Comments Posted By SpatialDan
Displaying 1 To 1 Of 1 Comments
I'm a spatial analyst. To Dr. Steve's first point, I do think they mitigate, but don't eliminate, the problem of overlapping deaths by confirming that the deceased resided in the house for three months before the death. This presumably increases the threshold for determining residence, since it would lead to fewer people reporting the deaths of transient family members who live next door. It's likely not possible to confirm this, though, because they did not record names of either respondents or deceased, given their IRB protocol.
With respect to spatial association, it is highly likely that the incidence of violent death is spatially correlated, for several reason. First, take the sectarian nature of at least recent stages of the conflict. We know for instance that Shiites and Sunnis living in mixed neighborhoods have been targeted in Bahgdad. There's a spatial relationship. It is also likely, though I am only speculating, that roving bands of militants do not rove randomly, but target certain areas, perhaps those close by. There's another potential spatial relationship.
The concern here is that both the overlapping deaths error and the spatial correlation error are likely to produce overestimated death rates. (All things being equal, they would also apply to the original 2004 study.) This would seem to undermine the tally of the study, particularly in the face of vastly different counts from other sources, which if not necessarily accurate themselves at least serve as a check for validity.
Comment Posted By SpatialDan On 11.10.2006 @ 16:10
