Have children's car seats not been proven to be more effective than seat belts alone for kids older than 24 months?

Question

A comment on a question on the Travel Stack Exchange claims that "child seats [in cars] have never been proven to be more effective than seat belts for kids older than 24 months", and links to a Freakonomics article published in the New York Times (archive link). That article summarises an analysis of data from the Fatality Analysis Reporting System (FARS) as follows: "there is no evidence that car seats do a better job than seat belts in saving the lives of children older than 2". The article does not describe the actual analysis conducted in much detail, and does not cite any other studies on the subject.

A Google Scholar search does not find recent articles that investigate this question, research seems to be focused on measuring prevalence of child seat use, effectiveness of training programs on child seat use, or correctness of child seat use. Is there evidence to support or refute the argument made in the Freakonomics article?

Answer 1

There is evidence otherwise

The second author of the NYT article is Steven D. Levitt, who also wrote Evidence that Seat Belts are as Effective as Child Safety Seats in Preventing Death for Children Aged Two and Up, published in the peer-reviewed journal The Review of Economics and Statistics. From the abstract:

Using data from the Fatality Analysis Reporting System (FARS) on all fatal crashes in the United States from 1975-2003, I find that child safety seats, in actual practice, are no better than seat belts at reducing fatalities among children aged 2-6. This result is robust to a wide range of sensitivity analyses, including controlling for sample selection that arises because the FARS data set includes only crashes in which at least one fatality occurs.

While the NYT article makes its own conclusions regarding risk of injuries, the study did not even attempt to research that, nor could it with the dataset. Digging deeper, I found a paper (Effectiveness of Child Safety Seats vs Safety Belts for Children Aged 2 to 3 Years) citing that one which explains a lot more:

[Levitt's] work is out of step with the literature. It contrasts with the literature establishing the effectiveness of child safety seats relative to unrestrained travel. More important, subsequent analysis by Elliott et al found that child restraints for children aged 2 through 6 years, when not seriously misused, reduce the risk of death by 28% compared with seat belts. Prior studies also found that child safety seats were more effective against fatalities than safety belts. Relative to unrestrained travel, child safety seat restraints reduced the risk of death by an estimated 47% for toddlers aged 1 to 4 years. At most, though, for this age group, they offered a modest fatality risk reduction vs safety belts, which were estimated to be 45% effective against deaths among people 12 years and older and 36.5% to 39% effective for toddlers.

The analyses by Levitt and Dubner and Levitt have 3 major limitations. First, it included children up to the age of 6 years, meaning it confounded seat effectiveness for children of appropriate weight and stature for a child seat and larger children whose restraint in a child seat constitutes misuse. Second, it used a fatal crash data file that lacked a good control variable for crash severity. Third, it largely restricted its analysis to fatalities, with some glancing attention to nonfatal injuries coded by police using a coarse inaccurate severity measurement system. Fully analyzing the appropriateness of child safety seat laws for toddlers, however, also requires considering the relative impacts of child safety seats and lap-shoulder belts against nonfatal injury. A few studies have examined the effects on nonfatal injuries to specific body regions (eg, severe abdominal injuries) but not on total injuries. Only 2 studies used a design similar to that used by Levitt and coworker to analyze effects on nonfatal injury. One limited the analysis to significant injuries, which were defined as an Abbreviated Injury Scale (AIS) score of 2 or more. Neither distinguished lap-shoulder from lap-only belts. One found that children aged 2 to 5 years in seat belts had a 3.5 times higher relative risk of significant injuries than those in child restraint systems. The other found an odds ratio of 0.22 (95% confidence interval, 0.11-0.45) of significant injury in a child seat vs a safety belt, but found no difference in the odds of minor injury. Our study covers all nonfatal injuries and, like the work by Levitt and coworker, it specifically examines lap-shoulder belts. It addresses the probability of being injured, analyzes crashes that resulted in at least 1 vehicle being towed away, and is restricted to passenger vehicles with occupants aged 2 to 3 years.

