3
$\begingroup$

I have been looking into interpreting feature coefficient in a bounded capacity (0,1) as probability scores. So for example, if one has the following linear model for classification(Logistic regression), then inverse logit should return one linear probability scores right as it is the inverse logs of odd ratio?

y<- c(1:3, 7, 6)
x1<- 1:5
x2<- c(20:25)
temp<-coef(lm(y ~ x1 + x2))
library('arm')
invlogit(temp[1]+temp[2]) 

However, in-case of linear regression for continuous value prediction,

coeff1/sum(coeff)

should return one probability scores right ? Did I miss something ?

$\endgroup$

1 Answer 1

2
$\begingroup$

No, coeff1/sum(coeff) cannot be a probability due to the fact that it can be bigger than 1. For instance: coeff1 = 1., coeff2 = -.5, then coeff1/sum(coeff) = 2. The interpretation of coefficients in linear regression is the derivative of the output variable with respect to the given predictor.

$\endgroup$

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.