Does EM-algorithm only work with missing data? If not, what is the idea to assume that we have missing data?

  • 1
    $\begingroup$ Any density that can write as$$f_\theta(x)=\int_\mathscr{Y} g_\theta(x,y)\text{d}y$$qualifies for EM, not truly a restriction. Provided you can implement the E step. $\endgroup$
    – Xi'an
    Sep 21, 2016 at 8:31
  • $\begingroup$ Check also: stats.stackexchange.com/questions/72774/… $\endgroup$
    – Tim
    Sep 21, 2016 at 8:42

1 Answer 1


I think the Wikipedia for EM algorithm has everything you want and it probably explains better than I do. You can also use the EM algorithm for latent variables; no missing data but variables that you can't observe and measure. The idea of the algorithm is to iteratively adjust your missing data/latent variables until your maximum likelihood estimate converges.


Your Answer

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