I'm reading a chapter on categorical outcome variables (chi-square & log linear analysis) and the author, in an effort of fitting a linear model, said that because the outcome variable is a categorical one, we have to use log values. So the model changed from :

outcome = (b0 + b1*variable_a + b2*variable_b + b3*interaction) + error

to : ln (outcome) = ln(model) + ln(error)

  • 1
    $\begingroup$ The outcomes are counts of the number of cases that fall into combinations of categories — also known as contingency tables. I believe that the log in loglinear is the log of a ratio of expected versus actual counts for each cell in the contingency table, so I’m not sure how your restatement is correct. I could be wrong, but are you sure your question is properly stated? $\endgroup$
    – Wayne
    Apr 30, 2019 at 22:34

1 Answer 1


As @Wayne above stated, the outcomes are counts. Counts are strictly non-negative, and so must be the modelled means. The $\log$ serves to make sure that all modelled means are positive, as its inverse transformation ($\exp$) has a positive range.

Note that in this case the concept of outcome = mean + error no longer holds.


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.