I presume that for the Model Effects Table, you are looking at
a 6 degree of freedom test, and you see that it is "significant."
Given that overall result -- it bothers you that not even one of
the 6 tests provided for separate parameters is also "significant."
What this means is that the particular set of parameterizations
for the interactions does not include any that is so extreme.
This is just like any regression or ANOVA result, where a test on
several variables taken together is significant, but the separate
tests are not: Even if the contrasts are orthogonal, the total
sums of squares accounted for may be significant while the separate
ones are not. If the contrasts are correlated, this result is even
more likely. (I don't know whether your parameter estimates are
correlated or not; I suspect that they are.)
However, it is certainly true that there exists some arbitrary, 1 d.f.
contrast that would be significant. Off hand, I don't know how
those contrasts for "parameter estimates" are constructed, but
I suspect that you will see a different set if you happen to switch
around the numerical labels for the categorical variables.
--
Rich Ulrich
> Date: Fri, 28 Feb 2014 06:19:29 -0800
> From:
[hidden email]> Subject: GENLIN problem with model effects and parameter estimates
> To:
[hidden email]>
> hello all,
> this might be too trivial for some, but I was wondering whether someone
> could enlighten me on the following:
> I'm running a Poisson regression under the Generalized Linear Model, and I
> have 3 categorical variables, e.g. A with 4 levels, B with 2 levels and C
> with 3 levels. My model includes the main effects and up to 3-way
> interactions of A,B,C. My problem is that for A*B*C none of the parameter
> estimates in the respective Table are significant for A*B*C, but the term
> A*B*C is significant in the Tests of Model Effects Table. Could anyone
> explain in simple terms what this means please? what is the test done for
> the parameter estimates Table and what is the test done for the Model
> Effects Table?
>
> Thank you all in advance
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/GENLIN-problem-with-model-effects-and-parameter-estimates-tp5724670.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> 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
