That has a very official-sounding name in statistics: "Going fishing."
The rule, in statistics, as in science, is you start with a hypothesis, and you test that specific hypothesis. If you feed a system endless amounts of data, it will find a correlation somewhere. That 95% correlation means there's a 95% chance of that correlation being valid; if you've tested for more than 20 correlations, you have around even odds of having found one.
See
[link]However, in this case the scientists have done their homework; they ran their tests, and then sought out other data sets to examine as well. Rick Nevin, one of the statisticians doing the research, has run the data on nine countries, and found strong correlations. Jessica Reyes, another statistician, ran the data on different states, comparing states on the basis of leaded gasoline phase-outs (different states followed different schedules on eliminating leaded gasoline); it tracked exceptionally well. There have been hundreds of citations and follow-ups on their work, and only one study out of all of them contradicted the findings. (And, for what it's worth, that study was funded by the largest lead fuel additive company, and engages in a lot of statistical hijinx, some of which are mentioned elsewhere in this thread.)
Moreover, studies performed have shown demonstrably higher blood levels in criminals than in the general population. The link between lead and aggressive behavior has long been known, as well as its effect on intelligence. Neurologist David Bellinger is one of hundreds of individuals who has published studies on the subject.
Unfortunately, virtually all of these studies are behind paywalls, so unless you have access to a university library I can't give them to you to examine yourself.