Projections based solely on history lead to ruin. Projections seduce us into a false sense of security. We don't know what's behind them, what assumptions they make, and yet we take great comfort in the precision of the numbers they proffer.
The recent earthquake exceeded the engineering assumptions by a long shot. What was supposed to happen once in a million years took only eight. Yet nothing about the situation seems surprising. Earthquakes happen, especially in Japan.
When building the nuclear power plant in Japan, it would seem reasonable to assume: 1) earthquakes happen; 2) they will continue to happen; 3) history has not witnessed the largest earthquake that will happen; 4) earthquakes generate tsunamis; 5) tsunamis generated by large earthquakes are huge; and 6) the frequency of earthquakes will likely increase. Or, you know, just don't build nuclear power plants in seismically charged areas — I'm sure a reasonable person, without a presentation filled with statistics, would conclude that's just asking for trouble.
The power plant should have, probabalistically, withstood an earthquake the likes of which the world has never witnessed, not fall to one so easily predictable. Some might call this hindsight bias, I call it margin of safety.
Things will not always continue as they always have:
Until the 35W bridge in Minneapolis collapsed, all the statistical data we had said highway bridges were very safe. Collapse was unthinkable based on the numbers. The problem with those statistics is that they were looking at a bunch of bridges that were built around the same time. The data also couldn't account for the pattern of neglect that U.S. infrastructure had undergone for a couple of decades (a form of political corruption). Once we looked at actual bridges instead of historical data, we discovered that many bridges were downright dangerous, on or near the point of serious failure. Without repair and replacement, bridge safety statistics were about to become obsolete in a big way.
We're at a very similar point with what we know about nuclear power production. We have an aging infrastructure, with plants nearing (or past) life expectancy. In order to determine what effect that's likely to have on safety, we need honest evaluation of the current situation, not just the assumption that things will continue as they always have. We are currently reliant on the industry for that evaluation. The question of how much we trust the industry is highly relevant.
Interested in learning more? Try reading Normal Accidents: Living with High-Risk Technologies.