In our reading group, we discussed a landmark paper of Paul Meehl’s, “Why summaries of research on psychological theories are often unintepretable” (1990). The paper ends with a very strong statement (p. 242), written by Meehl in italics for extra emphasis:
We should maturely and sophisticatedly accept the fact that some perfectly legitimate “empirical” scientific theories may not be strongly testable at a given time, and that it is neither good scientific strategy nor a legitimate use of the taxpayer’s dollar to pretend otherwise.
This statement should bring up all kinds of stages of grief in psychological researchers, including anger, denial, guilt, and depression. Are we really just wasting taxpayers’ money on studying things that are not studyable (yet)?
We sometimes have ideas, theories, or models, which cannot be tested given our current measurement devices. However, research is a process of incremental progress, and in order to make progress, we need to first understand if something works or not, and if not, why it doesn’t work. If we close our eyes towards all of the things that don’t work, we cannot progress. Even worse, if we find out that something doesn’t work, and don’t make any effort to publicise our results, other researchers are likely to get the same idea, at some point in time, and start using their resources in order to also find out that it doesn’t work.
To illustrate with a short example: For some reason or another, I decided to look at individual differences in the size of psycholinguistic marker effects. With the help of half a dozen colleagues, we have collected data from approximately 100 participants, tested individually in 1-hour sessions. The results so far suggest that this approach doesn’t work: there are no individual differences in psycholinguistic marker effects.
Was I the first one to find this out? Apparently not. When sharing my conclusion with some older colleagues, they said: “Well, I could have told you that. I have tried to use this approach for many years with the same results.” Could I have known this? Did I waste the time of my colleagues and the participants in pursuing something that everyone already knows? I think not. At least myself and my colleagues were unaware of any potential problems with this approach. And finding out that it doesn’t work opens interesting new questions: Why doesn’t it work? Does it work in some other populations? Can we make it work?
All of these questions are important, even if the answer is that there is no hope to make this approach work. However, in the current academic reward system, studying things that may never work is not a good strategy. If one wants publications, a better strategy is to drop a study like a hot potato once you realise that it will not give a significant result: throw it into your file drawer and move on to something else, something that will be more likely to give you a significant p-value somewhere. This is waste of taxpayer’s money.
Interesting. Very reason why I want to write a report of all failed attempts to make our statistical learning experiment.ReplyDelete
Thank you! I would be very interested to read about your statistical learning experiments. It seems to be one of those areas where there are a lot of experiments that didn't work and were never published.Delete