interpreting main effects with significant interaction term

[Kyra]

Hi,

I have a linear regression model Proc GLM where my outcome is medication dose prescribed

and variables being studied are patient gender, physician gender, their interarction term and there are few more variables.

Proc GLM gives significant p-value for the main effects and interaction both.

Medication dose dispensed= 232.65 +8.7 Male patient +23.3 female surgeon + 8.83 Male patient*female surgeon

Please let me know how to interpret association of patient gender and surgeon gender with continuous outcome( medication dispensed).

Thanks,

[SteveDenham · Accepted solution]

Well, lets plug in values and see what comes out.

Female patient, male surgeon = 232.65

Female patient, female surgeon = 255.95

Male patient, male surgeon = 241.35

Male patient, female surgeon = 273.48

So the differences between male and female surgeons are not the same for female and male patients (23.3 in the first case, 32.33 in the second),  Also, the differences between female and male patients for male and female surgeons (8.7 vs 17.53).

Does that help your interpretation?  If you used GLM, you might want to look at the interaction plot that should be in the output.

SteveDenham

[PaigeMiller]

Very good suggestion from @SteveDenham, which I agree with. Look at the interaction plots. The plots will help you understand what the interaction is, better than any words on a computer screen can.

If there is no interaction, the lines will be parallel (or very close to parallel if the interaction is not zero but also not statistically significant). With interaction, the lines are not parallel, and so you can see how the slopes change depending on the level of a 2nd variable (that's the definition of interaction)