Hello Anh,

The presence of interaction among factors in anova/manova indicates that marginal means for the main effects do not correspond to the differences between levels of one factor at specific levels of the other factor. Hence, the appropriate follow-up is often referred to as simple effects tests (e.g., each level of factor A can be compared, as in a one-way test first at level 1 of factor B, then at level 2 of factor B, etc.). These tests can be elicited in spss, but you'll need to use syntax, not the dialog box. Here's a good resource on how to do just that: https://www.ibm.com/support/pages/significant-interaction-anova-how-obtain-simple-effects-test

It can be the case that simple effects show the same general trend as does a main effect, except for the magnitude of the differences observed. So, even when a trend is congruent, it still differs (in magnitude) enough that the best information about what's going on comes from the simple effects tests (following up the interaction), and not from the main effects tests.

Good luck with your work.

Hello Anh,

The presence of interaction among factors in anova/manova indicates that marginal means for the main effects do not correspond to the differences between levels of one factor at specific levels of the other factor. Hence, the appropriate follow-up is often referred to as simple effects tests (e.g., each level of factor A can be compared, as in a one-way test first at level 1 of factor B, then at level 2 of factor B, etc.). These tests can be elicited in spss, but you'll need to use syntax, not the dialog box. Here's a good resource on how to do just that: https://www.ibm.com/support/pages/significant-interaction-anova-how-obtain-simple-effects-test

It can be the case that simple effects show the same general trend as does a main effect, except for the magnitude of the differences observed. So, even when a trend is congruent, it still differs (in magnitude) enough that the best information about what's going on comes from the simple effects tests (following up the interaction), and not from the main effects tests.

Good luck with your work.

Hi Dr. Morse,

Thank you for the good reference source. I tried it and I worked. I can do now the pairwise comparison for interaction in SPSS

My data have

• DV1 with 2 levels (Male, Female)
• DV2 with 3 levels (drug A, drug B, drug C)

So I should have 2*3 = 6 combinations. From 6 combinations, I should have 6C2 = 15 pairwise comparisons. However, if I do the /EMMEANS, two times, I can get only 9 comparisons. As shown in the image, we will miss some comparisons (not highlighted row) such as:

Female+drugA vs. Male+drugB

Female+drugA vs. Male+drugC

/EMMEANS = TABLES(gender*drug) COMPARE (gender) adj (LSD)

/EMMEANS =TABLES(drug*gender) COMPARE(drug) adj (LSD)

Do you know how to overcome these issues SPSS?

Should we report only 9 comparisons instead of 15 comparisons?

Hello again, Anh,

The simple effects tests for two factors (IVs), one with 2 levels and one with 3 levels would total 5: (a) Sex (M vs. F) at drug A only; (b) Sex at drug B only; and (c) sex at drug C only; (d) Drug (A vs. B vs. C) at Males only; and (e) Drug at Females only.

Note that the first 3 are pairwise tests, because there are only two levels (of sex) being compared. The last 2, however, are tests of 3 drugs for equality. This syntax doesn't give all possible pairwise comparisons. Generally, you would only pursue a follow-up of the simple effects tests of drug levels if one or both were significant.

However, you'll have expected marginal means for the relevant combinations, with their standard errors, so it you feel the need to run pairwise tests among the drugs (e.g., drug A vs. drug C when looking only at males), you have the information you'll need to set up those t-tests.

Good luck with your work.