1

I am doing a regression on the big 5 personality traits, and how birth order affect those traits. First I am trying to build 5 variables based on surveys that captures those traits. I have thought about creating dummies for each question in the category (trait) and then taking the average, but some of the questions are highly correlated, so the weight would be wrong.

I have made a principal components analysis, which gives me four components with an eigenvalue over one. The problem is that none of them accounts for over 40 pct. Of the variance.

Is there some way that I can merge the four into one variable? It is the dependent variable, so there can only be one.

Otherwise do you have another idea of how the index can be constructed?

Stephen Rauch
  • 47,830
  • 31
  • 106
  • 135
MissM
  • 11
  • 2

1 Answers1

0

It doesn't really make sense to merge your principal components as they are by definition orthogonal/uncorrelated.

I can't advise on how an index could be constructed as I think this requires subject matter expertise, but you might want to consider using a multivariate technique that allows for multiple response variables. See this answer for a possible approach (assuming your response variables are ordinal).

jruf003
  • 980
  • 5
  • 19