0

Good afternoon.

A research project I am part of is using multiple demographic variables to predict performance on a continuous scale item measuring a specific behavior so that we can describe people of trait X+Y+Z are likely to have the greatest elevations on this scale. Included variables are both continuous and dummy coded binaries (for example, income is an included continuous variable and gender is a binary). The variables have been entered into the linear regression model in blocks consistent with theoretical realms of influence for the dependent variable (e.g., social factors, physical factors, etc). One of these factors is ethnicity which we have dummy coded for this analysis.

When we first ran the regression analysis, it excluded the first dummy coded variable from the output and it appeared to use that variable related to the regression constant. So that we could include, and easily interpret, factors related to elevations in the dependent variable we re-ran the analysis and requested that a constant be excluded. When we did that, relationships to all the other variables changed. This change in the analysis led to things which were not significantly predictive before now being significant.

I have two questions

  1. Generally speaking, does this approach make sense to examine what factors predict elevations in a given dependent variable?

  2. Why would the regression results differ substantially with the constant removed / what does that mean interpretatively for the first (constant included) versus the second (constant excluded) model.

Thank you for the help. I apologize if I missed an answer to this elsewhere. I had no luck with search.

Paul
  • 1
  • Welcome to CV. Since you’re new here, you may want to take our [tour], which has information for new users. Is there a specific reason to remove the constant? It is not recommended as you can see here: https://stats.stackexchange.com/questions/7948/when-is-it-ok-to-remove-the-intercept-in-a-linear-regression-model – T.E.G. Jul 20 '17 at 00:08
  • With the constant included, the results exclude any beta/standardized beta information related to ethnicity as that information appears to be used as the reference for the model block. This seems to not be the clearest way to interpret the results when asking 'to what degree does membership in different ethnic groups predict a given score'. Perhaps my understanding of how to interpret the results when the constant is included is incorrect in this case? – Paul Jul 20 '17 at 00:28

0 Answers0