I’ve finally taken the courage to publicly ask whether psychologists (or empirical scientists in general) at all should continue to teach statistics. And I pinged some of the most prominent statistically skilled psychologists in the meta-science movement. See the Twitter thread.
My original tweet:
> Provocative question to all psychologists teaching statistics: Shouldn’t you just stop & let mathematically trained statisticians take over? Empirical research shows that psych’s teaching statistics don’t know statistics well enough. So how can still teaching it be justified?
Perhaps unsurprisingly, this turned out to be a sore topic. And this is where I really want this forum to take over; having these types of discussions on Twitter is so frustrating; it’s so difficult/impossible to be polite enough while also trying to explain thoroughly.
So, trying to move the discussion here now and we’ll see how that works. And I’ll continue with one possible (provocative) interpretation and implementation of the responses I received from meta-scientists: If we as psychologists can teach statistics just as well (or even better, as some suggested) than mathematical statisticians, is mathematical statistics a superfluous scientific discipline?
Several people on Twitter express the view that it’s better if students in the empirical sciences get to learn how to practice statistics directly - without understanding the theory behind it - because whenever these students try to learn statistics from the mathematical statisticians they simply don’t understand anything at all. See e.g. comments by Priolo Daniel, Katie Corker, and Pam Davis-Kean.
I have great respect for everyone involved in this discussion, most or all of them are deeply involved in the movement for replicable science and I know they really aim for exactly the same thing as I do: a more rigorous science. However, I do find this particular view remarkable.
Mathematical statistics is an own scientific discipline. We use knowledge from their discipline in order to practice research in our own disciplines. It should be a warning sign that our students (and thus also our active researchers?) can’t understand the introductory statistical lessons when taught by mathematical statisticians. Instead do we need people from our own field - without the deeper understanding - to explain a more shallow version. With that shallow knowledge do we then go directly to practicing our empirical science. I don’t understand how this can be considered preferable to the alternatives.
Here is one alternative: Psychologists don’t learn statistics at all and are instead taught that they don’t know statistics and that they need to collaborate with mathematical statisticians in their research. Instead of themselves being trained in employing statistics are they trained in understanding that there are many different statistical methods and theories out there and that they are continuously developing, just like theories and methods in our own fields. And they are being taught where and what to look for when searching collaborations with mathematical statisticians for different types of projects.
I think one of the problems with your proposition – psychologists don’t learn stats, they must collaborate with statisticians – is that knowing about statistical tools makes you think differently about research design. I see two possible consequences: (i) a psychologist develops a design whose data are difficult to analyze even with appropriate statistical tools (e.g., because inadequately powered); or (ii) a psychologists plays it safe and always runs factorial designs that can be analyzed with simple stats (and I have seen this since the beginning of my academic career). We would still need a conceptual overview of statistical techniques, what they are good for, and (ideally) who to ask for help within our university. Or am I missing something?
My understanding is that mathematical statisticians want to be and should be involved already on the planning stage. It’s too late to do anything after the data has already been collected. Many psychologists claim that statisticians don’t know anything about study design etc. That may be true, but they do in fact know how data must be collected in order to employ e.g. frequentist statistics. The collaboration with mathematical statisticians should thus not be limited to the statistical analyses. If they are, I think much of the current problems will persist: underpowered studies, not ideal post hoc solutions to make something out of data that violates basic assumptions, etc.
I do not suggest that psychologists should not learn statistics. I suggest that we are properly trained. That includes being trained to understand that most of us do NOT have enough training to start practicing statistics independently, and definitely not enough knowledge to teach it. Some of us may know enough, but most of us don’t.
During my undergraduate years we had lectures with a clinical psychologist. He ended one of his lectures by saying that it may feel like we are now ready to diagnose people we meet or otherwise assist them with psychological questions they might ask. And he bluntly continued: “be assured you are NOT ready, so PLEASE refrain from doing that, even if people ask you to do it”.
I’m thinking a similar warning would be relevant concerning statistics: “you may FEEL ready to employ it now and teach it to others, but be assured you are NOT”.
Psychologists know enough about psychology to give out proper warnings about how it is employed. But we don’t know statistics well enough to do the same; we practice and teach it with too much confidence. That strong confidence is possible to have only when you don’t know a subject well enough.
Based on the reasonings on this page, i fully agree that specialists from each discipline should collaborate in research endeavors. That means, yes, statistics should be taught by its specialists.
My goodness, I am a mathematical statistician, and proud of it. Science needs good mathematical statisticians. But not necessarily as teachers of applied statistics. That needs good applied statisticians. The two are not the same. What it doesn’t need is people of either category who are not aware of their limitations with respect to the other category.
Statistical science, and statistical thinking, are not part of mathematical statistics. Highly trained mathematicians are not necessarily any good at all in analysing data and in particular understanding how to quantify or somehow measure the reliability of what has been learnt.
