0
$\begingroup$

I have been provided a sample logistic regression as follows:

glm(formula = output ~ X1 + X2 + X3 + X4 + X5 + X1:term + 
               term:X5 - 1, family="binomial", data=mydata)

I am not too familiar with logistic regression, so I have a few questions about how to properly predict on a new test set using this model:

  1. Unlike a regular regression, I cannot simply 'plug-in' the variables and get a meaningful numeric output. Instead, I must first set a threshold probability above which values will be 1 and below which values will be 0. Is this correct?

  2. I cannot make use of this sample model or get the same results as the person who provided it until I have the probability threshold that was used for prediction. Is this correct?

  3. If I wanted to split the outputs into tiers, would I use the probabilities for that and map them to some other value? How would that process work (feel free to let me know if this is out of scope).

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ Don't split the probabilities. They are meant to be used continuously. For the other questions some in-depth background reading about logistic regression would help. $\endgroup$
    – Frank Harrell
    Commented Mar 16, 2016 at 17:02
  • $\begingroup$ @FrankHarrell Agreed, though I imagine this could help others who are being introduced to logistic regression through glm and are unclear on interpretation. $\endgroup$
    – 114
    Commented Mar 16, 2016 at 17:35

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.