“In this world, nothing can be said to be certain, except death and taxes.” Benjamin Franklin’s letter made this expression famous, but the phrase pre-dates him. Death is certain for any individual, but death statistics are not that certain, robust and reliable. So far, around 4,12,500 people have died because of the Wuhan virus in India and there are those who question this number. Modelling by The Economist is an instance, with numbers till 10 May 2021. “We estimate that, by May 10, there was a 95% probability that the pandemic had brought about between 2.4m and 7.1m excess deaths in Asia (official Covid-19 deaths: 0.6m), 1.5m-1.8m deaths in Latin America and the Caribbean (v 0.6m), 0-2.1m deaths in Africa (v 0.1m), 1.5m-1.6m deaths in Europe (v 1.0m) and 0.6-0.7m deaths in America and Canada (v 0.6m) … On the basis of the model, it would appear that around 1m people may have died of Covid-19 in India so far this year.” Around 4,12,500 is a current figure. On May 10, the figure was around 2,50,000. Therefore, The Economist’s model suggested the actual number of deaths in India was four times the official figure. Note that (a) this is a stochastic model, with a 95% probability, and (b) under-reporting of deaths, as per the model, is not sui generis to India.
The memory of another model is fresh, that of Goldman Sachs predicting the outcome of the Euro Cup 2020. As per that model, Belgium was going to win, beating Italy in the final, with England eliminated in the quarter-final. Memories of two other models are buried deep in our subconscious now. Between 18 March and 21 March 2020, Dr Ramanan Laxminarayan, Director of the Center for Disease Dynamics, Economics and Policy, granted interviews to all and sundry and projected that India would have 700 to 800 million infections and 2 to 2.5 million dead. Towards the end of March 2020, a high-ranking government official told us there would be no new cases beyond 16 May 2020. Both assertions were based on models and there were equations and graphs to back up the assertions. We are inordinately impressed with models, graphs and equations. This reminds me of Catherine the Great and the Diderot-Euler anecdote. Briefly, atheist Diderot (who knew no maths) pestered Catherine and she roped in Euler, the mathematician. Euler went up to Diderot, mouthed an equation that was gibberish and said, “Therefore, God exists.” Diderot took to his heels and fled the court. Any model is based on assumptions, conditioned by data availability and is probabilistic. A modeller should say, with 95% (or whatever number) probability, the range of deaths in India is between X and Y. True to the art, a modeller should always do interval estimation, never point estimation. But interval estimation doesn’t impress. It doesn’t suggest precision.
In discussing models, we shouldn’t muddle two issues. (a) Are deaths registered? (b) Are deaths correctly attributed to Covid? The Registration of Births and Deaths Act, 1969, makes death registration mandatory, with responsibilities fixed and penalties for violation. But that doesn’t mean all deaths are registered, especially in rural India, though numbers have increased over time. Even in rural India, people rightly perceive benefits associated with death registration, such as in inheritance matters. There is a sample registration system (SRS) to validate whether the civil registration system registers deaths accurately. The last SRS numbers are for 2019 and they tell us not all deaths are registered. States that stand out in non-registration are J&K (66.7%), UP (63.3%), Bihar (51.6%), Jharkhand (58.8%), Arunachal Pradesh (38.6%), Assam (74%), Nagaland (30%) and Manipur (21.4%). If all deaths are not registered, it stands to reason all deaths due to the pandemic will also not be registered. This isn’t a problem sui generis to India either. For upper middle-income countries, the World Bank tells us 75.97% of deaths are registered. At 92.7% in 2019, India is far ahead of this average and there have been extremely sharp improvements over time, perhaps even improvements between 2019 and 2021. If someone tells me deaths are not registered in India, the (a) proposition, that is certainly possible, but cannot logically be on a large-scale. While the law requires mandatory registration within 21 days, penalties for registration beyond this timeline are minor. I think it is plausible that during Covid and lockdown, people postponed registration. The updating of death numbers, as has been done by some states, may be explained by this factor, not deliberate fudging.
The (b) proposition is different and we have several problems with medical certification of cause of death (MCCD). Roughly 20% of deaths registered are covered by MCCD. Even in those cases, it isn’t guaranteed that the “cause” has been reported adequately. Even if the cause is generally reported adequately, it doesn’t follow that one has pinpointed Covid as the cause adequately. There are anomalies and inconsistencies in defining what a Covid death is. In West Bengal, in 2019, around 14% of deaths were medically certified. That cannot become 100% overnight simply because there was a pandemic. In principle, the pandemic will have differentially affected the cause of death. For example, fewer deaths from road accidents. But deaths from other ailments would have increased, since Covid has meant relative neglect of other diseases. Policy formulation requires good data, apart from uncertainty and time lags. Data is generated from below and aggregated upwards. The pandemic not only flags weaknesses in health infrastructure (both physical and social), particularly in some states, but also data issues.
Bibek Debroy
Chairman, Economic Advisory Council to the PM
(Tweets @bibekdebroy)
(Views are personal)