Statistics and Insomnia

Some years ago, I saw a snippet on the ABC science show Catalyst about insomnia – in particular, the flavour of insomnia where a person has trouble falling asleep at all. They reported on a trial study investigating the effectiveness of a tortuous new treatment for chronic insomnia. (You can find the published research here: Click here to go to insomnia article.)

The usual way to cure insomnia is to retrain your brain and your body to associate the bed with sleep rather than wakefulness. What they recommend is to only go to sleep when you're really really tired, and if you don't fall asleep in quarter of an hour, to get up and go to some other room until you feel tired enough to go to sleep again. Eventually, you'll fall asleep in bed. Then you try again tomorrow night, and the next night, and the next night... Usually it takes a month.

The big problem with it is that people just don't have the stamina to put themselves through all this for four weeks. Here's where the radical treatment comes in: you compress the month of practice into 24 hours. The poor participant is put in a windowless room and practises going to sleep, and when they finally do fall asleep, they only get four minutes to sleep before they are woken up to try and fall asleep again. In this way you fit a month's worth of falling-to-sleep practice in one day. Imagine how desperate you would have to be to sign up for this sort of thing!

Recently, it occurred to me that there are a lot of other skills that take a lot of practice to learn and this practice is usually drawn out over such a long period that people just don't get through it all. One of these is statistics -- in particular, the process of deciding which statistical procedures should be used to analyse your data.

In your standard stats course, the approach to teaching students to make decisions is to get them to do a project. This gives them practice at making decisions a grand total of once. And so students need a whole degree's worth of projects, and probably years of working as a statistician, to learn how to make decisions. Hence, very few people ever get very good at making them. It's just like the poor insomniac trying to cure their own insomnia once a night.

But what if you could, like the new insomnia treatment, compress all that practice into a short amount of time? What if you could pick out just the part where you make the decision and get students to make a lot of decisions all at once? Then they might get the necessary experience rather more quickly than the standard approach.

I tried it out last year with the med students. I gave them a quick lecture about how you make the decision of which hypothesis test to use. Then, I gave them 30 research questions and got them to make a decision for each one. They seemed to get the idea of how it worked. So much so that they actually had intelligent questions to ask afterwards!

I'm trying again this year, only this time the Medical School is letting me help design the whole stats teaching program, not just one lecture. Here's hoping that a little bit of torture for a short time can alleviate months of pain later...

Theis comment was left on the original blog post: 

Richard Knowling 27 January 2012:
This is an awesome idea David! I only wish Mike Roberts had still been alive to hear about it!