How many people have died in Ukraine?

NOTE: I now write for Aporia Magazine. Please sign up there!

The Kyiv International Institute of Sociology, a respected polling outfit, recently asked Ukrainians whether any of their friends or close relatives have died in the war. (The sample comprised adults living on territory that was controlled by the Ukrainian government as of 24 February, 2022: those living in the DPR, LPR and Crimea were not included; nor were refugees.)

63% of respondents said they had at least one friend or close relative who died in the war. The percentage was highest in the West and lowest in the East. This is somewhat surprising as the fighting is taking place in the South and East. Yet the West has long been the most pro-Western/anti-Russian part of the country, so it is likely that more men from the West have volunteered to fight.

Table from Kyiv International Institute of Sociology.

63% seems like a very high percentage, but it’s not obvious what the corresponding death toll is. One way of estimating it is to use polling data from the Covid pandemic.

At the end of January 2021, YouGov asked people in 16 different countries whether any of their close friends or relatives had died of Covid. Since we have accurate data on each country’s Covid death rate at the time, we can work out the relationship between the death rate and the percentage of people who say they’ve lost a friend or relative. Using this relationship, we can estimate how many people have died in Ukraine given that 63% of the population say they have lost a friend or relative.

To measure Covid death rates, I used cumulative excess deaths per million people on 31 January, 2021 – taken from Our World in Data. This measure was available for 12 out of the 16 countries in YouGov’s poll. There were no data for UAE, India, Indonesia or China.

Interestingly, the relationship between deaths per million people and % with a friend or relative who died was extremely strong (r = .96, p < 0.001). To estimate the death toll in Ukraine, I ran a linear regression model. It implies that there would be 6,268 deaths per million people in a country where 63% of the population had a friend or relative who died. This is shown graphically below.

Blue dots are the countries featured in the YouGov poll.

To get the total death toll, this number must be multiplied by the population of Ukraine. As of January 2022, this was 41 million (not including Crimea). However, we have to subtract the 6–8 million people who’ve fled the country since Russia’s invasion began, as well as the 3 million people living in the DPR or LPR. To be conservative, I’ll use a figure of 30 million for Ukraine’s population.

Multiplying 6,268 by 30 yields a total death toll of 188,000. This is substantially higher than official estimates of civilian deaths plus military deaths (though it is consistent with Colonel Douglas Macgregor’s estimates and with a report that 7% of alumni from a program at the Kyiv School of Economics had been killed).

In May, US officials told Fox News that 42,000 civilians and 20,000 soldiers had been killed. In June, the UN reported that at least 9,000 civilians had been killed. And back in December, the Ukrainian government reported that 13,000 soldiers had been killed. 188,000 is still vastly higher than US officials’ combined estimate from May.

There are a number of important caveats. The first is that Covid deaths are obviously different from deaths in a war. If the average person who has died in the war has more friends and relatives than the average person who died of Covid, then 188,000 will be an overestimate. This is plausibly the case. Many people who died of Covid were very old and may have had relatively few friends.

Second, a linear model isn’t the only one that could be fit to the data. A decelerating function would yield a lower death toll. Yet it is theoretically implausible. The true, causal relationship must be from the death rate to the percentage of people who say they have lost a friend or relative. But in order for the inverse relationship to be decelerating, that relationship would have to be accelerating – which isn’t plausible. Also, the linear model fits the data well and is more parsimonious.

Third, I used reverse regression to predict the death rate corresponding to a particular value of the % with a friend or relative who died. An alternative method is inverse regression: regressing y on x, and then rearranging the equation to find the value of x corresponding to a particular value of y. When I tried this, the resulting death toll was even larger (it was 203,000).

Fourth, due to population ageing, the cumulative numbers of excess deaths per million people are likely to be inflated. When I reduced them all by 10%, the resulting death toll was slightly lower (it was 169,000).

Fifth, the analysis was based on a regression model with only 12 observations. Although the relationship was highly significant, it is possible that if more data had been available the coefficient on % with a friend or relative who died would have been smaller, yielding a smaller total death toll.

Sixth, the KIIS poll and the YouGov poll used slightly different wordings. The former referred to a “family member and/or a close friend”, while the latter referred to "close relatives or friends”. It is not clear that this had any impact on the results, but it’s worth mentioning.

Overall, the analysis suggests that up to 188,000 Ukrainian soldiers and civilians may have died in the war – though this figure is likely to be an overestimate, since people who died of Covid likely had fewer friends than those who’ve died in the war. By how much it’s an overestimate is difficult to say.

Note: I’m not suggesting that the figure of 188,000 should be considered in any way definitive. However, it does seem hard to reconcile official estimates with the result from the KIIS poll.

Image: State Border Guard Service of Ukraine, Bakhmut, 2023

---

Thanks for reading. If you found this newsletter useful, please share it with your friends. And please consider subscribing if you haven’t done so already.