Mixing Memory

Every once in a while I run across a paper that I have no idea what to make of. That happened earlier today, when I read a paper titled “Does television cause autism?” by Waldman, Nicholson, and Adilov (you can read the entire paper at that link). Television causes autism? If you’d asked me this morning, I’d have told you that was crazy talk, but this as of yet unpublished (and unsubmitted, apparently, which means unreviewed) paper takes the idea very seriously. Well, they don’t really set out to show that television is the cause of autism, but that it’s part of a causal chain that begins with genetic predispositions toward autism spectrum disorders. One of the explanations for the increase in the incidence of autism spectrum disorders over the last few decades is that the genetic predisposition requires some sort of environmental trigger, and that the prevalence of this trigger has increased during the period that has seen an increased incidence of autism. Waldeman et al. hypothesize that television is that trigger.

Why do they pick television? They list four reasons, which, when taken together, suggest television as a candidate for the environmental trigger. The reasons include the fact that television viewing has increased among children over the last few decades (due largely to increased access to cable television), the connection between television watching and ADHD, and behaviors consistent with television watching among “at risk” infants. That’s only three, right? I saved the last for a sentence of its own, because once again, it’s just odd: autism rates are extremely low among the Amish, who don’t watch any TV at all (you have to have electricity to watch TV).

As if things weren’t already surreal by this point, the first study is just odd. They start by using the Bureau of Labor Statistics’ American Time Use Survey to show that the amount of television households watch, on average, is positively correlated with the amount of precipitation an area gets. When it rains, people watch more TV. They then show that autism prevalence is positively correlated with the amount of precipitation an area gets. This indirectly associates television watching and autism. To strengthen their case, they also show in a second study that in two states (California and Pennsylvania), the prevalence of autism in a county during the 70s and 80s is positively correlated with the percentage of households in that county that have cable.

So that’s the evidence. It’s not just correlational, it doesn’t actually involve correlating autism and television watching, but autism and something that is correlated with television watching, and autism and cable — not watching cable, but owning it. Like I said at the beginning of the post, I have no idea what to make of this. I suppose it’s definitely enough to foster future research on the question, but before anyone starts fearing that their infant’s TV watching habits might cause him or her to develop autism, keep in mind that this study is pretty convoluted. I don’t like to use the old “correlation is not causation” objection because, well, in the absence of alternative explanations, correlation is often really good evidence of causation. But in this case, we don’t actually have a correlation between autism and television viewing rates, so we’re left with all sorts of potential third (fourth, fifth, sixth… nth) variables. Hell, maybe precipitation causes autism. Why not?

Comments

#1 jbark
October 16, 2006

Oh dear.

Well, at least I can hope that this nonsense creates less of a lasting fuss than mercury, facilitated communication, or cold mother syndrome.

In the meantime though, it’s led to a far greater sin: encouraging Greg Easterbook! http://www.slate.com/id/2151538/

Save me jeebus!!!!
#2 Patricia Mathews
October 16, 2006

I ran this little gem past the aspie mafia on lj. Silent Generation opinion thereon ranged from “Bah, humbug!” on up to, shall we say, cattle shed sweepings?
#3 Chris
October 16, 2006

My suspicion, whatever reasons they give, is that they tried to correlate autism and TV use (they had the data! they used it for the precipitation correlation), and couldn’t, so they went with something more indirect. I just can’t help feeling like this is some sort of hoax, though.
#4 BWV
October 16, 2006

What is more bizarre is that the authors of the study are on the faculty of Cornell’s business school and have no background in medicine or psychology.
#5 Todd
October 16, 2006

Yes, very strange indeed. I am generally supportive of creative research, but this just seems to be too much of a stretch. After all, if one were to look hard enough, I am sure a link (or a link of a link) could be found between just about anything. Heck, if the Washington Redskins can determine the outcome of all but one presidential election since 1936, then anything seems possible…

< http://www.blackwell-synergy.com/doi/abs/10.1111/j.1467-9639.2006.00243.x>

Cramer, K. M., & Jackson, D. L. (2006). Fans, Football and Federal Elections: A Real-World Example of Statistics. Teaching Statistics, 28, pp.56-57.

“This article evaluates and explores the correlation (+0.892) between the United States federal election winner and the most recent Washington Redskin home-game winner, a relation perfectly linked for 17 of 18 elections since franchise inception in 1936.”

#6 jbark
October 17, 2006

That’s awesome Todd – thanks for the heads up on that paper.

There appears to be a thriving cottage industry these days of people just essentially randomly correlating columns in Excel and publishing the results. Public Health type stuff in particular seems rife with it.
#7 Brian
October 17, 2006

Wow, that is an interesting paper! Cool topic for a post. You should definitely re-post if it does make it through peer review. Some problems I had:

The authors use previous and current estimates of autism rates rather selectively to show that there is a 10-fold increase in autism rate, when in fact, the epidemiology has always found highly variable rates across studies. They also fail to distinguish between incidence and prevalence, while at the same time claiming to be able to control for the many factors that distinguish the two. I don’t think I understand their methods well enough to assess whether that’s really the case, but they are, in my opinion, overly dismissive of the confounding nature of this problem. To quote Eric Fombonne:

As it stands now, the recent upward trend in rates of prevalence cannot be directly attributed to an increase in the incidence of the disorder. [emphasis his]

-Journal of Autism and Developmental Disorders, Vol. 33, No. 4, August 2003

In the paper, they pretty much take an increase in incidence as a given.

