1
$\begingroup$

I have tested gender difference on the relationship between Variable A and B using linear model (GLM command in SPSS).

I have found that the interaction of A (covariate/independent variable) with gender (factor) on B (dependent variable) is statistically significant. This means that the relationship between A and B differ based on whether one is a male or a female.

I am not familiar with GLM procedure, so I am wondering what is the next step.

Question 1: Is there a way to find out whether the difference in the relationship between A and B is greater for males over females or vice versa. (i.e. how do I interpret the difference?)

My preference is to just stop at this stage and look at the correlation coefficients between A and B for males and females separately and based on the GLM result, discuss the two coefficients. e.g. if r for males is 0.64 and r for females is 0.72, then I just conclude, based on the GLM, that the correlation is stronger for females over males.

Question 2: Is it possible to combine GLM and correlation (Pearson) in the above manner?

Your thoughts will help me better explain the findings to my target audience!

$\endgroup$
  • $\begingroup$ How is your outcome variable distributed? Try graphing outcome vs. predictor. Compare the slope of the predictor for males and females. $\endgroup$ – Michael Bishop Nov 16 '11 at 15:00
  • $\begingroup$ Yes, perhaps a graphic illustration will be better. Can I do this in SPSS ver 17 $\endgroup$ – Adhesh Josh Nov 17 '11 at 22:21
2
$\begingroup$

Let Y be your DV, X be your independent continuous variable (covariate), G be grouping factor (gender, 1=male, 2=female). You want to know how males differ from females in dependency of Y on X.

GLM Y BY G WITH X /DESIGN= X G X*G /PRINT= PARAM.

Look in Parameter estimates table.

  • Intercept = intercept in regression of Y on X for females.
  • Coefficient for G=males = intercept in regression of Y on X for males minus intercept in regression of Y on X for females.
  • Coefficient for X = regression coefficient of Y on X for females.
  • Coefficient for interaction G=males * X = regression coefficient of Y on X for males minus regression coefficient of Y on X for females. Thus, if this value is positive then b for males is higher than b for females; and its significance is the significance of this difference in b.
$\endgroup$
  • $\begingroup$ Thanks Does Pearson r have any merit in this approach. (I think the answer is no, based on your response!) $\endgroup$ – Adhesh Josh Nov 15 '11 at 20:13
  • 1
    $\begingroup$ In your simple situation (one X) consulting with r is justified. In general situation (several Xs) the answer is no. $\endgroup$ – ttnphns Nov 16 '11 at 6:05

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.