Following my question here, I will be most grateful for further assistance to clear my confusion.

The confusion is because I have read in many places that when you use ANCOVA in GLM (in SPSS), the effect of the covariate is controlled. I am unsure why it should be controlled in my case below (unless I am wrongly interpreting what a 'covariate' means). Or using the wrong test!

Question 1

What is the best method to analyse the following problem in SPSS?

I am looking at how the relationship between A and B behaves or differs based on gender. In other words, how does gender impact on the relationship between A and B. A and B are interval variables. A is the independent and B is the dependent variable. Gender, of course, is a categorical variable. (The nature of the variables, as I understand it, makes factorial ANOVA inappropriate, leading to a choice of ANCOVA for this example.)

My understanding is that B goes in the dependent variable box, gender goes in the factor box and A goes in the covariate box in SPSS GLM (univariate).

Question 2

*****What does the interaction mean?****

I understand I can get an interaction between A and gender, which then means that A and gender have a joint effect on B. If this is significant, does it mean that gender affects the relationship between A and B.

Question 3

Should I look at the effect of A and gender separately or only focus on the interaction between A and gender?

I am very new to GLM and it is fascinating (if I can get it right!).

  • 2
    $\begingroup$ In addition to any answers below, I find Faraway's Practical Regression and ANOVA using R to be a concise and helpful reference for regression basics (available here: cran.r-project.org/other-docs.html). You can either go through the examples in R, or translate them to SPSS. I get the impression that your overarching question is, "How do I perform and interpret regression analyses". A good question, but one that most here studied in semester-long university courses, and some have spent careers studying. $\endgroup$ – jthetzel Nov 19 '11 at 17:36

Be sure that you added the interaction term (and the main effects) in the Model subdialog and parameter estimates in the Options subdialog. With just one covariate and one dichotomous categorical variable, you are just estimating two separate regression lines. If there is no interaction term, the lines are parallel. the gender=0 * A and gender=1*A terms tell you the two slopes (assuming gender is coded 0/1) The Test of Between-Subject effects tells you whether the interaction is significant.

If you haven't already done so, look in Case Studies>GLM for a short tutorial on how it works and how to interpret the output. That's no substitute for a real textbook, but it's a good quick start.


I think @jthetzel's comment is dead on. What you most need is a more solid basic understanding of the general linear model, and that's more than can be provided by answers on CV. Faraway may be helpful there. In addition, it's been a long time since I've worked with SPSS, so I can't help with #1 from memory. But explaining what an interaction is may help with #2 & #3.

Interactions seem mysterious to people, but there is no magic there. An interaction just means the effect of one variable depends on the other variable. In your example, there is one relationship (i.e., slope) between A and B for men, and a different relationship (i.e., slope) between A and B for women. To understand your results more clearly, and to present them clearly to others, you look at the effect of A on B twice, once looking only at the men, and then again looking only at the women. For example, you might make a scatterplot of A and B with different symbols for men and women and overlay two different lines in different colors to display the two different relationships.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.