All my predictors are binary in nature. So far I have been building the model with the logistic fuction. Could anyone suggest any appropriate statistical technique keeping the following points in mind:

  1. Dependent variable - binary
  2. All Independent variables - binary
  3. Response rate - On the lower side (I only include those which have a respose rate higher than 1%)

Thanks for your time in advance Kind regards

  • 1
    $\begingroup$ Can you please clarify what you actually want to do...what question are you trying to answer with these data? Can you also clarify your point about response rate? $\endgroup$ – D L Dahly Apr 3 '13 at 9:58
  • $\begingroup$ Hi, thanks for your interest. I would like to give a simple & most common example to clarify this. Dependent variable: defaulted/ did not default: Yes/No............ Independent binary variables: say, there are 2 binary variables.......(1) Employed/ Unemployed - 1/0 &&&&....(2) Married/ Unmarried - 1/0 >>>>> The response rate of an independent variable means -> (Sum of all 1)/Total number of observations>>>>>>>>. So, I would like to build a logistic model with the above 3 variables.. Hope this answers your query $\endgroup$ – Peter Apr 3 '13 at 10:54
  • 1
    $\begingroup$ But what question are you trying to answer with these data? Also, what is the unit of observation (a person, a business, etc), and do you have repeat measures, i.e. does each unit of observation have it's own "response rate"? $\endgroup$ – D L Dahly Apr 3 '13 at 11:14
  • $\begingroup$ A logistic model is to be built which will predict the probability of an individual defaulting on a bank loan..which means a discrete output (1/0).>>> The independent variables of this model which are Employed/ Unemployed & Married/ Unmarried -- binary in nature.>>>>>>here is my final equation>> Log (P(default)/(1-P(Default)) = A + B(Employed/ Unemployed) + C(Married/ Unmarried)____So, I am trying to answer whether the individual falls in the default loan category or not____All variables are individual (person) specific___ Lets' not discuss the response rate here, let's keep it out of this :) $\endgroup$ – Peter Apr 3 '13 at 11:46

You could use a decision tree: http://en.wikipedia.org/wiki/Decision_tree_learning

It should give similar results as logistic regression, but it's a really easy way to present the effects of binary variables.

After you have run both models, you can look at a contingency table of correct and incorrect classifications to determine which model to use.

  • $\begingroup$ Thank you for your time & response. I tried it and looks good but I need some quantifiable output similar to odds ratio. Here, I get the 'p' value but could you elaborate a bit on a hypotheical interpretation of the above three variables based on the decision tree. Many thanks $\endgroup$ – Peter Apr 3 '13 at 13:13
  • $\begingroup$ You are now insisting on two different questions, Peter. In another comment you stipulated you wanted "a discrete output (1/0)" but now you say you "need [an] ... odds ratio." Could you please make up your mind what your question is and edit it to make that clear? $\endgroup$ – whuber Apr 3 '13 at 15:16
  • $\begingroup$ Kindly note, that when I spoke about the discrete output, I was explaining him as to what I am doing and not what I want as he asked me - "what question are you trying to answer with these data". Since this is a predictive model with discrete dependent then the output will always be discrete. Only change in my project is that the odds ratio is the mandatory field and it is importat so we were trying to look out for other(better) predictive models which will give an output such that each indep variables will have their respective betas or other quantifiable values (like odds ratio in logistic) $\endgroup$ – Peter Apr 3 '13 at 16:42

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.