Can you please help me with 2 questions (both to do with dummy variables) -

1) If I have 2 sets of nominal variables (eg ethnicity [white, Indian, African] and smoking status [current smoker, ex-smoker, non-smoker]) and I turn each set into 2 dummy variables (+1 baseline reference variable for each set) for entry into a multiple linear regression in SPSS. How does SPSS know which baseline reference variable to use for which set?

At least in logistic regression you have to option to label each variable as categorical (so there seems to be some way of grouping them) - this does not exist for linear regression.

For my example given above, does it mean that the multiple regression would produce 2 constants (one for each set of dummy variables)?

2) If I recode ethnicity (White, Indian, African) into 3 dummy variables, why can't these be treated as 3 separate binary predictor variables and entered into the model? Ie why can't I enter 'White / Non-White', 'Indian / Non-Indian' and 'African / Non-African' : and thus why can't I interpret the resultant coefficients as White compared to Non-white, Indian compared to Non-Indian and African compared to Non-African?


Question 1 is really something you should look up in SPSS help.

Question 2 you seemingly already answered: the proper way of coding a polytomous variable is creating contrasts for a referent group of YOUR picking. Take the race variable. You describe 3 possible binary variables: white vs nonwhite, indian vs nonindian, and black vs nonblack. In any regression model you must pick 2 of the 3, with the omitted third variable being the referent group. Supposing you picked "white" as the referent, then the effect you estimate for the black variable is an estimated difference in the outcome comparing blacks to whites.


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.