When X is nominal (a grouping variable), it should be coded, for example, using dummy coding. With 3 groups, you would need to create 2 coding variables (X1 and X2, e.g., Group 1 vs. control and Group 2 vs. control). For a moderation analysis, you would additionally have to create two product variables (X1*W and X2*W) to reflect the potential interaction between X and W and include all 4 variables (and W) in the regression equation.

Hello Antonio,

Yes, you can conduct moderation analysis with a categorical X variable. If genuinely nominal, you'd have to re-express it as two dummy variates to carry the full information associated with differences in Y due to X.

But the basic logic is exactly the same:

Step 1: Regress Y on dummy1 and dummy2 (from the X variable) and W scores.

Step 2: Regress Y on dummy1, dummy2, W, and the two interactions (W*dummy1, W*dummy2).

If the increase in R-squared from step 1 to step 2 is noteworthy (based on whatever criterion you impose), and at least one of the two interaction terms is significantly different from zero, then you have evidence for moderation in the relationship.

Good luck with your work.

Hello Antonio Leite Coutinho Neto. The question appears to be whether the PROCESS macro will work when X is categorical. I do not use PROCESS, and therefore do not know the answer to that. But I do know you can obtain the results you want without using PROCESS. If I understood your model correctly, the following example (used with your data rather than the sample data file I used) should produce the output you want. HTH.

NEW FILE.

DATASET CLOSE ALL.

* Modify path on next line to point to the sample data files that come with SPSS.

GET FILE = "C:\SPSSdata\survey_sample.sav".

*

X (ordinal) = 0, 1 or 2 - which are groups (control, group 1 and group 2)

Y (continuous)

W (continuous; conditioning values are 16, 50 and 84 percentile)

.

* Generate some variables with the desired names.

RECODE degree (0=0) (1=1) (2 3 4 = 2) INTO X.

FORMATS X (F1).

COMPUTE Y = speduc.

COMPUTE W = age.

DESCRIPTIVES X Y W.

* NOTE: You want the code from this point on.

FREQUENCIES VARIABLES=W

/FORMAT=NOTABLE

/PERCENTILES=16.0 50.0 84.0

/ORDER=ANALYSIS.

* Use UNIANOVA to estimate the model, and use the EMMEANS

* sub-command to get the fitted values of Y for the 3 groups at

* selected values of W--i.e., the 28th, 50th & 84th percentiles.

* NOTE: You must plug in the values for your own data.

UNIANOVA Y BY X WITH W

/METHOD=SSTYPE(3)

/INTERCEPT=INCLUDE

/EMMEANS=TABLES(X) WITH(W=28) COMPARE ADJ(LSD)

/EMMEANS=TABLES(X) WITH(W=42) COMPARE ADJ(LSD)

/EMMEANS=TABLES(X) WITH(W=65) COMPARE ADJ(LSD)

/CRITERIA=ALPHA(0.05)

/DESIGN=W X W*X.