1

Doesn't the different distribution shapes of the respective variables (e.g., binary vs. fractional) influence the regression coefficients per se?
As the binary variable represents only "extreme values" of the fractional variable, 0 or 1.

Example: dependent variable: payment yes/no 0,1
independent variable a): gender male/female 0,1
ind. var. b): percentage of items in the basket that are green [0,1]
=> IV a) can only have 2 values 0 or 1, whereas IV b) can have infinite values between 0 and 1

Chase McIntosh
  • 23
  • 1
    Do you mean to ask whether you can have, for example, both age and sex as independent variables? (Yes) Or do you mean to parameterize a continuous variable in a multiple ways (eg. categorizing, dividing by another variable, or other mathematical transformations, etc...) and add all of those in your model (Probably not) ? – IWS Jun 13 '17 at 13:56
  • 4
    Logistic regression is used for binary dependent variables. You can use whatever independent variables you can reasonably code. Period. You might give a more complete description of the variables you have in mind and why you think they are problematic. Their problems are most unlikely to have anything to do with logistic regression per se but are likely to be more general. – David Smith Jun 13 '17 at 13:58
  • I added an example in my question of one binary and one fractional variable – Chase McIntosh Jun 13 '17 at 14:25
  • There is aboslutely no problem with the variables in your example. – kjetil b halvorsen Jun 13 '17 at 16:15
  • No issue but keep in mind that different "format" for the predictors will have an impact on the estimates (As your predictors won't be measured on the same scale (binary, fractional, etc) you cannot directly compare the model estimates) – Nicolas K Jun 13 '17 at 22:22

0 Answers0