Autism, TV, Precipitation: Dismal Science

There's a lot of buzz around a recent web publication by professors of business, policy, and economics at Cornell and Purdue Universities, who theorize that allowing children aged 0-3 to watch television causes autism. The authors compared TV exposure times to annual precipitation and come up with a theory that they outline in 67 pages of analysis and findings. Finally they conclude:

"Hence, our results suggest that early childhood television watching, or whatever is the trigger driving our finding of a positive correlation between autism rates and precipitation and autism rates and cable, is an important factor in autism diagnoses both from statistical and absolute standpoints." [emphasis ours]

67 pages for the "whatever is the trigger" theory?

Scientists are sure to question the ambiguous conclusion. But maybe they should take heed. Maybe the researchers are right. Perhaps more economists, policy and business professors should do science research. Scientists struggle to understand the causes of autism. They undertake extensive studies trying to isolate the myriad factors underlying increasing rates of autism diagnosis. They analyze a dizzying number of possible genetic and environmental triggers. Are they making mountains out of molehills?

The authors show how to cut this tedium by parsing the plethora of possibilities and inputs with their special brand of academic/business/economic/management Jujitsu. They derive a simple conclusion: TV viewing causes autism. For whatever reason, this theory eluded scientists, public health experts and physicians for decades. Worse still, according to these brazen science barons, even "the possibility" was "ignored" by researchers.

The methods may seem unconventional by science standards, but by simply ignoring the time intensive processes scientists habitually undertake, these authors stride forward with unprecedented speed to reach conclusions and propose future research and policy. We sifted through their study aiming to convince ourselves of their theory. Could we resolve autism by reducing TV time? How simple, how easy -- what a perfect solution. Why didn't these new kids on the block show up earlier to let the air out of the tires of the tall truck stuck under the short bridge?

Of course, we don't for a moment pretend to understand all their research, we've only dabbled in any of these subjects - autism, economics, business or policy (we didn't even have TV growing up). But here we'll take the authors' cues and audaciously deconstruct their methodology. Step by step, we'll show how we think they did it following ten simple rules (1-10), and how you too, can arrive at such compelling conclusions. When we don't understand their methods or magic we'll wave our hands, skip over the details, bluff, and make a joke of it all.

1) First, disarm your reader by acknowledging many possible causes for autism. Continue to do this throughout the study, but shed doubt on these possibilities to bolster support for your own theory. Choose any simplistic theory, make it your theory, then without hesitation, **prove it**. The authors declare that TV is the environmental trigger for autism. Sure, autism could be genetically linked or triggered by toxicants or air pollution. There are hundreds of possibilities. But trivialize these briefly and dismissively en route to proving your own theory.

2) Slight other theories about autism by pointing out flaws in decades worth of scientific research, but cite only a handful of studies. Say that genetics is altogether "discredited". Say the air pollution links are "intriguing". Suggest the preposterous with a straight face: "families who are more prone to have autistic children for other reasons, tend to locate in areas characterized by higher pollution levels." It's well-accepted that the perception of higher rates of autism is because we have better diagnosis methods. Seed doubts about this by saying that researchers have "mixed conclusions".

3) Having quickly acknowledged then dispelled previous scientific research, move on to explain the "rational" behind your theory. Call your impressions and ideas "reasons", and keep the number of them low, ie: "Four reasons to suspect TV".

  • First Reason: "Historical data are not very good" (the case of California):

    The US Department of Education only changed requirements that effected autism reporting in the 1990's, making rises in the incidence of the condition difficult to discern. However California passed laws in 1969 requiring the establishment of service centers to provide services for developmentally disabled children. Rates of autism at these centers increased during this time. So did TV watching. Aahha!

