By Atanu Biswas
As Covid-19 emerged, with tons of readily available data, academic journals and media were flooded with lots of conflicting predictions about Covid-19 numbers and the beginning times of different waves. Most of these eventually turned out to be completely wrong. There’s a new trend now – estimating excess death from Covid, which is primarily unverifiable, and thus even the odd estimates can hardly be proven wrong. Fun indeed!
‘Excess death’, however, is not an alien concept. When Covid ravaged Italy in 2020, one of my Italian acquaintances wrote me an email: “Many people die who have never been checked for Covid-19, so they do not go into the official counts.” Well, is that just Italy’s history? Possibly not.
There is reason to believe that many Covid-related deaths were not recorded as ‘Covid’ elsewhere – though not a fault of that authority. No country was prepared for the pandemic. Due to the initial scarcity of test kits, overwhelmed healthcare and of course social stigma, such underreporting is inevitable – especially if the disease in question is highly contagious in nature.
However, defining “excess deaths” is a difficult task. Leading magazines and top medical journals argued that those who had Covid but might have died on a similar time frame of other ailments should be excluded from the count. However, it is recommended that people who have died of preventable causes during the pandemic – because the overwhelming health infrastructure could not treat them – be attributed to Covid. Similarly, quarantine measures may have reduced deaths due to accidents and work-related injuries, and social distancing also led to a decrease in deaths due to influenza-like illnesses. These should also be kept in mind when estimating the Covid number.
The World Health Organization’s latest estimate of more than 4.7 million Covid deaths in India – almost 10 times higher than official records suggest – sparked controversy. However, there are other estimates. A study by the University of Chicago even estimated 6.3 million excess deaths in India during the pandemic up to August 2021. In fact, a March 2022 Lancet article estimated that 18.2 million died worldwide. The dynamic estimate made by The Economist magazine shows about 14.7-25.1 million global exaggerated deaths, which is 2.3-4 times the official figure so far. The U.S. Centers for Disease Control and Prevention also estimated more than 1 million excess deaths in America.
Such widely differing estimates are based on different public and private data and different types of models. Understandably, Covid data from different sources contain errors and biases of different size and direction. And to estimate something, in general, based on possible biased data of unknown nature, is subject to errors of an unknown quantity. Partial data is another serious problem. For example, if the Covid mortality rate in Maharashtra or Kerala were used to draw a pan-India picture, it would give a strongly biased estimate.
Data-based conclusions are not unique, we know. Interestingly, the above-mentioned Lancet article used six models to estimate expected mortality; final estimates were based on an ensemble of these models. Well, other models could not give other estimates? No one really knows how appropriate a model is in general. ‘Wisdom’ certainly plays an important role in choosing an ‘appropriate’ model.
While this may sound surprising, Covid data may not be necessary at all to estimate Covid’s excess death for any country. Mortality data for all causes in the last few years may be sufficient if they are available and credible. One can build an ‘appropriate’ time series model for the total number of deaths up to 2019, keeping in mind the trend and the possible disasters in the country in these years. If there had been no Covid, the expected number of deaths in 2020 and 2021 could be obtained from the fitted model. If we subtract this from the total death toll for all causes in 2020-21, it would give the estimated Covid figure. In fact, using Covid data for this purpose may even induce an error of unknown magnitude. However, deaths from all causes cannot be fully registered or released in a timely manner, or even be questionable in a country. With so many conflicting and vastly different estimates, finding Covid’s exact death toll can be like chasing an unknown in an unknown direction. And it would possibly remain intangible unless a credible large-scale survey is conducted across the country.
Yes, one can remember a seven decade old story. During the Bengal famine of the 1940s, Prasanta Chandra Mahalanobis, India’s ‘Plan Man’, conducted a large-scale sample survey of famine-ravaged villages for causal analysis and to assess the scale of the disaster and an estimate of the number of people affected. The study yielded very useful results, such as a quarter of the number of families (1.5 million people) who had owned rice land before the famine had either wholly or partly sold such land or had mortgaged it, and the economic situation of almost 4 million people was worsened during the famine. Well, only a credible study like it can bring the actual Covid story clear. However, it is not easy – Mahalanobis’ shoes are not yet full. Conflicting images of the Covid disaster would continue to create buzz.
The author is Professor of Statistics, Indian Statistical Institute, Kolkata.