I'm running a binary logistic regression from 5 predictor variables. The outcome variable is match/no match. Although I've run this for four different groups of data with varying distributions on the outcome variable (i.e. Group 1 has match = 500/no match = 1500 and Group 4 has match = 900/no match = 1100, for example) my classification table only predicts 'match'.

                     No match           Match 
         No Match       0                3048
         Match          0                3132
 Note: Cut value = .500     

This outcome persists regardless of the number of variables included in the model, the proportion of outcome responses, and the cut value. Why might this be happening?

  • 2
    $\begingroup$ It means you dont have good predictors? Also, you are using an improper score function. Logistic regression gives you a fitted probability, use that directly, do not dichotomize it. For example, show us plots of fitted probability against each of your predictors. $\endgroup$ – kjetil b halvorsen Apr 20 '18 at 17:02
  • $\begingroup$ See for instance: stats.stackexchange.com/questions/73165/… for discussion $\endgroup$ – kjetil b halvorsen Apr 20 '18 at 19:52
  • 1
    $\begingroup$ I second @kjetil's recommendation that you should not cut predicted probabilities from the logistic regression model as SPSS does by default. Also, if you raise the cut value high enough, you'll eventually see some no match cases. But I feel the way this table is generated makes it pretty much useless for evaluating the model. $\endgroup$ – Heteroskedastic Jim Apr 21 '18 at 1:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.