That study is cited in every debate about helmet use, yet it was barely scientific -- the researcher was his own test subject.

It would be interesting to see if the effect is reproducible under better experimental controls, such as analyzing video recordings from places where there is a lot of mixed car and bike traffic. Of course the analysis would have to be automated to avoid bias.

The data could then be combined with an estimate of the actual increase of crash risk correlated with passing distance, e.g., how many crashes are due to drivers passing too closely. What actual risk is being influenced here?

In my experience (anecdotal of course) of near-misses, the drivers were not aware of my presence at all, typically because they only look for what they expect to see. It's hard to imagine a mechanism in their brains, scanning for helmets but not cyclists.

Startup opportunity: bike helmets with wigs on top.