Several other factors might affect the results from the death toll study:

Baseline mortality rates—The researchers inferred the number of deaths caused by the war by comparing pre- and post-invasion mortality rates. The number they came up with for the pre-invasion rate was 5.5 deaths per 1,000 people. As the authors point out, this figure matches up with estimates from some other sources, like the CIA. On the other hand, numbers from the United Nations suggest a higher baseline. If there really is a discrepancy here, it's not clear how to interpret it: The researchers gathered pre- and post-invasion mortality rates in exactly the same way, so you might expect any bias to affect both figures equally. In that case, the error would disappear in the analysis.

Deception—It's possible that some respondents lied about deaths in their household, either by giving spurious casualty reports or by concealing real ones. The researchers discount this problem by pointing out that they were able to corroborate reported deaths with death certificates in 92 percent of the cases. That doesn't explain why their results are so at odds with the government numbers, which apparently draw from the same official documents.

Gaps in the data—When the researchers knocked on a door and didn't get an answer, they moved on to the next house. Empty houses—in which every resident might have been killed—were therefore omitted from the study. Families that refused to answer questions were also omitted. It's not clear whether this sampling bias would inflate or deflate the study's death count.

Status of the victims—Not everyone is counting the same thing. The death toll figure from the Lancet study doesn't distinguish between combatants and noncombatants, while the Web site Iraq Body Count—which pulls casualty numbers from the media—only counts civilian deaths.