MI Estimate Problem

[Davia Downey]

Has anyone run into the problem of mi estimate NOT bringing in the newly created observations in a model? These are the commands so you can see the structure of the data.

mi misstable patterns $ylist $xlist

            Missing-value patterns
               (1 means complete)

              |   Pattern
    Percent   |  1  2  3  4    5  6  7  8    9
  ------------+--------------------------------
       14%    |  1  1  1  1    1  1  1  1    1
              |
       64     |  1  1  1  0    0  0  0  0    0
       11     |  1  0  0  0    0  0  0  0    0
        7     |  0  1  1  1    0  0  0  0    0
        1     |  0  0  0  1    0  0  0  0    0
        1     |  1  0  0  1    0  0  0  0    0
        1     |  1  0  0  1    1  1  1  1    1
  ------------+--------------------------------
      100%    |

mi register imputed log_accessions pct_maori deprivationindex log_level6 pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd

mi register regular funding policyscore

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
log_access~s |      3,024    9.940385    .6635382   8.234169   11.85503
   pct_maori |      2,906    .1795299    .1433115  -.3225543   .7145595
deprivatio~x |      2,922    7.440636    9.274504  -21.92738   45.97867
  log_level6 |      2,906    9.275616    1.420161    3.15082   15.58429
pctbachelors |      2,906    45.69963    5.780976   20.04118   69.70129
-------------+---------------------------------------------------------
  pctmasters |      2,906    8.057414    3.164138  -2.479166   21.62572
pctdoctorate |      2,906    2.482619    2.351187  -6.384878   12.85397
    turnover |      3,024    16.32612    2.038624   10.07816   21.82732
unemployme~d |      3,034    5.312884    1.585719   .1699752   12.83865
 policyscore |      3,048    .5964567    .4906884          0          1
-------------+---------------------------------------------------------
     funding |      3,048    .0485564    .2149741          0          1

mi estimate: xtreg log_accessions pct_maori log_level6 deprivationindex pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd funding policyscore, fe i(region)

Multiple-imputation estimates                   Imputations       =         20
Fixed-effects (within) regression               Number of obs     =        168

Group variable: region                          Number of groups  =         14
                                                Obs per group:
                                                              min =         12
                                                              avg =       12.0
                                                              max =         12
                                                Average RVI       =     2.6497
                                                Largest FMI       =     0.8825
                                                Complete DF       =        144
DF adjustment:   Small sample                   DF:     min       =      10.94
                                                        avg       =      26.36
                                                        max       =      99.99
Model F test:       Equal FMI                   F(  10,   87.8)   =       1.18
Within VCE type: Conventional                   Prob > F          =     0.3127

-----------------------------------------------------------------------------------------
         log_accessions |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
------------------------+----------------------------------------------------------------
              pct_maori |   .1583934   .4750647     0.33   0.742    -.8195537    1.136341
             log_level6 |   .0427818   .0647897     0.66   0.519    -.0947651    .1803286
       deprivationindex |   .0111996   .0088192     1.27   0.226    -.0078017    .0302009
           pctbachelors |  -.0013473   .0161937    -0.08   0.935    -.0356404    .0329458
             pctmasters |    .027238   .0284658     0.96   0.349    -.0316455    .0861216
           pctdoctorate |  -.0306891   .0358613    -0.86   0.406    -.1071646    .0457863
               turnover |  -.0258731   .0439013    -0.59   0.564    -.1193276    .0675814
unemploymentrate15_oecd |  -.0646297   .0379148    -1.70   0.100    -.1425142    .0132547
                funding |   .0458863   .1544445     0.30   0.767    -.2605275    .3523001
            policyscore |  -.0347175   .0677549    -0.51   0.612    -.1732905    .1038555
                  _cons |   10.13297   1.211425     8.36   0.000      7.46498    12.80097
------------------------+----------------------------------------------------------------
                sigma_u |  .48764784
                sigma_e |   .2173891
                    rho |  .83421637   (fraction of variance due to u_i)
-----------------------------------------------------------------------------------------
Note: sigma_u and sigma_e are combined in the original metric.

I'm stumped (again) and starting to think mi estimate is not all its cracked up to be.

[Rich Goldstein]

I'm confused (maybe I missed some previous posting), but where is your -mi impute ... - command?

[Davia Downey]

My two impute commands are below.
For variables with missing data:

mi register imputed log_accessions pct_maori deprivationindex log_level6 pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd

For variables with full data (note these are dummies, so I'm not sure if I actually need these to be set as regular:

mi register regular funding policyscore

[Rich Goldstein]

these commands do NOT impute any data, they just register the variables; please see

help mi impute

[Davia Downey]

The mi impute command I used is:

mi impute mvn log_accessions pct_maori deprivationindex log_level6 pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd, add(20) rseed(1234)

I've also tried the following mi impute commands:

mi impute mvn log_accessions pct_maori deprivationindex log_level6 pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd cera_funding_assistance eqc_funding= totalregionalemployment mean_earn_nzstat employmentrate15_oecd, force add(20) rseed(1234)

 mi impute chained (regress) log_accessions pct_maori deprivationindex log_level6 pctbachelors pctmasters pctdoctorate turnover unemploymentrate15_oecd cera_funding_assistance eqc_funding= totalregionalemployment mean_earn_nzstat employmentrate15_oecd, force add(20) rseed(1234)

It does not matter which method I use (although mvn is appropriate here because log_accessions is continuous, not binary), when the mi estimate command is run it still only estimates the original 168 observations and not the imputed values.