> He flat out admits that his only sources of his "research" are 1) the DoT NHTSA's FARS [0] and 2) four crash tests that were run by an anonymous test center.

> He flat out admits that his only sources of his "research" are 1) the DoT NHTSA's FARS [0] and 2) four crash tests that were run by an anonymous test center. Nope. You didn't watch it to the end, did ya? From the transcript: "In my data, and in another data set I've looked at for New Jersey crashes, I find very small differences in injury. So in this data, it's statistically insignificant differences in injury between car seats and lap-and-shoulder belts. In the New Jersey data, which is different, because it's not just fatal crashes, but all crashes in New Jersey that are reported, it turns out that there is a 10 percent difference in injuries, but generally they're the minor injuries." > The FARS data is heavily biased as well. First, it only reports _accidents with fatalities_. The NJ data reports injuries, too. > At the time of the video it spanned 30 years of data, 1975-2005. There's incredible amounts car safety improvements in that time frame. That point was addressed, by analysing more recent subsets of the data. From the transcript: "And the other thing you might argue is, 'Well, car seats have got a lot better over time. And so if we look at recent crashes -- the whole data set is almost 30 years' worth of data -- you won't see it in the recent crashes. The new car seats are far, far better.' But indeed, in recent crashes the lap-and-shoulder seatbelts, actually, are doing even better than the car seats > The only source that I found which I actually liked is called "What If These TED Talks Were Horribly Wrong?" Skepticism sells. Hate to be the one who appeals to authority, but I have more faith in a data scientist (with a PhD and published articles and all that jazz) doing analysis, than a meta-analysis by a journalist whose main achievement is writing plays, and whose main aim is create controversy for clicks.