There seems to be a serious bug in spss 7.5 glm. When I run the syntax below, spss version 6.1 (manova) and version 7.5 (glm) give different output. I believe the 7.5 output is wrong. The df for subject is 6 in spss 7.5 vs 7 in spss 6.1. Also, the degrees of freedom for the estimated effects don't add to the total Corrected Model df, in version 7.5.

The data below are for a 2 x 2 cross-over design. Outcome variable is the time required for a task. Each subject does the task twice, with tool 1 and with tool 2. Question is whether an 'experts' do better with tool 1 and non-experts better with tool 2 (interaction tool by expert).

=== cut here === data list free /subj expert tool order time. begin data. 1 1 1 1 6.9 1 1 2 2 5.0 2 1 1 2 4.7 2 1 2 1 5.1 3 1 1 1 6.6 3 1 2 2 7.5 4 1 1 2 6.2 4 1 2 1 5.7 5 2 1 1 4.5 5 2 2 2 5.4 6 2 1 2 6.6 6 2 2 1 7.0 7 2 1 1 6.8 7 2 2 2 5.6 8 2 1 2 4.4 8 2 2 1 3.1 end data.

; for SPSS 6.1 : MANOVA time BY tool(1 2) subj(1 8) order(1 2) expert(1 2) /NOPRINT PARAM(ESTIM) /METHOD=UNIQUE /ERROR WITHIN+RESIDUAL /ANALYSIS time /DESIGN subj order tool expert*tool .

; in SPSS 7.5 use : GLM time BY tool subj order expert /METHOD = SSTYPE(3) /INTERCEPT = INCLUDE /CRITERIA = ALPHA(.05) /DESIGN = subj order tool expert*tool .

;==== end of syntax ====

Thanks, Arnold Kester