Explanation of "covariate" in general linear model in SPSS I am completing a project for a client using general linear model (GLM command) in  SPSS/PASW (Ver. 17)
Basically, the project is designed to find out if factors such as gender and age affect the relationship between Variables A and B.
In this case, 


*

*Variable A is the independent variable (IV)  

*Variable B is the dependent variable (DV)

*Gender or age are the factors


In GLM command, IV goes in the covariate box and gender goes in the factor box.
Many textbooks I have consulted say one of two things (generally):


*

*GLM can be used to assess the joint significance of the predictors (A and gender in the above example) on a continuous outcome (B in the above example)

*GLM can be used to assess the significance of the factor (gender in the above example) on the outcome (B in the above example) by controlling for the effect of the covariate (A in the above example).


Obviously, these two different usages lead to different outcomes. I am interested in the joint significance.
I find the above a bit contradictory but I am not statistically trained. Can any one explain the key difference above (when the same test is in question i.e. IV in the covariate box and gender in the factor box).
 A: In my opinion the two statements are not really contradictory. 
The first statement regarding 'joint significance' simply states that GLM can be used to test the joint hypothesis that: "Gender's impact on B is different from 0 and A's impact on B is different from 0".
The second statement essentially indicates that GLM lets you assess if the individual predictors are significant after controlling for other factors. Thus, the statement implicitly states that using GLM we can do a simple hypothesis test such as: "Gender's impact on B is different from 0 provided everything else (in particular A in your case) is held constant".
A: Good question.  I interpret the "joint significance" line differently from @varty.  Usually, a "joint effect" means the same thing as a multiplicative effect or product effect  or interaction effect.  In other words, one could test whether gender's link with B differs depending on A.  Or whether A's link with B differs depending on gender.  SPSS's dialog boxes for GLM ("Univariate") allow you to specify such an interaction term.  Designing such a model will also allow you to see how A links with B, controlling for gender--assuming the interaction effect is nonsig.*  Just make sure you include the main (plain/vanilla) effects for each predictor as well as any interaction effect, or else you will get inaccurate results.  
*In which case you really should rerun the model without the interaction term, for best accuracy.
(If you want to see a basic, visual and nonmathematical summary of the difference between control and interaction, please see this short page of mine.  Sorry if this is a duplicate or self-referential citation, but I think it'd be useful.)