As the title would indicate, the injury angle is the main focus of that newer paper, which examined data on toddlers (ages 2–3) who were in accidents where at least one car had to be towed, over the years 1998–2004 in the US. Its summary:

The adjusted odds of injury were 81.8% lower (95% confidence interval, 58.3%-92.1% lower) for toddlers in child seats than belted toddlers.

Answer 2

I shall start by saying that this is a somewhat sensitive subject, as giving the wrong advice could have terrible consequences. If you are reading this as a way to make a decision about how to transport your child(ren) then I would encourage you to have a look at the data. Cars are really dangerous, and it seems there are things you can do to reduce the risk. Do not listen to me, I do not have children or any professional experience in road safety.

In trying to answer this as data question I started with a 25 minute youtube video that introduces most of the papers referenced below.

It is easy to find correlation between injury and death rates and child seat use. For example Elliott et al. 2006 found a relative risk for death of 0.72 (0.54-0.97 95% CI) associated with the use of a child restraint system in children of ages 2 to 6, and 0.79 (0.59-1.05) when excluding cases of serious misuse. This represents a 21% reduction in the chance of dying in an accident. Arbogast et al. 2009 find reductions of adjusted relative risk for injury in children of ages 4 to 8 of 0.55 (0.32–0.96), representing a 45% reduction in risk. This effect appears particularly pronounced in younger children, with Zaloshnja et al. 2007 finding a 81.8% lower chance of injury in children aged 2 to 3.

However this sort of study has been questioned in a number of ways. Probably the hardest to deal with is the question of how much of this correlation represents the use of child restraint systems causally resulting in lower rates of death and injury of the individual, and how much this represents the use of child restraint systems serving as a marker for wealth, the use of safer cars and/or more responsible driving behaviour. There are also questions about the validity of the data. The presence of child restraint systems is easier to prove than the use or not of a seat belt, so it is postulated that when insurance professionals ask parents of children who where unrestrained they may be incentivised to say they had a seat belt on. This would inflate the rate of harm associated with seat belt only use.

One study that attempted to control for these variables is Levitt 2005 who finds that child safety seats are no better than seat belts at reducing fatalities among children aged 2-6. Another is Anderson et al. 2009 that separates out booster seats and full seats with their own four point harness. This finds the estimates for chance of dying actually higher for booster seats than just seat belts, though none of the differences are statistically significant (Table 3, Panel II). Ma Et al. 2013 look specifically at booster seats compared to seat belts, and find significantly increased rates of some injuries, specifically "AIS > 0 injury to the neck and thorax".

It must be noted all these studies highlight the problem for incorrect installation of child safety seats as an unmodelled variable in the data, with the first two estimating this rate at 80% and >50% respectively. The importance of properly fitted seats is illustrated by Ciccone et al. 1997 who find at 25 mph "For the properly restrained dummies, the safety seats provided excellent protection from the crash forces. For the improperly restrained dummies, protection ranged from adequate to nonexistent, depending on the type of misuse." This could disguise a significant positive effect of properly installed seats.

On a similar vein Edgerton et al. looked at children between 20 and 40 lb in either a full child seat or booster seats and found odds ratios between 4.4 and 29 for various injury metrics. They interpreted this as the harm from switching from full child seats to booster seats too early.

There are few crash tests that directly compare seat belts to child seats. Kriston et al. (which is not peer reviewed, but published by the US DoT) compared a booster seat to seat belts with and without belt guides with a crash test dummy "aged" approximately 12 years, and find the booster seat 33% better (16%-50%). Talimian et al. 2023 do computer modelling and find a four point better that a three point, but their models do not look much like what I see sold as car seats.

Another point that is worth mentioning is the importance of position in the car on the chance of injury. As I have mentioned on mechanics.SE Arbogast et al. 2008 found "children seated in the center rear have a 43% lower risk of injury compared with children in a rear outboard position".