Naive observation of conflict data severely underestimates the tail risks of conflicts.

True if statistical analysis of war casualty data (e.g. Cirillo & Taleb 2015/2016, or independent analyses) shows that conflict casualties follow fat-tailed distributions, so that naive or sample-based estimates severely understate tail risk (e.g. the true mean or the probability of extreme events is much higher than the observed figures suggest). Mostly true if fat tails and underestimation are supported but the degree of 'severe' underestimation is disputed. Mixed if the literature is split, e.g. agreement on fat tails but substantial disagreement on whether naive observation underestimates. Mostly false if fat tails are confirmed but the evidence shows naive estimates are not materially biased downward. False if conflict data are shown to be thin-tailed, or naive estimates to be unbiased or overestimates.

Set before the research began.

1. E1 Fat tailed distributions for deaths in conflicts and disasters arXiv (preprint; authors not shown on the fetched page)

   “and this suggests a heavy tailed distribution where the average is controlled by the deadliest of wars.”

2. E2 Change Point Analysis of Historical Battle Deaths Journal of the Royal Statistical Society Series A (Oxford Academic) · 1 June 2020

   “It has long been understood that battle deaths data are characterized by heavy tails and are typically modelled by using a power law distribution”

3. E3 Trends and fluctuations in the severity of interstate wars Aaron Clauset (author PDF hosted by University of Oslo FocuStat blog)

   “These models indicate that both the long peace, and the period of great violence that preceded it, are not statistically uncommon patterns in realistic but stationary conflict time series.”

4. E4 The Decline of Violent Conflicts: What Do The Data Really Say? TU Delft Research Portal (Nobel Foundation Symposium 161 record) · 15 June 2016 Own record

   “We also establish that violence has a “true mean” that is underestimated in the track record.”

Independent sources agree that war casualty data are heavy-tailed. E2 says this "has long been understood" and that the data are usually modelled with power laws. E1 finds that the average is controlled by the deadliest wars, and E3 finds that long peaceful stretches are not statistically unusual in a stationary heavy-tailed process. These findings imply that sample-based estimates can understate tail risk. However, the explicit claim that the "true mean" is underestimated in the track record comes only from the subject's own co-authored work (E4), so how severe the underestimation is has not been independently confirmed.