Feature: To rubbish WHO’s 2020 COVID death estimates, India used flawed data, analysis

 

  • On May 5, 2022, the WHO estimated that 100.51 lakh people died in India in 2020. Of this, it estimated 8.30 lakh were COVID-linked deaths.
  • In its May 5 press release rejecting the WHO figures, the Union health ministry claimed that 81.20 lakh people died in 2020 – even lower than the number estimated to have died in 2019.
  • But the health ministry was wrong by a long margin and by its own findings in a national report released earlier this year: the National Family Health Survey.

The Indian government has used a flawed dataset to rubbish the WHO’s 2020 COVID-19 death estimates, according to the fine print of official death data and a recently released government survey analysed by The Reporters’ Collective.

On May 5, 2022, the WHO estimated that 100.51 lakh people died in India in 2020. Of this, it estimated 8.30 lakh were COVID-linked deaths.

The Union government, which has often claimed to have done a splendid job to soften the impact of the COVID-19 epidemic in India, rubbished the WHO’s numbers.

In its May 5 press release rejecting the WHO figures, the Union health ministry said, “Registration of births and deaths in India is extremely robust and is governed by decades-old statutory legal framework i.e. Births & Deaths Registration Act, 1969.”

It claimed that 99.9% of all deaths that occurred in India in 2020 were recorded and that a total of 81.20 lakh people died that year – even lower than the number of people estimated to have died in the non-pandemic year of 2019.

Ergo, there was no way that 8.30 lakh excess deaths could have occurred due to COVID-19, the ministry’s statement indicated. It used these statistics to dismiss the WHO’s projection as untenable.

But the health ministry was wrong by a long margin and by its own findings in a national report released earlier this year: the National Family Health Survey (NFHS), which offers another official estimate of the death registration rate.

The report of the fifth NFHS – the latest – for 2019-21 shows that an average of 70.80% of deaths that occurred during the years 2016 to 2020 were registered. It was the first NFHS report to include data on levels of death registration.

Going by the NFHS death registration rate and the actual number of deaths recorded in the country’s Civil Registration System (CRS), a total of 114.07 lakh people can be estimated to have died in 2020.

This puts the WHO estimate of total deaths near the Indian government’s own data and punches holes in the government’s defence.

The 99.9% mystery

So where did the Indian government get its statistics from while rebutting the WHO? Answer: From two government data gathering and estimating exercises.

One of these sources is the CRS, which records the total number of people whose deaths are recorded in India’s villages, towns and cities. Not all deaths are recorded every year. Those deaths that do get recorded are enumerated in the CRS.

The total number of people dying each year is estimated using the results of an annual sample survey called the Sample Registration Survey (SRS).

In this survey, officials go to a select number of households to ask how many people in a family have died in the year. Based on this, the Office of the Registrar General calculates the death rate (the number of deaths per 1,000 people) in each state and then for the whole country. Remember, this is an estimate.

The government multiplies this estimated death rate with the population of the country to reach the total number of people estimated to have died in that particular year.

The government did this for 2020 also and estimated that the total number of deaths in the country was only 81.20 lakh while rebutting WHO (see table below). This official estimate of total deaths in the pandemic year of 2020 is lower than the total estimated deaths in 2019, when there was no pandemic.

These figures run contrary to all scientific modelling on COVID-related excess deaths published thus far.

The CRS 2020 report was the first one in at least a decade to not include key statistics like the number of estimated deaths or the level of registration. The Union health ministry published its claim of 99.9% registration of deaths in a press release instead.

With 81.15 lakh registered deaths according to the CRS, the ministry then went on to state that 81.2 lakh were estimated to have died in 2020. The SRS data, released on May 25, 2022, was in line with this claim.

Made with Flourish

A close look at the SRS data reveals that the estimate of the total number of deaths that the government is relying on is highly unreliable.

Where did it go wrong?

The SRS death rate estimates are deeply flawed. If we were to go by the SRS estimate, in 20 out of 36 states and Union territories, there were more deaths registered than people who are estimated to have died. This is impossible.

Consider the case of Chandigarh. According to the SRS, the number of death registrations was nearly four times (394.82%) the number of people estimated to have died in 2020. In Delhi, the registered deaths were nearly double (196.42%) the number of people estimated to have died in 2020. And so on.

Government statisticians do warn that the SRS can get the death estimates wrong for small states and Union territories. “Confidence limits of death rates and infant mortality rate for small states and union territories have not been presented due to small sample size and very large variations between the upper and lower tolerance limits,” according to the SRS report’s authors.

In other words, we can’t trust the estimates for states and UTs with smaller populations.

However, the government’s flaw is visible even in the case of many large states with much higher populations as well. For example, in Tamil Nadu, the number of recorded deaths was 148.14% of the deaths estimated through SRS. In Andhra Pradesh, the registered deaths were 137.56% of the estimated deaths.

The flaw in the SRS

The Reporters’ Collective spoke to a top official in the Office of the Registrar General, which conducts the sample surveys to estimate death rates (SRS) and maintains the country’s death registry (the CRS).

“Registered deaths being higher than estimated ones is not unusual,” the official, who wished to remain unnamed, said. “SRS is a sample-based survey, unlike CRS.”

He provided one reason to explain how the number of registered deaths in a state could be higher than the number of estimated deaths: migration.

