1

I am currently studying the effects of intercontact time in a customer's purchase journey related to his/her purchase probability, as well as the influence of the number of contacts. Based on extensive literature I expected an inverted U-shaped relationship between intercontact time and purchase probability, which is why I developed the following model:

mylogit_poly <- glm(purchase_any ~ poly(intercontact_time, 2)  + number_contacts_mc, data = Subs1, family = "binomial"(link="logit"))
summary(mylogit_poly)

Which results in:

[Output 1[1]

Now, based on feedback I received from my supervisor, I tried different levels of the polynomial term. When I got to the model below (with poly being 4), I suddely get all significant results.

mylogit_poly4 <- glm(purchase_any ~ poly(intercontact_time, 4)  + number_contacts_mc, data = Subs1, family = "binomial"(link="logit"))
summary(mylogit_poly)

[Output 2[2]

My question now is what to make of this? Does this mean that the relationship can be considered fully non-linear? Can I infer something from this?

PS: My full model exists of an additional 15 control variables that I left out for the sake of complexity. The output shown above comes from the full model.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Daniël Lutjens
  • 131
  • 1
  • 3
  • Strongly related [question](https://stats.stackexchange.com/questions/198484/why-the-significance-of-terms-in-orthogonal-polynomial-regression-changes-with-t) maybe even duplicate? – Joel May 11 '18 at 18:33
  • 2
    You should consider a spline model as an (often better) alternative to high order polynomials – kjetil b halvorsen May 14 '18 at 07:29

0 Answers0