I'm trying to estimate a fractional outcome through weighted logistic regression. One of my IVs is continuous and cannot be negative. And about 45% of the values are 0. Most of the zeros are related to another IV which is binary and contains the status of the observation (active vs inactive - 70/30). In other words, 98% of the observations that are Inactive have a 0. The distribution on the Active side is 21% (zeroes) and there is a positive correlation between higher values of this IV and the response. My sample size is very large.

How do I approach modeling this situation? Specifically, what extra care is required to handle the large concentration of 0s?

  • 3
    $\begingroup$ Maybe no extra care is needed at all: as always, everything depends on the relationship between the response variable and the explanatory variables. Nothing general can be said based on the properties of the explanatory variables alone. Could you provide information to narrow the scope of your question? $\endgroup$ – whuber Jan 11 at 21:47
  • $\begingroup$ Thanks, @whuber. Here is a summary of the relationship. The numbers are fractional outcome (response) that I am trying to model. Active / 0 - 12% Active / >0 - 44% Inactive / 0 - 8% Inactive / >0 - 33% $\endgroup$ – StatsStudent Jan 12 at 0:15
  • $\begingroup$ @kjetil b halvorsen - thanks for the edits. $\endgroup$ – StatsStudent Jan 12 at 0:40
  • 1
    $\begingroup$ You sould give some more context! Do you have only the proportions, or do you have numerator and denominator also? Proportion of what? "sample size is very large"--- some substructure to the sample? different country/regions/schools/...? sampled over time? ... $\endgroup$ – kjetil b halvorsen Jan 12 at 13:21

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.