Hi all, many further thanks for the help I recently received on setting
up data for repeated measures analysis, thanks to the help I got here I
have been able to do that analysis.
A further question: it seems one could come up with justifications for
several ways to compute outliers in a dataset. For example with the DV
"reaction time," one could exclude all measurements > 2 STDs away from
mean. But, the overall mean of all measures? or just for each level of
one of the factors? Or considering both factors, as in, calculate
outliers for each cell?
In the latter case, one would be decreasing within-cell variance
relative to overall variance, and thus, increasing the F statistic
calculated between cells. ...right?
But if one calculates the outliers from the entire dataset, one
decreases overall variance relative to between-group or between-cell
variance, thus artificially deflating the F value.
What approach should one take? Reaction time outliers are most often due
to distraction or a missed cue, so that it is usually not necessary to
individually justify excluding each single outlier, and a way to "en
masse" delete the offenders without harming the statistical model is
sought.
Many Thanks
Jim
--
WATCH OUT for spam block
To reply by email, remove the spam block
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at
http://jse.stat.ncsu.edu/
=================================================================
