MSE
Message/Author
 Eric Teman posted on Thursday, May 17, 2012 - 6:26 pm
I am looking on p. 363 of the Mplus manual (version 6).

It looks like the std. dev.'s are nearly identical to MSE's. Does this indicates small bias in general? In other words, if MSE's diverged greatly from empirical st. dev.'s, would that indicate bias?
 Eric Teman posted on Thursday, May 17, 2012 - 6:45 pm
On p. 364, it says, "To determine standard error bias, subtract the population standard error..." My questions is, how would one know the population standard error? It can only be estimated with MSE, right?
 Linda K. Muthen posted on Friday, May 18, 2012 - 11:52 am
It says subtract the population parameter value not population standard error. The example uses numbers taken from the output. Take another look at it.
 Eric Teman posted on Friday, May 18, 2012 - 8:35 pm
If you look at the MSE's on page 363 of the manual, they appear to be calculated incorrectly. MSE = bias^2 + variance of the estimate.

If you look at variance I, for example:

bias = (.5 - .4820) / .5 = .036
variance = .1950^2 = .038025
MSE = .036^2 + .038025= .039321

However, the MSE produced by Mplus is .0493. What is causing this discrepancy?
 Eric Teman posted on Friday, May 18, 2012 - 8:37 pm
I mean the MSE produced by Mplus is .1954. But why isn't it .039321?
 Linda K. Muthen posted on Saturday, May 19, 2012 - 9:04 am
The output in the user's guide is not correct.

The calculation for MSE for the example is:

MSE = (.5 - .4820)^2 + .1950^2

The value -3.6 is percent bias. The actual bias is the difference between the population values and the average value.
 Eric Teman posted on Saturday, May 19, 2012 - 10:53 am
OK, but if you calculate the MSE you just gave me, you would get .03825, not .1954 as the manual states. It still appears incorrect in the manual. Please verify.
 Eric Teman posted on Saturday, May 19, 2012 - 10:58 am
I just realized you had said NOT correct above. That answers my question. Thanks!
 Bengt O. Muthen posted on Saturday, May 19, 2012 - 11:09 am
The V6 UG uses old output. MSE is correct in Version 6.12.