0

I have two variables both are binary.

I would like to see how well one predicts the other.

Is it possible to do this in logistic regression? variables are coded as 0 and 1 but 1 does not indicate that something is 'correct'. can I still use logistic regression to see how well it is predicted?

upabove
  • 2,657
  • 10
  • 30
  • 37

1 Answers1

5

You can, but you can get all the information you need from the simpler Chi-Square test and/or from a Phi correlation. Easier to conduct and more straightforward to interpret and report results.

rolando2
  • 11,645
  • 1
  • 39
  • 60
  • so I can run a simple chi square on two binary variables with around 160 values? – upabove Aug 25 '11 at 22:44
  • 1
    @rolando2, could you show an example of where chi-square / phi correlation would be better than logistic regression for estimating effects of other binaries? I use logistic regression on survey data quite frequently `1=increasing / 0=decreasing ~ personality attributes (0 or 1)...` to understand the effect of personality attributes on likelihood to increase or decrease business with a company. – Brandon Bertelsen Aug 26 '11 at 00:05
  • @Brandon I won't say "better" in an absolute sense, but better in the ways I mentioned. – rolando2 Aug 26 '11 at 18:21
  • @Daniel In your original post you indicated that your variables took on one of only 2 values--0 or 1. Do you mean 160 cases rather than 160 values? If so, yes, you can run a Chi-Square test in that situation. – rolando2 Aug 26 '11 at 18:23