Hi,
I'm hoping that someone here can answer a simple question for me. I'm
teaching university psychology students methodology (up to postgraduate
level), and not only do I want to show them the calculations that are
being done (easy), but also get them to see why the calculations are being
done that way.
First, the easy bit. If I run a 2-way ANOVA with one random and one fixed
factor and replicates, I get three F ratios. Each F ratio has, of course,
a mean-square error term
F1(Fixed factor): MSError = mean square of the interaction term
F2(Random factor): MSError = mean square of the interaction term
F1XF2 interaction: MSError = mean square of the replicates
So far so good: it seems that SPSS uses the unconstrained model to
determine the expected mean squares, hence the interaction term for the
same MSE term for both main effects (most sources I've read/come across
use the constrained model in which the random factor has a MSE from the
replicates, but at least I can explain why)
Now my problem. When calculating LSD post-hocs, I was under the impression
that you should used the MSError associated with the F ratio (this makes
perfect sense to me). That is, in my example above, for running LSD on the
main effects you would use the MSE of the main effect (the MS of the
interaction term).
However, when I run this on SPSS 19, the post-hoc analysis uses the mean
square associated with the replicates (i.e. what I would expect if all the
factors were fixed), even though the F ratio is calculated using the MS of
the interaction. Can someone explain why the change in the error term?
Thanks in advance
Mike
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
