* First instinct is that the fall-off should be more gradual if it's based on people's ages, not have that sudden sharp descent. I don't put a lot of weight to this though since the 20-year gap makes it sound like the exposure has to be under a certain age (and so within a relatively small range), which would appear as a relatively sharp drop.
* Second, after noticing the graph doesn't actually look like the one I posted (which jtmarmon posted confirmation to in a google doc), I took a closer look: It claims to be using "violent crimes per 100,000 population", but instead matches this chart from wikipedia [0], which isn't scaled by population - it's using raw numbers from the UCR data tool [1]. If the gasoline line is labeled wrong as well, then there's probably no issue, but if it's labeled correctly then half the correlation for this story isn't actually a correlation. Not sure where to look for that one, though.
