Beyond ANOVA: from repeated measures to multilevel models

I have just finished giving a talk on repeated measures ANOVA and multilevel models at the PsyPAG Postgraduate Mathematics, Statistics and Computing workshop. I promised I would upload a link to my presentation. Here it is:




I managed to correct a couple of minor errors - though the transfer online seems to have added several more (though these disappear - or rather the errors mutate - if I swap browsers). I'll add further links and corrections if I get time.

This is run every two years or so in conjunction with the Mathematics, Statistics and Computing section of the British Psychological Society (BPS). The 2008 workshop was at Nottingham Trent University. The full schedule is here:

http://www.psypag.co.uk/msprogramme.html

The statistics clinic was fun - as was my confrontation with Mark Shevlin (of which more later) - but I missed the signed statistics text book raffle (proceeds to charity). I was very interested in John Reidy's talk on moderated multiple regression. Unlike my talk he covered mediation as well, but possibly covered moderation in less detail as a consequence. I think anyone who found John's talk useful will probably want to look at my moderated multiple regression blog entry if only for the link to Kris Preacher's web resources.

Mark Shevlin and I had a (friendly) dispute over the relative merits of SEM and multilevel models and MPLUS versus MLwiN for the latter. For experimental psychologists I think multilevel models are definitely the way to go - it being easy to extend classical experimental designs to multilevel models. For non-experimentalists who will have messier studies to analyze SEM is probably more useful. Although all multilevel models are probably just SEM in disguise I think SEM is harder to pick up and (at least until recently) was not great for repeated measures and clustered data. As for software ... because MLwiN is free for UK academic users I think it would be my top tip for anyone in the UK learning multilevel modeling (with R a very close runner up).

Postscript. The Centre for Multilevel Modeling have just launched their online multilevel modeling course. At first glance it looks very useful, as it starts with simpler statistical techniques (e.g., multiple regression) and builds up to multilevel models. I'm hoping to audit all the modules at some point ... They are released under a creative commons license so it should be possible to use them in Education.