0
$\begingroup$

I have a question on the coding/use of categorical variables: In short, I have two control variables in an SPSS regression model. One is participant (15 levels, or participants) and one other categorical variable (10 levels). And there is just one predictor variable in the model. I've tried two different approaches to the control variables:

1) Coding all 25 as separate 0,1 variables. This allows us to see more nuanced predictions, which is preferred!

2) Coding the two variables as two dummies (0, 1, 2, 3 etc), and (0, 1, 3, 4, etc.).

My question: I assumed I would get nearly the same result given this is essentially controlling for the same thing. But I'm seeing different results for the IV: a weakly positive coefficient for approach 1, and a weakly negative coefficient for approach 2. So my question is why is there a different result/sign reversal of a pretty weak predictor variables? This, to me, is weird.

I would really appreciate any input you may have on this - and what approach is correct - or if there may be something else wrong. Thanks so much!

$\endgroup$
0
$\begingroup$

First, from what you've said, it's not at all clear that "participant" should be a variable at all. For participant to be a variable you'd need to have multiple measures on each participant or have some form of clustering and be working on a multilevel model. Then participant would be a random effect. But I am pretty sure that is not what you are doing.

2) For the 10 level categorical variable, you would not code it as (0, 1, 2, 3....) but as a series of 1's. SPSS will do this for you, you don't need to do it yourself.

3) Your description of your results is very odd. You can't have "negative" or "positive" relationships with a categorical variable.

I think you need to take a course on regression; these are fundamental issues.

$\endgroup$
  • $\begingroup$ Thanks, Peter. To clarify: We do have multiple measures per subject, thus justifying why we're coding for each subject to control for subject-level effects. On point 3: I was referring to the estimate of the independent variable. It changes from a negative to a positive in the model depending upon the two approaches used for controls described above. $\endgroup$ – Kacie Feb 14 '18 at 14:42

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.