This has very little to do with multiple imputation or multilevel
models. This is work for -nlcom-, and using anything else is likely an
inferior solution:
nlcom exp(_b[lns1_1_1:_cons])
As it was mentioned, exponentiation is a pretty brutal operation
skewing the distributions quite markedly. There might be hope for
asymptotic normality of the logged variances; there is less hope for
asymptotic normality of the variances themselves, hence an additional
reason for log transformation.
On 11/11/08, Nailing Xia <[email protected]> wrote:
> Hi,
>
> I have a question regarding the standard error of the random-effects
> parameter in multi-level models (-xtmixed).
>
> I am estimating a multilevel model using -xtmixed-, and the dataset I
> use is from multiple imputation procedure using -ice-. I use -mim- for
> estimation and the command is : mim: xtmixed irtm pexpect pbelmth
> mclsize mtexp if [parwgt] || schid: || childid:. The output gives the
> random-effects paramters in logrithm of the paramters:
> ------------------------------------------------------------------------------
> irtm | Coef. Std. Err. [95% Conf. Int.] MI.df
> -------------+----------------------------------------------------------------
> /lns1_1_1 | 1.20731 .052497 1.10429 1.31034 923.0
> /lns2_1_1 | 2.42047 .009313 2.40208 2.43886 164.9
> /lnsig_e | 2.21744 .00496 2.2077
> 2.22718 597.9
> ------------------------------------------------------------------------------
>
> I know I can get the paramter by using di exp(1.20731), but how can I
> get the corresponding standard errors?
--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
