I'm trying to understand the explanation of how multinomial logistic regression works from the point of view of a set of independent binary regressions.

The wikipedia page seems to provide a fairly good explanation. However, I'm stuck on an algebra step that I can't seem to get.

Namely, how do we go from 1-sum product of p(y = k) and e^coef to the next step? enter image description here


I realize it may be an easy solution that I'm just missing.


$\Pr(Y_i = K)$ can be taken out of the sum as

$$\begin{align}\Pr(Y_i =k)&= 1 - \Pr(Y_i =k)\, \sum_{k = 1}^{K-1}\,e^{\beta_k\cdot X_i}\\[2ex] 1&=\Pr(Y_i =k) + \Pr(Y_i =k)\, \sum_{k = 1}^{K-1}\,e^{\beta_k\cdot X_i}\\[2ex] 1 &=\Pr(Y_i =k)\,\left(1+\sum_{k = 1}^{K-1}\,e^{\beta_k\cdot X_i}\right)\\[2ex] &\implies \Pr(Y_i =k) =\frac{1}{1+\sum_{k = 1}^{K-1}\,e^{\beta_k\cdot X_i}} \end{align}$$


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.