Miguel Verdu wrote:
>
> Posted also to comp.soft-sys.stat.spss where the same question appeared
> (and nobody answered).
>
> Hello.
>
> This is an output of GLM from SPSS 9.0 where the dependent variable
> FLOR (log transformed)
>
> was analysed by crossing 2 levels of the FIXED factor SEX with 2
> levels of the RANDOM factor
>
> POP. The F for POP has been obtained by dividing MS POP/MS (SEX*POP)
> (0.06349/0.199=0.319).
>
> I think this is wrong because the F could be obtained by dividing MS
> POP/MS error
>
> (0.06349/0.05737=1.10). Can anybody tell me if I am right or wrong?
>
> Dependent variable: LOGFLOR
>
> Source TypeIII SS df
> MS F Sig.
>
>
>-----------------------------------------------------------------------------------------------------
>
> Intersección Hipótesis 694.987 1
> 694.987 10945.783 .006
> Error 6.349E-02 1
> 6.349E-02a
> SEX Hipótesis .449 1
> .449 2.256 .374
> Error .199 1
> .199b
> POP Hipótesis 6.349E-02 1
> 6.349E-02 .319 .673
> Error .199 1
> .199b
> SEX * POP Hipótesis .199 1
> .199 3.467 .066
> Error 4.360 76
> 5.737E-02c
>
>
>-----------------------------------------------------------------------------------------------------
>
> a MS(POP)
> b MS(SEX * POP)
> c MS(Error)
A colleague wrote to SPSS a month or two ago about this issue.
Following is the response that he received from them. You can also look
in Searle (1971) Linear Models (Sec. 9.7, pp. 400-404) for a discussion
of this.
Response from SPSS:
There are two different sets of assumptions that are commonly made
concerning the status of interaction effects in models with random
components. They can be assumed to be fixed effects restricted to sum to
0
over the levels of the fixed effects within each level of the random
effects, or they can be assumed to be random variables. The error term
assignment you want comes from the first set of assumptions. The error
terms
in SPSS are based on the latter, which is the more commonly used set of
assumptions for general handling of potentially unbalanced data (BMDP
and
SAS, for example, also do it this way).
David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.
