Obermeyer and colleaques1 compare their estimate of 5,393,000 war
deaths in 13 countries, 1955-2002, with a PRIO battle-death figure of
2,784,000, a ratio of 1.9 war deaths per battle death, and claim that PRIO
data “indicated a figure of only a third” of their own estimate.2 The 3.0
that appears in Table 3 turns out to be an unweighted average of the 13
ratios of the Obermeyer estimates to the PRIO figures: one for each
country. Thus, Georgia with 0.6% of the estimated war deaths and a ratio
of 12.0 gets the same weight as Vietnam, with 71% of the war deaths and a
ratio of 1.8. Without Georgia the mean ratio falls to 2.2, close to the
weighted average of 1.9 as well as to the median of the 13 ratios (2.1).
The authors also conclude that “there is no evidence to support a
recent decline in war deaths.” They base this claim on an extrapolation
from 13 data points. Their regression equation is completely dominated by
the huge Vietnam point. The equation is:
Estimated War Deaths = 27,380 + 1.81*(PRIO Battle Deaths)
The estimated constant in this relationship, 27,380, is far from
statistical significance. All PRIO figures, 1955–94, are then passed
through this equation, including the insignificant constant without which
the PRIO trends would have been preserved exactly. In fact, Obermeyer et
al.’s own data show a clear decline in war deaths over time. Had they
included the Korean War and the two world wars, as did Lacina, Gleditsch
& Russett (2006)3, the decline would have been even more evident.
1 Obermeyer Z, Murray CJL, Gakidou E. Fifty years of violent deaths
related to war from Vietnam to Bosnia: analysis of data from world health
survey programme. BMJ 2008 doi: 10.1136/bmj.a137.
2 This is an “apples versus oranges” comparison since the category of
“battle deaths” is substantially narrower than the category of “war
deaths”. By PRIO’s definition, battle deaths can only occur during
contested combat incidents involving two opposing sides, one of which must
be a State. War deaths include many additional types of incidents
including massacres and inter-ethnic violence.
3 Lacina B., Gleditsch NP, Russett B. The declining risk of death in
battle. Int Stud Q 2006; 50: 673-80.
Competing interests:
None declared
Competing interests:
No competing interests
03 July 2008
Michael Spagat
Professor of Economics
Department of Economics, Royal Holloway College, University of London, Egham Surrey TW20 0EX, UK
Rapid Response:
Inconsistent Comparisons and Unsound Conclusions
Obermeyer and colleaques1 compare their estimate of 5,393,000 war deaths in 13 countries, 1955-2002, with a PRIO battle-death figure of 2,784,000, a ratio of 1.9 war deaths per battle death, and claim that PRIO data “indicated a figure of only a third” of their own estimate.2 The 3.0 that appears in Table 3 turns out to be an unweighted average of the 13 ratios of the Obermeyer estimates to the PRIO figures: one for each country. Thus, Georgia with 0.6% of the estimated war deaths and a ratio of 12.0 gets the same weight as Vietnam, with 71% of the war deaths and a ratio of 1.8. Without Georgia the mean ratio falls to 2.2, close to the weighted average of 1.9 as well as to the median of the 13 ratios (2.1).
The authors also conclude that “there is no evidence to support a recent decline in war deaths.” They base this claim on an extrapolation from 13 data points. Their regression equation is completely dominated by the huge Vietnam point. The equation is:
Estimated War Deaths = 27,380 + 1.81*(PRIO Battle Deaths)
The estimated constant in this relationship, 27,380, is far from statistical significance. All PRIO figures, 1955–94, are then passed through this equation, including the insignificant constant without which the PRIO trends would have been preserved exactly. In fact, Obermeyer et al.’s own data show a clear decline in war deaths over time. Had they included the Korean War and the two world wars, as did Lacina, Gleditsch & Russett (2006)3, the decline would have been even more evident.
1 Obermeyer Z, Murray CJL, Gakidou E. Fifty years of violent deaths related to war from Vietnam to Bosnia: analysis of data from world health survey programme. BMJ 2008 doi: 10.1136/bmj.a137.
2 This is an “apples versus oranges” comparison since the category of “battle deaths” is substantially narrower than the category of “war deaths”. By PRIO’s definition, battle deaths can only occur during contested combat incidents involving two opposing sides, one of which must be a State. War deaths include many additional types of incidents including massacres and inter-ethnic violence.
3 Lacina B., Gleditsch NP, Russett B. The declining risk of death in battle. Int Stud Q 2006; 50: 673-80.
Competing interests: None declared
Competing interests: No competing interests