in a different way to the other states (i.e. the y axis numbers change). That makes comparisons across states difficult I know, but it was the only way to fit in such a wide range of data.

Perhaps the biggest problem with my calculations is that it pertains to one year only – 2010 – when the numbers should ideally be averaged out over a few years at least.  Not doing that is probably why, for some age-groups for some states, I actually end up with a surplus, not a deficit. While this is plausible for states like Kerala, it’s less so for states like Haryana, and I can only attribute this to problems with the ‘noisiness’ of the data. In all such cases, I’ve simply put the excess deaths number to zero.

Charts made with the D3 javascript library.

April 21, 2013

6 responses to Accounting for India’s ‘Missing’ Women

  1. shankar said:

    Hi,
    It would be useful if you could make such a trend line for the US, UK, Japan and some of the Scandinavian countries that are considered the best in class.

    • Administrator said:

      that is implicit in each chart. excess mortality is calculated with reference to the standard of ‘developed regions’, which includes these countries collectively.

  2. pramod said:

    I read the paper a month or two ago, so my recollections may be inaccurate. With that disclaimer out of the way, I seem to remember that their methodology for computing missing adult women was the following.

    The mortality rate for adult men is higher than the mortality rate for adult women in the west. I assume this is because because men are more likely to be involved in high risk endeavors like crime, police and the military but I would appreciate more insight on why this might be true. So what they did was compute the “differential” mortality between men and women in the west broken down by age groups, and use this to compute what the adult population of women should have been in India.

    The big question for me was: how reliable is this? Is it really fair to assume that the differential mortality in the west is a good predictor for the same variable in India. I realize they correct for this to some extent by using Kerala as a benchmark in the end of the paper, but that seemed like it was done as an afterthought.

    I don’t work in the social sciences and I admit my knowledge of the field is cursory so I would appreciate comments on how valid their methodology really is from those more knowledgeable than me.

    • Administrator said:

      Well, I’m no expert either, but its worth reading their earlier paper written in 2010, to get some insight into the questions you are asking (linked above). Its a fair question about whether we can use the west as a benchmark for mortality here. In particular there can often be genetic differences, with respect to susceptibility to disease which might explain some of the disparity in female mortality between the West and other continents and countries such as Africa or Asia. But there is a major component of excess female mortality in INdia which is still explainable by social, economic and cultural disparity between the sexes and to that extent, a comparison is fair. To quote (from earlier paper):
      “In India, communicable,preventable diseases explain missing girls in childhood. Maternal mortality and injuries are important at the reproductive ages. Cardiovascular deaths are an overwhelmingly strong source of missing women at older ages in India and dominate all other sources of excess female
      mortality. Finally, congenital deaths at infancy, as well as Injuries, account for a suspiciously large total of excess female deaths in India. These excess deaths easily outnumber maternal mortality.”
      On higher male mortality in the west compared with women see this:
      http://www.demogr.mpg.de/Papers/Working/WP-1999-009.pdf
      From the above, smoking for instance, is an important cause of higher male mortality in the West.

  3. Aatish said:

    I’m a little confused about this analysis. If you are indeed comparing deaths per 1000 people in each age group to numbers in the West, wouldn’t this be strongly skewed by the fact that life expectancy is lower in India? In other words, showing a peak in the 70-75 age group for missing women might be slightly misleading, because it needn’t be a female specific effect – I would expect there to be a similar peak in 70-75 age group for ‘missing men’ simply because life expectancy is lower in India. I haven’t read the paper, so please correct me if I’m misunderstanding something basic here.

    • Administrator said:

      no, its an absolutely fair question and I should probably have dealt wih it in the main post. This problem is adjusted for by considering female mortality in the two regions (India and the West), not in absolute terms, but relative to male mortality. think of it this way. Suppose Indian women and men at age 75 die at rates of 15 and 20 respectively (per thousand). In the west assume they die at rates of 6 and 12 at the same age. The standard of comparison is 15/20 (for india) or 75%, vs 6/12 (for the West), or 50%. By using male mortality as the denominator for both regions, the authors adjust for overall adult mortality being higher at older ages. The point being made is that even after you adjust for lower life expectancy/higher adult mortality for India at later ages, there is an excess female mortality whose extent, in our example, is indicated by 75%/50%.
      The relevant part of the paper says as follows:”all groups exhibit excess female mortality at all ages. But relative excess female mortality [75%/50% in our example:ed] is generally higher at adult ages, sharply highest at reproductive ages, with another gradual peak at old age. Adult excess female mortality is not just a reflection of higher adult mortality.”