They are also suspiciously dismissive about potential confounds in the Amish cohort. While they acknowledge the problem of the Amish being a very different genetic population (I don’t think we can safely assume no random mating but can assume no migration), and probably very inbred compared to the populations of CA and OR, they just wave this away in favor of on (count it, one) “informal investigation” by a journalist. Not particularly compelling…

It’s also not clear to me that they really do a very good job of documenting TV viewing. They talk about subscription rates, introduction of the VCR, children’s networks, etc., but when it comes to the question of how much TV kids under three (and I’d argue that’s not nearly young enough) are actually exposed to, it seems like they rely on a lot of speculation. Later, they don’t talk about this, but are using these ATUS surveys, and there seems to be a correlation within a correlation in the way they assemble the data. Maybe I’m getting lost here – is it me or is this paper overly long?

Also, I have to disagree with you about the correlation/causation thing. I see way too many places where the causation could be the reverse of what they say it is – and in fact the authors themselves point out how this could be the case in other studies/theories, but are (I would say unjustifiably) resistant to the idea that this could also be the case with their data. It seems to me their conclusions ultimately rest on too many correlations that all need to be both accurate and causative.

Still, that precipitation correlation is pretty interesting…
#8 Chris
October 17, 2006

Brian, great points. I was as dumbstruck by their use of the Amish as you were. I mean, the number of differences between Amish populations and the general population is, well, very large, even if you discount genetic differences (the potential environmental triggers to which the Amish are not exposed, or are much less likely to be exposed, is significant, even excluding TV).

What I meant about the correlation/causation thing was that I hate the statement “correlation is not causation,” because it’s too general. For this paper, it applies, though, because there are obviously all sorts of confounds and potential third variables.

Oh, and yes, I did think it was too long. I suspect that if it is published (and it probably will be, given that it’s getting press), I suspect the editor will demand that they shorten it.
#9 CA
October 17, 2006

I would be embarrassed if it were published in my journal, but there are plenty of “alternative’ journals that are “peer-reviewed” by limited populations of sympathetic reviewers. There are so many variables that are not corrected for that it is dizzying. If they want to prove their point, they need to compare actual TV viewing at different ages of autistic vs. control kids. Then, there are still confounding variables and chicken-and-the-egg issues. for instance, do parents of autistic kids tend to use TV as a sitter because of the unique behaviors and demands presented by kids who are autistic? So does autism result in more TV time, or does TV time result in autism. I think it is bizzarre primarily because it is amateur. They are business dudes, not psychologists or phaysicians.
#10 Brian
October 18, 2006

For this paper, it applies, though, because there are obviously all sorts of confounds and potential third variables.

Yeah – I just can’t get over that precipitation correlation among different counties… That’s just wierd. Well, that’s the fun of these correlative studies, either they turn out to be right, or there’s an even more unexpected explanation that pops up down the road…
#11 Pat Mathews
October 18, 2006

Maybe autism causes TV watching!
#12 Chris
October 19, 2006

Yeah, I think a couple of commenters have actually suggested that. I keep waiting for someone to suggest that autism causes precipitation, though.
#13 Tammy Glaser
October 19, 2006

There were many things that bugged me about this “research”.

1 – Temperatures were found not to be statistically significant when correlated to television viewing. Excuse me, but, in extreme weather (Minnesota winters, Lousiana summers), where will you find the children? Uh . . . indoors. Sh! That does not count!

2 – They never actually sampled children (autistic and otherwise) and found out their television watching habits at different ages.

In my sample of two, my autistic child watched far less television under age two because she could not visually track nor process television. She did not become interested in television until after we started her evaluations. My son, neurotypical, watched way more television than her–he ought to the autistic one.

Rule number one of statistical research–you need to understand the issue you are studying. Clearly the authors of the study did not.

3 – Why is it so important that the Amish do not watch television and that overrides other factors of their lifestyle less exposure to indoor pollution, outdoor pollution, immunizations, lack of genetic variability, etc.? (Hint: only factors that match the hypothesis are *really* important–everything else you can pooh pooh to “prove” your point.)

4 – Underlying factors: autism tends to run in families with engineers/math types. Wouldn’t they tend to live in areas with greater access to technology? People familiar with autism know this.

5 – CORRELATION IS NOT CAUSATION!

6 – I seriously doubt this study could ever make it passed people who understand how and how not to conduct sampling and regress data.

For a laugh, compare this study to the circular logic of Monty Python.
#14 Alvaro
October 21, 2006

OK. So this is a random analysis. Which doesn’t merit so much attention, if any, until/ if it is published in a decent journal.

Question: Why has the media focused so much on this “thing”? How can we better educate the media and prevent wasting so much time and energies? I can imagine parents now being more confused than we are.

I am sure many journalists out there are not autistic. It is not raining today. Let’s explain the filtering virtues of PubMed. Does someone here know the Slate author?
#15 Andrew Wade
October 26, 2006

For this paper, it applies, though, because there are obviously all sorts of confounds and potential third variables.
Speaking of which, I’ve just run across a post looking at a confound for this:
http://autismnaturalvariation.blogspot.com/2006/10/precipitation-rates-vs-population.html
#16 chiz
October 26, 2006

I wonder if vitamin D is the link. Spend less time outside when you’re pregnant and have lower levels. There have been similar suggestions for MS and schizophrenia.
#17 JG
February 24, 2007

I love tv, but I already thought tv was the cause of autism on my own. It is so easy to see. Evolution. Autism (now affecting 1/166 people) – evolution of our brains due to eveyones hours spent watching television over the years. Plain to see. Didn’t anyone think there would have to be a major downside to something we all love so very much. Just explaining something to an autistic person alone means the need for visuals. The brain has come to rely on visuals now because of tv and can’t grasp ideas well without them. (I have an autistic son.) I’m now checking out if others thought of the same idea as me (television causing autism)that is why i’m here at this site typing this comment right now.