I am wondering what are the advantages/disadvantages of breaking down a logistic regression in multiple steps, when they are available.

Let me explain what I mean by multiple steps: Think of it like the customer journey: A cold lead (A) becomes a prospect (B) who then becomes a customer (C).

A -> B -> C

I'm interested in predicting the conversion from A to C, which can be done with a logistic regression.

I wonder if I could also do two logistic regressions, first from A to B, then from B to C, and multiply the predictions.

What are the differences between the two approaches?

Things to consider:

  • What if the conversion rate from A to B is small? (Then the sample size for the 2nd model is small as well)
  • Where does most of the signal come from? Maybe my explanatory variables explain most of A to B but nothing of B to C, or the other way around.

1 Answer 1


That sounds like a sequential logit to me. You can compute a "total" effect of explanatory variables on the finale outcome and decompose that into a weighted sum of the effect of that explanatory variable on each step/transition. See: http://dx.doi.org/10.1177/0049124115591014

  • $\begingroup$ Also take a look at state transition models. $\endgroup$ Jul 7, 2017 at 12:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.