Hierarchical Regression

[milehigh899]

Hi,

I need to perform hierarchical regression and include in the models independent variables (some interval and some dummy variables), moderator (interval scale), and the interactions of the IVs and the moderator. My question is: to create interaction of the dummy variable and the moderator, do I simply multiply the dummy variables with the moderator? My confusion is whether it makes sense to multiply a dummy (categorical) variable by an interval level variable.

Any comments are very much appreciated.

Kind regards
Muzi

[Jake]

Yes, it makes sense. Just note that in this model the coefficients for the dummies will represent the predicted effect when moderator = 0, and conversely, the coefficient for the moderator will represented the predicted slope for the baseline category (i.e., when the dummies = 0). Personally I would probably not use dummy variables in such a situation but some fields seem to be obsessed with them.

[milehigh899]

Yes, it makes sense. Just note that in this model the coefficients for the dummies will represent the predicted effect when moderator = 0, and conversely, the coefficient for the moderator will represented the predicted slope for the baseline category (i.e., when the dummies = 0). Personally I would probably not use dummy variables in such a situation but some fields seem to be obsessed with them.

Thanks for your quick reply.

I have three dummy variables, but I will enter only two in model and interpret them with reference to the third absent dummy variable. I hope this is right. For the predicted effect, do I look at standardized or unstandardized coefficients? Also, the moderator is significant alone too (when entered in a separate model like an IV) and in interactions also. How do I interpret such a result. Thank you very much for your time and expertise..

PS: I am working in the field of business management, and playing with the dummy variables is kinda necessary sometimes, particularly in this case the practice is supported by literature

[Jake]

I have three dummy variables, but I will enter only two in model and interpret them with reference to the third absent dummy variable. I hope this is right.

Yes, right.

For the predicted effect, do I look at standardized or unstandardized coefficients?

Personally I would stick to unstandardized coefficients but it doesn't really matter either way. It's the same model.

Also, the moderator is significant alone too (when entered in a separate model like an IV) and in interactions also. How do I interpret such a result.

As noted above, in the model that includes the dummies + moderator + interactions, the coefficient for the moderator represents the slope of the moderator for the baseline category. The model without the dummies or the interactions is just a simple regression of the DV on the moderator. So your results indicate that there is a relationship between the moderator and the DV both overall (i.e., the no-dummy model) and more specifically for the baseline category (i.e., the interaction model). It's not obvious whether this second result is of any particular interest.

[milehigh899]

Yes, right.

Personally I would stick to unstandardized coefficients but it doesn't really matter either way. It's the same model.


As noted above, in the model that includes the dummies + moderator + interactions, the coefficient for the moderator represents the slope of the moderator for the baseline category. The model without the dummies or the interactions is just a simple regression of the DV on the moderator. So your results indicate that there is a relationship between the moderator and the DV both overall (i.e., the no-dummy model) and more specifically for the baseline category (i.e., the interaction model). It's not obvious whether this second result is of any particular interest.

Thank you so very much. Your answer is clear and bang on!

[milehigh899]

Thank you so very much. Your answer is clear and bang on!

Hi,

Could you please spare a minute on my output. I take the liberty of attaching the output in PDF file. I'm aware that you will perhaps have many better things to do, but I would very much appreciate your comment on the attached output. Thank you again!