Re: Little's MCAR Test and EM algorithms

Posted by Rich Ulrich on Jan 10, 2012; 6:25am
URL: http://spssx-discussion.165.s1.nabble.com/GLM-repeated-measures-concern-question-tp5104022p5133464.html

If it never converged, then it never reached a stable answer. 

In general, that would say that the answer is not valid.

Also in general -- for results that are based on convergence after
iterations, there might be some clue in the progressive iterations
by single steps.  Big fluctuations suggest a problem with the problem.

--
Rich Ulrich

---

Date: Mon, 9 Jan 2012 20:14:24 -0600
From: [hidden email]
Subject: Little's MCAR Test and EM algorithms
To: [hidden email]

I am trying to determine whether my data are missing at random using Little's MCAR test.

When I run the test via the missing values analysis, I also get a notification that "The EM algorithm failed to converge in 25 iterations".

When I increase the iterations (i.e., 50,75,100), the Chi-square value for Little's MCAR test also changes. In some cases from significant to non-significant.

I'm wondering if anyone could provide me with more information about the relationship between these and whether or not the failed convergence on the EM algorithm makes the Little's MCAR test invalid.

Thanks!

[snip, sig; and previous post, not relevant]