# ANCOVA: ordinal covariate on SPSS

I'm trying to run an ANCOVA on SPSS with 3 variables: Gender, score 1, score 2.

I want to control for the possibility that gender has an effect the scores, in order to isolate just the relationship between score 1 and score 2.

However, I cannot enter gender as a covariate on SPSS as, by definition, it assumes covariates must be scalars.

So, how can I run my ANCOVA and make sure that gender isn't affecting my 2 other variables?

• Calling a variable a "covariate" is treated as meaningful in SPSS in terms of how you can use the wizards to call for your analysis, but it doesn't actually have an underlying statistical meaning. Just code Gender as a column of 1s & 0s (using whichever you prefer for "female"), & run a multiple regression w/ your scores & your dummy code. – gung - Reinstate Monica Apr 8 '14 at 0:43
• @gung is right. And also note that a binary variable is of its own kind. It can serve as factor or as a covariate, giving the same results. – ttnphns Apr 8 '14 at 2:28

If you want to retain the nominal coding of the Gender variable, you can include it as a fixed factor, then go in "Model...", then select "Custom" (instead of "Full factorial"), and enter all main effects, and the 2-way interaction between score1 and score2, but not the interaction between Gender and either score1 and score2 (i.e., enter only the main effect for Gender).

Alternatively (and this is what I would suggest), you can recode your gender variable into dummy codes or effects code (or any other two values). For example, you could call your variable Female, and code males as 0, and females as 1, and set it as a "scale" variable. Then you could include it in the "Covariates" section.

• Thank you, that's very helpful. If I were to go with the second option, would that still tell me the exact same thing as if I had left gender ordinal? (i.e. 'scalafying' gender would not affect the result of the analysis, only my ability to accomplish it, right?)? – SK1 Apr 8 '14 at 0:33
• Right. It's going to give you the same results. (If you feel so inclined, you can try both and compare; but they'll give you the same results.) – Patrick Coulombe Apr 8 '14 at 0:35
• FYI, for the sake of interpreting the Gender effect, I would recommend coding one gender 0, and the other 1, which is by far the most common method for coding a categorical variable (at least in the social sciences). – Patrick Coulombe Apr 8 '14 at 0:38