Questions: The identity link is the standard one with normal responses but is not often used with binary or count responses. Why do you think this is?

My idea: The range for a linear predictor, and the range of the identity link applied to a binomial probability or to a Poisson mean. I am kinda confused

  • 2
    $\begingroup$ It’s to map the real line to a range of plausible values of the parameter of the response distribution. Instead of $Poisson(-1)$, the $log$ link gives us $Poisson(e^{-1})$. $\endgroup$ – Dave Feb 14 '20 at 3:47
  • $\begingroup$ @Dave What do mean by Poisson(-1)? $\endgroup$ – Simpson's Paradox Feb 14 '20 at 3:51

If you use the identity link for the binary data, then your regression equation would be $$ {\rm Pr}(Y_i|x_i)\equiv p_i = \beta_0 + \beta_1 x_i. $$ Without any restriction on the coefficients, RHS can range in real-line (for example, a negative value). Since this does not make any sense for probability $p_i$, we transform it so that $g(p_i)$ takes its value in $[-\infty, \infty]$.


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.