[o09rdk]

I kinda wondered that looking at the paper. It's hard to know from eyeballing such low numbers relatively speaking, but it seems like there's not only a large peak in supercentenarians born shortly before the introduction of records (suggesting erroneously reported very old age), but maybe also a smaller peak after the records.

With name databases, there seems to have been a phenomenon where people retroactively created records when they were introduced (e.g., people registered for SSNs after the fact). Maybe there's something similar going on: people are overreporting their age when born with no records, but also are overreporting their age when they registered their birth after the fact.

It's a bit hard to explain in terms of underreporting age, and the trends are such that I think they still largely support the authors' hypotheses in terms of relative overreporting and underreporting. But I agree there's probably more fuzziness here.

The bigger issue for me that this raises are the perils of making inferences on outliers of any distribution. The further out you go on any characteristic, the more likely you are going to run into similar problems with errors, fraud, or unusual circumstances.