weighted random effects model with clustered SE

[Mywish Maredia]

Hello:

I want to estimated a weighted RE model with clustered standard errors. The xtreg does not allow weights with the re option. The xtregre2 allows weights but not clustering of SE. What are my options? Is there any command that will allow me to estimate a weighted RE model with clustered standard errors?

Thanks

[Scott Merryman]

-mixed- will cluster the standard errors if you used pweights, or you could use -vce(cluster var)- if you use fweights. For example:

. webuse nlswork,clear
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. 
. drop if grade == 0
(21 observations deleted)

. 
. mixed ln_wage age  [pw= grade] || idcode: ,   stddev   pw(grade) nolog pwscale(size)

Mixed-effects regression                        Number of obs     =     28,487
Group variable: idcode                          Number of groups  =      4,700
                                                Obs per group:
                                                              min =          1
                                                              avg =        6.1
                                                              max =         15
                                                Wald chi2(1)      =    1060.42
Log pseudolikelihood = -144650.68               Prob > chi2       =     0.0000

                             (Std. err. adjusted for 4,700 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   .0190463   .0005849    32.56   0.000        .0179    .0201927
       _cons |   1.146902   .0164237    69.83   0.000     1.114712    1.179092
------------------------------------------------------------------------------

------------------------------------------------------------------------------
                             |               Robust           
  Random-effects parameters  |   Estimate   std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
idcode: Identity             |
                   sd(_cons) |   .3643255   .0051169      .3544333    .3744938
-----------------------------+------------------------------------------------
                sd(Residual) |    .305044   .0037289      .2978223    .3124409
------------------------------------------------------------------------------

[Carlo Lazzaro]

Mywish:
welcome to this forum.
As an aside to Scott's helpful reply, you may want to consider -mixed- cousin -xtreg, re mle- (that allows -iweight-, though):

. use "https://www.stata-press.com/data/r17/nlswork.dta"
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. xtreg ln_wage age [iweight = nev_mar], mle

Fitting constant-only model:
Iteration 0:   log likelihood = -2896.4449
Iteration 1:   log likelihood = -2879.4203
Iteration 2:   log likelihood = -2879.0314
Iteration 3:   log likelihood = -2879.0311

Fitting full model:
Iteration 0:   log likelihood = -2323.9484
Iteration 1:   log likelihood = -2319.7536
Iteration 2:   log likelihood = -2319.6975
Iteration 3:   log likelihood = -2319.6975

Random-effects ML regression                        Number of obs    =   6,548
Group variable: idcode                              Number of groups =   1,972

Random effects u_i ~ Gaussian                       Obs per group:
                                                                 min =       1
                                                                 avg =     3.3
                                                                 max =      15

                                                    LR chi2(1)       = 1118.67
Log likelihood = -2319.6975                         Prob > chi2      =  0.0000

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   .0290693   .0008328    34.90   0.000      .027437    .0307016
       _cons |   .8884023   .0213848    41.54   0.000     .8464889    .9303158
-------------+----------------------------------------------------------------
    /sigma_u |   .3279617   .0068033                      .3148948    .3415708
    /sigma_e |   .2732508   .0028178                      .2677834    .2788298
         rho |   .5902535   .0117491                       .567083     .613113
------------------------------------------------------------------------------
LR test of sigma_u=0: chibar2(01) = 3267.48            Prob >= chibar2 = 0.000

.

[Mywish Maredia]

Thanks, Scott and Carlo for these helpful tips. Will try these options, and read more about them to decide which one is best for the data and study objective.

Thanks again.
Mywish

[Carlo Lazzaro]

Mywish:
take a look at -help weight- for a detailed description of different -weight- allowed.
For instance -pweight- (as per Scott's helpful example) differ from -iweight- (and -mixed- support -fweight-, too, whereas neither .xtreg, re mle-, nor -mixed- support -aweight-).