The Field Guide · No. 50
Excess deaths: what the count includes and why estimates differ
Excess deaths are the deaths that occurred minus the deaths expected without the crisis, so every figure is a model's estimate and only as good as its baseline.
Updated
Excess deaths are the deaths that occurred in a period minus the number expected for that period had the crisis not happened. The expected figure is never observed, so it has to be estimated, and every excess number carries that uncertainty. WHO's dataset page defines excess mortality as "the mortality above what would be expected based on the non-crisis mortality rate in the population of interest".
Excess is not a count of COVID-19 deaths. WHO's FAQ says the difference is assumed to include deaths caused directly by the virus, deaths caused indirectly when health systems are overwhelmed or people avoid care, and a subtraction for deaths that were averted, such as road deaths during lockdowns. Reported COVID deaths miss people who died without a test and countries that do not correctly record COVID as the cause of death. That is why WHO considers excess mortality a more objective and comparable measure.
The headline totals cover 2020 and 2021 together, and WHO has published two rounds. Its May 2022 release put the global figure at about 14.9 million, between 13.3 million and 16.6 million. The peer-reviewed paper behind those estimates, published online in Nature in December 2022, reports 14.83 million (13.23 million to 16.58 million) against 5.42 million COVID-19 deaths reported to WHO for the same period. WHO's revised estimates, posted on 19 May 2023 and calculated as of 5 April 2023, replace the 2022 round: about 14.86 million excess deaths (13.28 million to 16.63 million), 4.44 million of them in 2020 and 10.43 million in 2021. The revision also subtracts deaths from conflicts and natural disasters, using UN Population Division counts, so its total aims at the pandemic's toll alone. The three totals sit close together, but quote one with its date.
So read an excess figure as a modelled answer, and check three things before relying on it: which earlier years form the baseline, how the model fills gaps where a country has little data, and whether the figure comes with a range. WHO ties each country's range to its reporting capacity and the quality of its data, so the range is part of the answer.
How the expected baseline is built
Every source builds the expected line from earlier years, but the details differ. WHO's data story describes its baseline as "existing average mortality data from 2015 to 2019." The Nature paper models expected numbers from historic data, using 2015 to 2019 for most countries, with a negative binomial model that allows for variation beyond a plain Poisson model.
The UK Office for National Statistics describes its new method in an article dated February 2024. It fits a quasi-Poisson regression to death registrations, and each model uses five years of data with a one-year lag, so every week or month gets its own baseline. The method it replaces took a simple five-year average and removed all of 2020 from the baseline data. The new one removes only the weeks and months the pandemic substantially affected, and it accounts for changes in population size, ageing and mortality rates.
ONS defines the expected number as the deaths that would have been registered if mortality trends had stayed as they were in recent periods, with no extraordinary events such as the peak of the pandemic. Because it cannot be observed, it has to be estimated from the data that do exist.
Why two estimates can differ
WHO's FAQ names the sources of disagreement with other groups: the input data, how expected deaths are calculated, the statistical model, the variables used where data are thin, and the period or week assessed. A single modelling choice can move one country's figure a long way.
| Figure | Value | Range | Source and date |
|---|---|---|---|
| Excess deaths, WHO release | 14.9 million | 13.3 to 16.6 million | WHO news release, 5 May 2022 |
| Excess deaths, Nature paper | 14.83 million | 13.23 to 16.58 million | Msemburi et al., published online 14 December 2022 |
| Excess deaths, WHO revised estimates | 14.86 million | 13.28 to 16.63 million | WHO data download, 19 May 2023 |
| COVID-19 deaths reported to WHO | 5.42 million | Not given | Msemburi et al. |
WHO's dataset page says its estimates "may differ from, and should not be regarded as, the official national statistics of individual WHO Member States." A national statistics office and WHO can therefore publish different numbers for the same country and year.
The Nature paper's own estimates show the effect of one choice. For Sweden, the authors judged that a completeness adjustment to the raw 2019 mortality figure was unnecessary, and they switched the trend term from a spline to a linear one. The estimate rose from 11,300 (9,900 to 12,700) to 13,400 (11,700 to 15,200). WHO's May 2023 revision puts Sweden at about 12,200 (10,900 to 13,500), a third figure for the same country and years. For India, the 4.74 million estimate rests on subnational data from 17 of the country's 36 states and union territories, plus a proportionality assumption to reach a national total.
P-scores and negative excess
A P-score expresses the excess as a share of the expected deaths, which puts places of different size and age structure on a common scale. In the paper's example, 100 expected deaths and 140 actual deaths give 40 excess deaths and a P-score of 40%. The paper says the P-score "implicitly considers both the population size and the age structure." Globally, it reports P-scores of 7.97% for 2020 and 18.30% for 2021.
A negative excess means fewer people died than the baseline predicted. WHO's FAQ gives the usual reason: "For instance, during lockdowns, fewer deaths due to road traffic injuries and seasonal flu were observed due to the restricted movement of people." The Nature paper reports mortality deficits, or negative estimates, for certain months in several countries, most of them with high-quality reporting, and ties them to fewer deaths from other causes and milder flu seasons in 2020 and 2021.
Like any excess figure, a negative one is an estimate with a range around it.
What to remember
- Excess deaths are observed deaths minus expected deaths. The expected figure is a model, so every excess number inherits the model's choices.
- An excess count includes indirect deaths and subtracts averted ones, so it is not a tally of deaths from COVID-19.
- A P-score puts the excess over the expected count. 100 expected and 140 observed deaths give a P-score of 40%.
- Two estimates for the same place can differ through the baseline years, the model, the data gaps and the period assessed. Read the range before the headline.
From the record
Excess mortality quantifies the increase in mortality from all causes, including direct COVID-19 deaths, indirect COVID-19 deaths (for example, health-system overload) and strictly non-COVID-19 deaths (for example, those resulting from other health shocks such as violent conflict or disasters).
Asked often
Is an excess death the same as a COVID-19 death?
No. The count includes COVID deaths that were never tested or recorded, and deaths from other conditions that rose when health systems were overwhelmed or people avoided care. It also subtracts deaths that did not happen because of lockdowns and quieter flu seasons. WHO's FAQ says the difference is assumed to include these effects, which is why it is not a COVID-19 toll.
Why do two organisations publish different excess totals for the same years?
The baseline and the model differ. The choices that matter include which years form the baseline, how the trend is modelled, how gaps in a country's data are filled, and whether the raw count is adjusted. In the Nature paper, dropping one adjustment for Sweden and changing the trend model moved the estimate from 11,300 to 13,400.
What does a negative excess death figure mean?
It means fewer people died in that place and period than the baseline predicted. WHO's usual explanation is that measures such as lockdowns cut other deaths, including road injuries and seasonal flu. Like any excess figure, a negative one is an estimate, so it comes with a range.
Read the news better, every morning.
We send you the day's real progress in science, medicine and beyond, told plainly and sourced. Free, forever.
Free for good. One click to leave.