Thread by @iacobus42

@CityOfIowaCity raised the speed limit on a section of road. Reasoning that in the year has been open there have been no fatal crashes with peds = safe. Except their analysis is widely underpowered to detect an increase in risk. Like <5% power. #badstats #SafetyOverSpeed

If crashes happened before at 12 times a year, on average, they'd need the rate to *double* in order to have 80% power to find a signal. At 24 times/year, they'd need crashes to happen about 1.7 times more often.

For fatal crashes when happen less often (say 1 time per decade) they would need the rate to be well over 20 times worse than before to have a detectable signal.

Bad events, like crashes and especially pedestrian/bike crashes, are rare events. Most roads won't see any in a given year. Or even many years. But this does not mean the road is safe.

The same logic would indicate that playing Russian roulette is safe so long as the first pull doesn't fire the gun. That's crazy!

We can't use a single road segment over a short period of time to learn much about safety, especially of rare events, using the standard frequentist approach. That leads to the "Russian roulette is safe" problem.

Knowledge from context and other examples is required. A Bayesian approach would say things like "high presence of pedestrians", "entering urban area", "crosswalks", "recently exited an interstate" combined with high 85th percentile speeds leads to bad outcomes.

I also love how no one on Council or staff indicated that read either @onanov or @greg_shill's comments. But they must come 2nd to @iowadot: physical changes if volume > 20k are inappropriate. Is the corollary also policy: changes to increase safety are also inappropriate > 20k?