    But this could be explained by increased numbers of diagnoses of autism and more parents visiting service centers. Or the fact that there actually were service centers, could have, all on its own, led to the increase. Increases in toxicants or pollution also would cause increase visits. Fortunately, you've thrown out these possibilities already, no need to revisit them. Autism rates have increased in the past decades, most people agree, and now you've linked this to television viewing. It never hurts to reinforce your point, so cite some more irrelevant data to support it, like the increased sales of VCRs, cable deregulation, the rise of Nickleodeon and Disney, and increases in the number of television sets per household.

  • Second: Attention deficit hyperactivity disorder (ADHD):

    You managed to find one paper showing a correlation between ADHD and television viewing. Note that there were no controls and no proven cause and effect in this paper, but say that the results "are suggestive", "of interest", and "certainly suggestive". [We'd suggest something more, that since ADHD begins with "A" and Autism begins with "A" too, TV could cause both.]

  • Third: High risk kids "engage" with TV: Cite a study showing that children were more likely to become autistic if they were at "high risk" for autism, defined as those who had a sibling with autism. You previously discounted the genetic link, but whatever. These children were found to "disengage" more slowly from TV than their "low-risk" peers, therefore you theorize that they get more TV exposure and more autism.
  • Fourth: The Amish. There are low rates of autism in the Amish, according to some reporter somewhere. Acknowledge that this is sort of speculative, but say that "even with all these caveats", the reporter's "findings provide intriguing evidence". Important not to shy away from using words like "evidence". Sprinkle authoritative sounding science words liberally throughout your paper: "evidence" (30 times), "data" (85 times), and "results" (64 times).

4) Having explained the reasons you suspect the link, now explain your methodology, no matter how convoluted. You chose three states, California, Washington, and Oregon, because of their "high precipitation variability". This allows comparisons between areas of low precipitation and high precipitation with corresponding autism rates. Explain the onerous county and state data collection process that no doubt stymied other researchers. For instance of the three states, Washington was "unwilling to provide autism data". And "Oregon only reported the county autism count when it was greater than or equal to ten". Since the data are obviously totally inconsistent between states, counties, cohorts, explain how you resolved this. For example:

"For Oregon we use age-specific counts by county in 2005 and then construct autism rates by dividing by the corresponding county-level age-specific population taken from the 2000 census. For the case of California we focus on cohorts born between 1982 and 1997 (vs cohorts born between 1987 and 1999 for Oregon) and use the county autism count in the year a birth cohort was eight yrs old...dividing by that year's corresponding county-level age-specific population....For example, for children born in LA county in 1990 we use LA county's autism count of eight year olds in 1998...."

Continue like this, variable by variable, Black and Hispanic coefficients, income coefficients, education coefficients, sex of interviewee coefficients, household type, child gender, etc., paragraph after paragraph, page after page. This helps shed light on the morass of inputs and the challenges that have made conclusive answers so difficult, but also potentially increases the *wow* factor of your paper. Say that you've considered all of this and controlled for every nuance.

5) Since you may well have lost the reader by this point, model the data, as any economist would, into four easy to follow formulas, included below.

  • For example, this formula represents the television viewing of a child measured in precipitation at the respondent's location on the day of the survey, and other variables:

    "TVi= β12PRCPi + β3PRCPi²+ β4Xi+ <βZi+&epsiloni"

  • Most of this is fairly intuitive.

    "AUTk=&beta1+&beta2PRCP+&epsilonk", and

    "AUTk12PRCPk3logPOPk+ β4INCk + β5REGk+ β6HISPk7BLKk+ β8INDk+&epsilonk"

    Here in the second and third formulas "AUTk denotes the 2005 autism rate among school-aged children in county k. PRCPk is the average precipitation level in county k between 1987 and 2001, logPOPk is the logarithm of count k's total population in 2000, INCk is county k's per capita GNP in 1999, HIS and BLK and IND represent Hispanic and Black and Indigenous children...."

  • Finally, this fourth formula represents pooled data between the states, for another look at the relationships:

    "AUTk,b12PRCPk,b3TIMEb+ β4logPOPk + β5INCk+ β6REGk7HISPk,b+ β8BLKk,b+ β9INDk,b+&epsilonk,b"

6) Restate your conclusion, which unsurprisingly, was your theory. Dismiss any "anomalies." Now that you've modeled all of this like Play-Doh you can state authoritatively:

"There is a positive relationship between autism and precipitation in Oregon and Washington, but no [sic] in California...We present the precipitation and autism maps for each state[..] It is clear from the maps that there is a very strong correlation in each state between precipitation and autism."

Each state except "no California". Ouch. Tackle this conundrum, that even after your incredible machinations, one-third of your data set - California - doesn't fit your theory.

You can pull this out. The map shows that where it doesn't rain there are still high rates of autism--well it's more complicated -- higher than the median for the state. Suggest that there could be other factors, such as "urban density", concurring with high autism rates which might make precipitation levels irrelevant. [This theory would also fit the OR and WA data]. Manipulate the data to erase this aberrant possibility by using the "fixed-effects specification". Since your releasing this on the internet, most people don't know what that is, oh well. Just observe that autism rates vary as precipitation deviates from the "average". Does this prove that there aren't other inputs? Say it does. Now you've accounted for that whole "urban density" issue as well as other possible perturbations. The alarmingly titled "omitted variables problem" is eliminated, so phew, therefore California fits your theory.

7.) Note some problems with the study, then dismiss them. Suggest that one potential problem with this data is that indoor activities in general, not just TV watching, may account for higher rates of autism in places with greater precipitation. If this were the case, any indoor toxicant may result in higher autism rates since TV time is coincidental to indoor time. Explain that you've resolved this by comparing Pennsylvania and California cable subscription rates with autism rates. Let the reader wonder why you chose these two states. Pennsylvania? Ignore precipitation for this analysis. Does it matter, for instance, that that most of Pennsylvania has higher rates of precipitation than California? Whatever.

Model your findings for this: "AUTk,b=&beta1+&beta2CABk,b+&beta3TIMEk+&beta4POPk+&beta2INCk+&beta6REGk+&beta7HISPk+&beta8BLKk+&beta9INDkk"

Say that increased cable subscriptions correlate with increased rates autism. Does this prove that indoor toxicants are irrelevant? Or has increased technology coincidentally increased with autism? Never mind. Say indoor toxicants are irrelevant.

8.) State your theory again: Television watching causes autism "or whatever is the trigger driving our finding of a positive correlation between autism rates and precipitation and autism rates and cable". Humbly consider the flaws in your research, ie:

"because we do not provide a direct test of the effects of television watching on autism, we do not consider our results to be definitive evidence in favor of the television viewing as trigger hypothesis."[emphasis ours]

9.) Now that you've duly noted potential flaws, like, the whole study might be hogwash, ie: 'we do not consider our results to be evidence of TV viewing as a trigger for autism', pretend you didn't say that. Instead, forge ahead and state with confidence that autism is caused by cable viewing. Declare that cable TV viewing causes "seventeen percent of growth in autism". Assert that TV watching due to precipitation causes "just under forty percent..of diagnoses".

Then tell everyone else how to proceed. Describe some experiments that scientists can do. Make policy recommendations for the American Academy of Pediatricians (who already recommend that kids under 2 don't watch TV).

10.) Publish this paper on the internet ahead of extensive peer review, or any peer review whatsoever, for that matter. Disperse the information widely in the popular press. People will be intrigued, interested, scared, nervous, possibly guilt-ridden. Sure, they may note the problems with the research, or they may say that your conclusions were obvious, but some will also say, perhaps jealously -- 'we said that first!'.

Tsk, tsk, children, children, stop squabbling or no more TV rights! As for you, dear researcher, you've probably got dozens of parents toting TVs out of their houses. And you got credit for your intrepid ideas! True or not.

follow us on twitter!