Heresy again-- Child seats necessary?

The Freakenomics guys, Stephen J. Dubner and Steven D. Levitt, are at it again:

(Note: Longer than normal quotes below because of the difficulty of retreiving NY Times archives).

According to the NHTSA manual, more than 80 percent of car seats are improperly installed.

Perhaps the single most compelling statistic about car seats in the NHTSA manual was this one: 'They are 54 percent effective in reducing deaths for children ages 1 to 4 in passenger cars.'

But 54 percent effective compared with what? The answer, it turns out, is this: Compared with a child's riding completely unrestrained. There is another mode of restraint, meanwhile, that doesn't cost $200 or require a four-day course to master: seat belts.

For children younger than roughly 24 months, seat belts plainly won't do. For them, a car seat represents the best practical way to ride securely, and it is certainly an improvement over the days of riding shotgun on mom's lap. But what about older children? Is it possible that seat belts might afford them the same protection as car seats?

The answer can be found in a trove of government data called the Fatality Analysis Reporting System (FARS), which compiles police reports on all fatal crashes in the U.S. since 1975. These data include every imaginable variable in a crash, including whether the occupants were restrained and how.
Even a quick look at the FARS data reveals a striking result: among children 2 and older, the death rate is no lower for those traveling in any kind of car seat than for those wearing seat belts. There are many reasons, of course, that this raw data might be misleading. Perhaps kids in car seats are, on average, in worse wrecks. Or maybe their parents drive smaller cars, which might provide less protection.
But no matter what you control for in the FARS data, the results don't change. In recent crashes and old ones, in big vehicles and small, in one-car crashes and multiple-vehicle crashes, there is no evidence that car seats do a better job than seat belts in saving the lives of children older than 2. (In certain kinds of crashes -- rear-enders, for instance -- car seats actually perform worse.) The real answer to why child auto fatalities have been falling seems to be that more and more children are restrained in some way. Many of them happen to be restrained in car seats, since that is what the government mandates, but if the government instead mandated proper seat-belt use for children, they would likely do just as well / without the layers of expense, regulation and anxiety associated with car seats.

NHTSA, however, has been pushing the car-seat movement ever further. The agency now advocates that all older children (usually starting at about age 4) ride in booster seats, which boost a child to a height where the adult lap-and-shoulder belts fit properly. Could this be a step in the wrong direction? In 2001, the Insurance Institute for Highway Safety sent NHTSA a memo warning that its booster-seat recommendations were 'getting ahead of science and regulations' and that certain booster seats ''did not improve belt fit, and some actually worsened the fit.' (all emphasis added)

North Dakota has a booster seat law taking effect this year. I want what is safest for my son. It seriously concerns me that many states are passing laws without proper tests and may be making the situation worse.