What's interesting about the Fermi Estimate post is that its examples encourage you look for predictors that are unexpectedly reliable, rather than those that first jump to mind.

That I haven't heard of many plane crashes in the past decade, this sounds like something I might hear on a post arguing an opposite point. "Sure, you've haven't heard of plane crashes in a decade, but why suspect that reliable predictors are in the neighborhood of your daily activities rather than your knowledge about the world? And now I will eye you knowingly until you learn something of your cognitive biases!"

Although, perhaps there aren't any estimated-statistics approaches that would be any good here, that wouldn't rely on other more incidental bits of information you happen to possess. Sure, you could try to list the top causes of flight disasters (human error and mechanical malfunction?) and estimate the likelihood these things occur, and also estimate how many of these flights result in large scale deaths. But there may be too many variables; either that, or I have ways to go in making Fermi estimates. In any case, it would be hard to incorporate the time variation of flight disaster outcomes. For all I know, flight safety has skyrocketed in the past decade due to widespread process improvement resulting from studying past disasters. And how could I ever predict that effectiveness, or predict how long it would take to come about?