interactions- ANCOVA

[Nomi-2]

Hello all,
I'm running an ANCOVA and would like to examine interactions between the
independent variable and the covariates. Is it ok to include only
interactions in the model, without the main effects?
Thank!
Nomi



Nomi Werbeloff, PhD
Department of Psychiatry
Sheba Medical Center

=====================
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

[Bruce Weaver]

Nomi-2 wrote

Hello all,
I'm running an ANCOVA and would like to examine interactions between the
independent variable and the covariates. Is it ok to include only
interactions in the model, without the main effects?
Thank!
Nomi



Nomi Werbeloff, PhD
Department of Psychiatry
Sheba Medical Center

The usual practice is to include all lower order terms that comprise a higher order term.  I.e., if you include a two-way interaction, include both main effects; if you include a 3-way interaction, include all 3 main effects, and all 3 of the two-way interactions.  Etc.  

Just a note on terminology:  When you include the covariate x group interaction, your model is no longer an ANCOVA, strictly speaking.  ANCOVA is usually reserved for a linear model with some combination of continuous and categorical explanatory variables, but with main effects only.  E.g., one-way ANCOVA has one categorical variable and one continuous variable (or covariate).  So you might want to call your model a regression model, or a general linear model.