[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [AUDITORY] stats use in psychology and hearing science

Dear list,

important topic! A related while probably less general problem is that data from repeated-measures / within-subjects designs (that is, when a subject is tested in more than one experimental condition) are sometimes not analyzed properly. I've seen quite a few papers in hearing science where the authors used an ANOVA for completely randomized / between-subjects design for repeated-measures data (you can spot this problem by examining the degrees-of freedom for the F-tests), or fail to use the appropriate repeated-measures ANOVA approach (e.g., apply no correction for the degrees-of-freedom when using a univariate approach).

Another point that should be considered is that while for between-subjects designs the general linear model is very robust against non-normality, this is unfortunately *not* the case for repeated-measures designs. Therefore, when analyzing non-normal data like error rates, percent correct or response times from a within-subjects design, the p-values you get from the rmANOVA could be either on the conservative or on the liberal side.

You can find an in-depth discussion of these issues in a recent paper from my lab (http://link.springer.com/article/10.3758%2Fs13428-012-0281-2).



PD Dr. Daniel Oberfeld-Twistel
Johannes Gutenberg - Universitaet Mainz
Department of Psychology
Experimental Psychology
Wallstrasse 3
55122 Mainz

Phone ++49 (0) 6131 39 39274
Fax   ++49 (0) 6131 39 39268