0
$\begingroup$

In relation with my other question here where I observe a strange behavior of the residuals after fitting binary data using glm/glmer, I now wonder:

Are there boundaries on the proportion of 0 (or 1) necessary in order to be able to regress binary data? If there are too many 1 compared to the 0 or conversely, does it prevent attempts of modeling?

Cheers

$\endgroup$

1 Answer 1

1
$\begingroup$

There are no hard limits aside from the fact that your data can't be 100% one class. It is true, however, that the less balanced your classes are, the harder it is to learn to distinguish them.

$\endgroup$
1
  • $\begingroup$ Thank you for your answer. Would you happen to know what could cause the residuals to be biased and not have a null mean ? $\endgroup$
    – Tom Bug
    May 30, 2016 at 0:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.