“If somebody goes to Delhi for treatment and dies there, the death will be registered there, which will be reflected in the CRS data,” he explained. “The SRS takes only usual residents into account so SRS will cover the death in the place of residence.”

Put another way, in states where many people migrate out, the sample survey might capture their deaths in the states of origin – while the CRS will capture their deaths in the states to which they migrated.

This would imply that states with high inward migration will have more registered deaths than estimated deaths, and vice-versa.

This could possibly explain the death registration being higher in Delhi or Chandigarh, where people from a lot of states migrate to work.

But then how could 20 of 36 states and Uts record higher registrations of death than estimated deaths, that too by such high margins?

The Reporters’ Collective sent detailed questions to the Office of the Registrar General but didn’t receive a response at the time this report was published.

Experts debunked the official excuse for the flaw in the data.

“‘Out-of-state’ registrations are clearly an issue in Delhi and in Chandigarh, where many people come for treatment or work, but can’t explain why, say, Andhra Pradesh or Karnataka sees so many death registrations,” said Murad Banaji, a mathematician at Middlesex University, London, who has been closely following mortality in the pandemic.

“There is no substantial evidence showing that the excess levels of registration you see in some states is because of migration.”

“If anyone argues that the excess registrations are ‘out-of-state’ registrations, then the onus is on them to provide data which backs this up,” he added. “You can’t craft an argument against a claim when they have shown you no data.”

“You can expect some deaths of Uttar Pradesh to go to Delhi but you can’t expect that to happen in all the states with excess registration – they don’t fall within that pattern,” said T. Sundararaman, a public health expert and a former executive director of the Union health ministry’s National Health Systems Resource Centre. “Excess registration is happening because SRS death rates are an underestimate.”

“Yes, migration might reflect to a degree in this discrepancy,” said Dipa Sinha, assistant professor (economics) at Ambedkar University, Delhi. “But most migrants are of the age set where you will not see high mortality. And you would not see such high levels of differences across so many states.”

Banaji and others have also published studies saying as much.

Next, the SRS’s survey design and sampling method are revised every 10 years after the new population Census report comes out. This was last done in 2014.

Made with Flourish

The discrepancy between the number of registered deaths and the number of estimated deaths by SRS in states with more than 100% death registration has only widened after 2014, the data shows.

Correcting for the discrepancy

Banaji and his collaborator Ashish Gupta, the David E. Bell Fellow at Harvard University, have adjusted for this discrepancy in the SRS death estimates in their calculations. When dealing with a state where the actual number of death registrations was higher than the death toll as per SRS numbers, they used the actual number of registered deaths as the number of those who died in such states in a year.

They do so following the government’s practice of reconciling the CRS data with the SRS estimates.

Whenever the government finds the number of registered deaths to be greater than the number of estimated deaths, it says the level of registration is 100% – acknowledging that the death registration data can’t be disputed by the lower death estimate. Oddly, however, when calculating the total number of deaths, it still sticks to the lower figures of estimated deaths.

For example, in 2018, the government found the number of registered deaths in Andhra Pradesh to be 3.76 lakh against the SRS estimate of 3.53 lakh deaths (6.42% higher). So it said the registration was 100%. But to count the total number of deaths in India, the government used Andhra Pradesh’s lower estimated death count of 3.53 lakh!


CRS 2018 assumes 100% registration in the highlighted states (yellow highlight, table above). But the number of estimated deaths is lower than the number of registered ones (yellow highlight, table below). Source: CRS 2018


Banaji and Gupta account for this in their research by taking the number of registered deaths as the true count of how many people died that year in the state.

The Reporters’ Collective replicated their method to calculate the total annual death toll for 2020 and the previous 9 years. For 2020, the death toll would be 91.8 lakh, which would mean the death registration rate through CRS is 88.4% and not 99.9% as the government claims.

Even these numbers, arrived at after partly correcting the data in 20 states and UTs, seem to be an underestimate compared to the NFHS’s estimated death registration rate.

Why rely on NFHS estimates?

If we were to discard the calculations of independent demographers and experts, we could revert to the government’s own NFHS estimates of the annual death toll.

Unlike the SRS methodology, NFHS enumerators ask each surveyed household for data on both the number of people in the family that died in a year and what number of these deaths were registered.

“The figure of 99.9% death registration is based on comparison of death registrations and the deaths estimated through the sample registration system,” K.S. James, the director of the Union government’s International Institute of Population Sciences, which conducts the NFHS, explained.

“The accuracy of death rates given through SRS also varies and can have underestimates. It depends on the coverage of the SRS sample. That problem is there since it’s a sample.”

He also said that the NFHS collected data in two phases over 2019-2021 for the period 2016-2020. The death registration rate calculated on average would therefore reflect differences that manifested over this period.

In 2020, large states like Uttar Pradesh and Telangana officially admitted that their death registrations were hampered by COVID-19, leading to underreporting.

The NFHS death registration estimates were lower for 31 of 36 states and UTs in 2020 when compared to the registration rate arrived at through the SRS and the CRS. They also remain consistently lower for at least 19 states for the four previous years.

In sum, the NFHS survey results placed a big question mark over the Indian government’s claims of having handled the COVID-19 epidemic well and blew holes in its argument against the WHO’s estimates of COVID-linked deaths. So instead, the government decided to go with the unreliable SRS estimates to shore up its case.

Shreegireesh Jalihal and Tapasya are members of the Reporters’ Collective.

%d bloggers like this: