Health
014 Close encounters of the first and second kind
Type 2 error is like missing the needle in the haystack.
Posted July 16, 2010
To recap: Type 1 error consists of erroneously jumping to the conclusion that a cause-and-effect relationship exists ("umbrellas cause rain"). Type 2 error consists of overlooking a cause-and-effect relationship that actually does exist (failing to find the needle in the haystack).
One huge source of Type 1 error comes from misuse of the Vaccine Adverse Effects Reporting System (VAERS) data. Remember Mister Shady Broker, who only reported on 50 out of 800 subjects (the ones where he guessed correctly)? It's the same here. VAERS does not collect information on children who receive immunization without apparent problems, nor does it collect information on children who develop ASD even though they never received an immunization in their life. And it does not collect information about children who regress at other times in their lives than following an immunization. Nonetheless, there have been a couple of very prominent opinion-setters who have misused VAERS data in an attempt to prove that immunizations trigger autistic regression.
We also know from studies of home movies that autistic features can be present long before parents take notice. (See Chapter 4 of my book for references.) So, claims by parents that their child was "perfectly normal" before the immunization cannot be accepted uncritically. This does not diminish the concern that an immunization may have caused further impairment in a child with a pre-existing condition, but that is a separate question.
What about Type 2 error?
For any given risk factor and bad outcome, want to know if a particular subject has been exposed to the potential risk factor or not, and we want to know if the subject gets the disease or not. We show this in what's known as a "two by two table":
![](https://cdn2.psychologytoday.com/sites/default/files/styles/image-article_inline_full/public/blogs/43103/2010/07/45419-31396.jpg?itok=tFaUirho)
Population-wide studies (Cohort studies) go "across" the table. To run a Cohort study, we have to know the exposure status and the outcome for everyone in the population (for example, "all children born in the USA in 2005," or "all children under the age of 5 currently residing in New York City"). Cohort studies have proven that the population-wide risk of ASD, and regressive ASD, have not budged, when the MMR has been introduced or withdrawn in various industrialized nations over the past 20 years.
![](https://cdn2.psychologytoday.com/sites/default/files/styles/article-inline-half/public/blogs/43103/2010/07/45419-31397.jpg?itok=MblusLoJ)
Population-wide studies also show that the prevalence of ASD rose after thimerosal was removed from most shots in the early 90s. Here's the same figure you saw in Post 008, where we discussed changes in Federal education law and revisions of the DSM, only this version includes a star to show when thimerosal came out of most shots. From these data, we might argue that thimerosal prevents ASD. No one really thinks that, but at the very least, population studies seem to pour cold water on the argument that thimerosal causes ASD.
There is a catch, however: Cohort studies are useful in assessing the risk of disease following exposure to a given risk factor - as long as the risk factor and the disease are both common enough that we'll capture a large number of exposed and/or affected individuals in our cohort. But what if the risk factor or the disorder are rare? If either of these are the case, then we might not pick up enough children - even in a large cohort - to be able to detect the problem. (This is the PKU / diet soda dilemma: diet soda is common enough, but PKU is rare.) Is there a group of children who are uniquely at risk for autistic regression due to immunization? Hannah Poling, who deteriorated after receiving an immunization, has an underlying metabolic disorder. Did that disorder render her uniquely vulnerable to the stress of immunization? Perhaps. Might she have deteriorated anyway - perhaps after some "harmless" event like a minor febrile illness (very common in children with metabolic disorders)? Perhaps. And the big question: For children like Hannah, which risk is greater: getting the immunizations, or becoming ill with the diseases themselves? We have no idea. And are there other potential risk factors (say, immunologic conditions)? We don't know.
This is where Case-Control methodology come in handy. Rather than starting with an entire population of children and going "across" the 2x2 table, Case-Control studies start in the lower lefthand corner, with cases of identified disease (for example, all children with regressive ASD at University Hospital). "Cases" are then compared to suitable controls (we need 3 control groups: children who were on the spectrum from birth, children with other disabilities, and normally-developing children), looking for metabolic or immunologic differences between the "cases" and the controls. It would also be useful to look within the "Cases" group at the onset of regression compared to the timing of receipt of immunizations as well as other risk factors, such as colds, diarrheal illness, etc.) These data would still not prove cause-and-effect, but at least they would narrow down the search to a smaller portion of the haystack than the cohort data, which are all we have to go on at the moment.
At the moment, however, we don't have those data. So, what's a parent supposed to do? More on that next time.