Saturday, July 26, 2008

Biased sample fallacy in stuttering research

As I said before, I think of stuttering as a disorder coming in two phases: the dormant phase until onset of stuttering, and the reactive from onset onwards. A question of the causes of stuttering must then we split into two questions: What causes onset of stuttering? What causes recovery or non-recovery of stuttering children? In stuttering (and I would not be surprised in other disorders, too), the majority of children grow out of stuttering. Therefore, if we study the sample of persistent children, it will be a sub-sample and very likely biased, because the whole sample is the children who start stuttering.

Here is an example. Imagine, you interview 10 lottery winners, and they all tell you that they had a dream, wrote down the numbers and won! You are inclined to attribute some cause to the dreams. However, this is a fallacy, because you need to look at the whole sample, the set of all lottery participants who had a dreams and wrote down the numbers. Of course, if you look at this sample, there is no effect at all!

How could this happen in stuttering? Well, I just wrote a post on Kate Watkins' talk in Paris where she tentatively explained a reduced activity in auditory regions with a reduced input from motor regions (likely due to some bad connectivity). So one could conclude that bad connectivity causes stuttering. However, the reduced activity might just be an effect of a biased sample, because the persistent kids could not recover from stuttering due to a lower-than-average connectivity to the auditory regions. To put bluntly, if you better integrate motor and hearing data, you are more likely to develop a good compensatory system to speak (more) fluently! I am not saying that this is the case, but I am saying that it might be the case and that we must be very careful with interpretation. The 2-phase disorder framework is useful here.

No comments: