0

I have a linear regression model that predicts lifetime customer value. It has coefficients that tell me things like, if the customer is a VIP then they have +$100 value. However if i know that these coefficients actually add to values in more of a percentage way than absolute value, then how can i get an estimate of these percentages? A bit like the odds ratio in logistic regression.

Perhaps i shouldn't be using linear regression to understand these features? Any thoughts in how I can understand the percentage value increase of customer value for these features?

Kodiologist
  • 19,063
  • 2
  • 36
  • 68
khhc
  • 47
  • 7

1 Answers1

4

Linear regression is based on the assumption of additive effects. If you expect multiplicative rather than additive effects, you can change the model. For example, if you replace the dependent variable with its logarithm, then you get estimates of multiplicative effects, because additive effects on a log scale are multiplicative on the original scale.

Kodiologist
  • 19,063
  • 2
  • 36
  • 68
  • Thanks for the informative reply. Yes multiplicative rather than additive is what i was after i just didn't have the vocab to describe it! I have done what you said, took the natural log of the dependent variable. Now when a coefficient 'VIP' says, '1.2', it's stating that if a customer is a VIP they have a 120% higher value than a customer with the exact same properties (apart from their VIP status)? This seem to be what is suggested here: https://stats.stackexchange.com/questions/18480/interpretation-of-log-transformed-predictor – khhc Dec 12 '18 at 11:46