Estimate of mean squared error

[CEdward]

Hello,

Is there a Stata written command that allows one to determine the mean squared error from a mixed-effects linear regression model? I want to estimate this quantity for a simulation analysis that I am doing.

All best,
Jack

[Joseph Coveney]

You can square the root MSE that is returned.

sysuse bplong

anova bp sex / patient|sex when sex#when, repeated(when)

display in smcl as text "MSE = " as result e(rmse)^2

[CEdward]

Thanks Joseph.

Would this work for a mixed-effects model?

I have adjusted my simulation code, have I done it correctly?

capture program drop swcrt
program define swcrt, rclass
/*Code omitted*/
/*Fit multi-level model to simulated dataset*/

mixed y intrv i.time ||cluster: ||individual:, covariance(unstructured) reml dfmethod(kroger)

/*Return estimated effect size, bias, p-value, and significance dichotomy*/

tempname M
matrix `M' = r(table)
return scalar bias = _b[intrv] - `intrvcoeff'
return scalar mse = e(rmse)^2
return scalar p = `M'[r(p)]
return scalar p_= (`M'[`r(p)'] < `alpha')
exit
end swcrt

*Postfile to store results

tempfile powerresults
capture postutil clear
postfile step int num_clus intrvcoeff p p_ bias mse using `powerresults' /*Maybe add ICC*/

*Loop over number of clusters, ICC

foreach c of local num_clus{

[INDENT=2]display as text "Number of clusters" as result "`c'"[/INDENT]/*

foreach i of local ICC{

[INDENT=2]display as text "Intra-class correlation" as result `ICC' */[/INDENT]

forvalue i = 1/`nrep'{

[INDENT=2]display as text "Iterations" as result `nrep'
quietly swcrt `c' `intrvcoeff' `sigma_u3' `sigma_u2' `sigma_error' /*Maybe add ICC*/
post step (`c') /*(`ICC')*/ (`r(p)') (`r(p_)') (`r(bias)') (`r(mse)')[/INDENT][INDENT=2]}[/INDENT]

}

}
postclose step

*Open results, calculate power

use `powerresults', clear
levelsof num_clus, local(num_clus)

/*levelsof ICC, local(ICC)*/ /*Maybe remove or add ICC*/

matrix drop _all

*Loop over combinations of clusters and ICC
*Add power results to matrix

foreach c of local num_clus {

/*foreach i of local ICC { */

quietly ci proportions sig if clus_num == `c' & /*float(ICC)*/ == float(`i')
local power `r(proportion)'
local power_lb `r(lb)'
local power_ub `r(ub)'
quietly ci mean bias if clus_num == `c' & /*float(ICC)*/ == float(`i')
local bias `r(mean)'
local bias_lb `r(lb)'
local bias_ub `r(ub)'
quietly ci mean mse if clus_num == `c' & /*float(ICC)*/ == float(`i')
local mse `r(mean)' /*Would this be correct?*/
matrix M = nullmat(M) \ (`c', `i', `power', `power_lb', `power_ub', `r(mean)', `r(lb)', `r(ub)', ) /*May need to remove extra comma and space; why isn't rmean, rlb, and rub bias, biaslb, and biasub*/

}

}

[CEdward]

I reviewed the stored results section of the mixed command and it appears that the mean squared error is not generated for a multi-level model, but an anova. Does anyone know if this is possible in Stata?