The death toll is currently running at 116,000. This is almost an order
of magnitude higher than the initial 14,000 death estimates. By contrast,
if you recall the initial 9/11 estimates were about 10,000 and the final
number dropped to around 3,000. In 9/11, the estimates were mostly
generated by figuring out how many people were potentially killed and
then crossing them off the list as they turned up alive. This time,
however, it seems that the estimated get updated as bodies turn up.
It would be interesting to compare initial fatality estimates that appear in the news to the final accounts. Another interesting question is whether it would have been possible to generate better estimates. It would seem that some geographic-based sampling could have gotten us within the right order of magnitude within the first day.
As I watched these numbers change, I've wondered to myself, who creates the numbers? Presumably national governments - and what incentives do they have for making them large (or small). Do these incentives change as time goes on? For example, initially do you want to downplay the incident to reduce widespread disruption then as time goes on do you want to up the issues to make sure you get your fair share of relief efforts? I've got no idea - but would be interesting to talk to someone that did.
I would guess, off the top of my head, that death toll estimates peak and decline over time, as information on the total number of missing is gathered, then some of the missing are eventually accounted for. On 9/11, the first part of the curve was presumably very fast, so it looked like the "initial" estimates were overstated. I expect that if careful records continue to be kept, then the total death toll from the recent tsunami will be somewhat lower than the peak estimate. Of course, we may never actually see the headline, "tsunami toll lower than previous estimate", because it's not particularly interesting to faraway Western audiences.
Geographic based sampling is really really hard to effect well if you have no access to a very large percentage of the geography, say due to some recent natural disaster.