Currently I'm working on a model that predicts sale based on the activity of a rep. It can be 3 types: mail, phone call, or meeting. These variables are continuous and somewhat cross-correlated, e.g., a rep can send 10 emails, make 1 call and have 3 meetings - and the more calls a rep makes, the less emails he/she sends.

Ultimately I'm trying to understand proportional weight of each activity to create combined activity metrics and say something along the lines: To win a deal with X% probability you need to perform 10 activity "units" where a meeting gives you 5 units, a call 3 units, and an email 1 unit.

It looks that in this case logistic regression should be used to estimate probability of sales from combined activity metric, but how to get best weights for each activity type?


Do logistic regression with the three variables (and possibly interactions). From the results of the regression, you can estimate the probability for any combination. SAS and other programs can do this for you for combinations that you set, or you can use the formula:

$$ P = \frac {e^{b_0 + b_1 X_1 + b_2 X_2 + b_3 X_3 + \dots b_p X_p}}{1 + e^{b_0 + b_1 X_1 + b_2 X_2 + b_3 X_3 + \dots b_p X_p}}$$

| cite | improve this answer | |
  • $\begingroup$ Thank you! But i'm not sure I understand, can you please advise how to do it in R? When I do GLMLogit <- glm(formula = IsQuestion~ Metric 1+ Metric 2+ Metric 3+ Metric 4, data = Input data, family = "binomial") Is this it? Do I say P = intercept+Metric 1*Coeficient 1+ Metric 2*Coeficient 2 + .... Very new to this, so really appreciate your help! $\endgroup$ – Dmitriy Berdnikov Sep 18 '17 at 19:13
  • $\begingroup$ Questions about coding are off topic here, but that looks right to me. $\endgroup$ – Peter Flom Sep 19 '17 at 22:41

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.