TV causes autism? I doubt it.

An article in Slate yesterday argued that TV watching causes autism. The Slate article is based on research done by Cornell economists Michael Waldman, Sean Nicholson, and Nodir Adilov. You can download the academic working paper here.

The paper gives some theories why TV and autism might be linked, but the more interesting part of the paper is the data analysis. The researchers are trying to find a “natural experiment” that shifts around TV watching, but otherwise has no impact on whether a child is diagnosed as autistic. Rainfall is one of the things they use. In places where it rains a lot, kids watch more TV. Maybe rainfall doesn’t affect autism in any other way. This is a creative approach, although it suffers from the weakness (which they acknowledge in the paper), that rainfall changes other things, like how much time you spend indoors doing other things besides watching TV. They also use the arrival of cable TV in an area. This approach is potentially stronger, although it would be better if they used availability of cable TV, rather than the number of people who actually subscribe.

These are intriguing approaches, but personally I did not find the empirical evidence in the paper very compelling.

The rainfall evidence is based off of three states: Washington, Oregon, and California. It rains a lot in some parts of these states, but not others. There is more autism in the parts of the states where it rains more. The problem is that it rains on the coast of Oregon and Washington, and in Northern California. But there are a million other differences between the coast of Washington and the Eastern part of the state, and between Northern and Southern California. The researchers also look at how much rain there was when you were between the ages of 0-2, controlling for your county. This is more promising. The impact of rain gets smaller, but it is the most convincing evidence in the paper.

The data analysis of cable TV is limited to California and Pennsylvania and also finds positive results. The difficulty with the cable TV analysis is that there is an incredibly strong positive trend in autism. The cable TV data are basically on an upward trend. The regression analysis is going to have a very hard time sorting out between a steady rise in cable TV penetration and the time trend. In the current version they only include a linear time trend, which is an extremely powerful predictor. My guess is that if they generalize their specification to allow for non-linear time trends, the cable TV result will disappear.

The authors have done some interesting work, but the nature of the problem makes it a really hard one to answer convincingly. For instance, you might think that Oregon and California should have similar autism rates. Nope, Oregon’s rate is four times higher. That sort of gap is almost certainly due to differences in what is called autism in the official data in one state versus the other. The increasing time trend is also heavily influenced by what is labeled autism. When the outcome of interest is measured so poorly, it is hard to know what the analysis is really picking up — differences in the underlying symptoms or just in the reporting of them.

The more I thought about it, the more it seemed to me that there might be a causal link between rainfall, TV, and autism, but not the one suggested by the paper.

My theory: when it rains a lot, parents watch more TV, see more shows about autism, and this leads them to seek out a diagnosis of autism for their kids. They have the same kids, it is just that TV makes them believe that their kids are autistic.

I don’t mean to sound overly negative on this research. I applaud the authors for asking a daring hypothesis and gathering data to try to test it. My gut, though, tells me that this is a result which will not stand up to scrutiny.