If you search the forum for "freakonomics" (especially if it's spelled right--mine might not be), you'll find a tin of threads on the topic.
I can't go into great detail now, but the bottom line is that, yes, a seatbelt alone will likely keep a perfectly seated 2-year-old from flying out of the car, as the study found. However, the study didn't use dummies that measure things like abdominal injuries, nor is it in any way remotely likely that a 2-year-old will sit properly in a seatbelt.
So in reality, you're likely to have a dead and/or ejected child.
We enjoy the Freakonomics books (we own both) and the column that they write. However, I disagree with their assessment of the benefit of child restraints versus seatbelts alone. They based their assumptions by commissioning a limited crash test using dummies and I don't doubt their statistical analysis or the results of that testing - a dummy can use a seatbelt perfectly in a sled-based collision, but as has been stated a 2 year old is
highly unlikely to do so. I have a 2.5 year old and I'd never trust her to sit correctly with a seatbelt/booster!
Part of their assumption is that because most parents use/install car seats wrong, statistically kids would be just as safe with easier-to-use seatbelts. Now, I don't know the details of the study they did - whether it was just a sled and frontal impact testing, if any offset/side impact testing was done, or if any computer modeling was involved. I don't know if they purposely incorrectly used the test seat, or even what car seat(s) were used for the testing. I could pull out my copy of the book and check later in the day (once I'm snowed in
). I'm pretty sure that I recall them not giving a lot of detail at the request of the lab which performed the crash testing - they wanted to remain anonymous.
Without details of how the test was set up and instrumented, I'm not going to trust my kid's life to their conclusions. Real kids slump, shift, pull out the shoulder belt for comfort, etc. and that's where a simple crash test with a dummy or even a computer model breaks down. Real crashes are messy, not neat collisions of a sled into a barrier with no offset angle or other forces at play.
Yes, Levitt is a gifted economist who's far better at statistical analysis than I will ever be. But I'm an engineer who knows more than a little bit about solid body dynamics, and I'm a mom of a 2.5 year old - physics and common sense says that you can't look only at very limited crash test data (where the people who commissioned the study didn't even release actual engineering data!) to make an assessment like that, you need to review actual crash injury and fatality statistics. When you do, you see overwhelming evidence that for
actual use, car seats and boosters save lives.