Things move quickly in the modern world. Within two hours of posting my academic paper on car seats vs. seat belts on the Freakonomics web page (the first time this paper had seen light of day), another economist found the paper and tested its hypotheses on a very different data set and reported back the results.
The economist is Paul Heaton, a graduate student at the University of Chicago and a former co-author of mine. (Our affiliation is largely irrelevant here — he just happened to have data handy to test my results — but I mention it in the interest of full disclosure). The data he has are from New Jersey motor vehicle crashes. The big difference between his data and mine are that he has all crashes (even if no one dies. I only had access to crashes in which someone died. This difference is important, because a concern in the fatal crash data is what economists call “sample selection.” The choices people make about what safety device to use will affect whether they die, which in turn affects which crashes I see in my data.
Heaton replicated the most basic specifications in my paper. His results are remarkably similar. He found no difference in the death rates or incapacitating injuries for children in car seats versus children using adult seat belts. Like me, he found a slight advantage for car seats in preventing non-incapacitating injuries relative to adult seat belts (car seats offered a 10% improvement for these less serious injuries in both of our samples). The only difference is that in his data the gap in injuries between car seats and seat belts is statistically significant, whereas in my data set it was not.
When I compare my findings to the existing NHTSA estimates on fatalities, I can see how our approaches differ and why we get different answers. What is more puzzling to me is why my results and Heaton’s both suggest very little injury benefit of car seats, but the medical literature often finds 70% (!!) reductions of injuries with car seats relative to seat belts. We find reductions that are an order of magnitude smaller. They use very different methods — surveying people in the weeks after crashes for instance — but still it is really a puzzle. Which is why, when you read my paper, I am extremely cautious in interpreting the injury findings.
I hope that the medical researchers, Heaton, and I can all work together to try to make some sense of the conflicting results being generated by these different methodologies to resolve this important question.