Richard Ulrich wrote:
> Taking the logs of the durations is an obvious thing.  I believe it
> does not take special justification because a statistician should have
> recommended that before you began.

Yes - its fairly common in this field (as is taking the medians of
multiple trials).

> As an alternative, re-stating the durations as time-per-word seems to
> me to be a realistic version of the outcome, which also might
> avoid interactions and heterogeneity of variances.

No. The appropriate correction in the field is time/syllable. This tends
to produce cleaner results than time per word. There is a slight bias in
that the time to read a sentence is:

c + aS

(S is number of syllables, c is a constant related to processes that
don't depend on sentence length). The correction will over-correct short words.

An alternative is to include sentence length in syllables (or expected
mean sentence length if the stimuli are randomized) as a covariate. This
will produce an unbalanced ANOVA that can be analysed in standard
packages with a little tweaking:

Lorch, R. F., & Myers, J. L. (1990). Regression analyses of repeated
measures data in cognitive research. Journal of Experimental Psychology:
Learning, Memory and Cognition, 16, 149-157.

Thom
.
.
================================================================
This list will soon be replaced by the new list EDSTAT-L at Penn
State.  Please subscribe to the new list using the web interface
at http://lists.psu.edu/archives/edstat-l.html.
================================================================

Reply via email to