井出草平の研究ノート

固定効果SEM・ランダム効果SEM[Stata]

SEMモデル内に観測されない異質性を組み入れることができる。これを行うために、我々は、観測されない潜在変数を使用する。

このモデルでは、潜在的なUは独立変数と相関し、従属変数に一定の効果を持つ。

. sem (U -> dg02i1, ) (U -> dg02i2, ) (U -> dg02i3, ) (U -> dg02i4, ) (aid1001 -> dg02i1, ) (aid1002 -> dg02i2, ) (aid1003 -> dg02i3, ) (aid1004 -> dg02i4, ), covstruct(lexogenous, diagonal) cov(lexogenous_oexogenous@0) standardized latent(U ) cov( Uaid1001 Uaid1002 Uaid1003 Uaid1004 aid1001aid1002 aid1002aid1003 aid1003aid1004) nocapslatent

Endogenous variables
  Observed: dg02i1 dg02i2 dg02i3 dg02i4

Exogenous variables
  Observed: aid1001 aid1002 aid1003 aid1004
  Latent:   U

Fitting conditional model:
Iteration 0:   log likelihood = -3024.3611  (not concave)
Iteration 1:   log likelihood =  -2790.155  (not concave)
Iteration 2:   log likelihood =  -2777.439  (not concave)
Iteration 3:   log likelihood = -2746.3103  
Iteration 4:   log likelihood = -2735.1368  
Iteration 5:   log likelihood = -2701.0888  
Iteration 6:   log likelihood = -2676.4515  
Iteration 7:   log likelihood =  -2651.019  
Iteration 8:   log likelihood = -2594.8923  
Iteration 9:   log likelihood = -2556.8301  
Iteration 10:  log likelihood = -2548.6717  
Iteration 11:  log likelihood = -2547.9887  
Iteration 12:  log likelihood = -2547.9869  
Iteration 13:  log likelihood = -2547.9869  

Fitting target model:
Iteration 0:   log likelihood = -2547.9869  
Iteration 1:   log likelihood = -2547.9869  

Structural equation model                                  Number of obs = 140
Estimation method: ml

Log likelihood = -2547.9869

 ( 1)  [dg02i1]U = 1
-------------------------------------------------------------------------------------
                    |                 OIM
       Standardized | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
--------------------+----------------------------------------------------------------
Structural          |
  dg02i1            |
            aid1001 |   .0944771   .0248838     3.80   0.000     .0457058    .1432483
                  U |   .9867268   .0087286   113.05   0.000      .969619    1.003835
              _cons |    1.78054   .1363356    13.06   0.000     1.513327    2.047752
  ------------------+----------------------------------------------------------------
  dg02i2            |
            aid1002 |   .0795519   .0238115     3.34   0.001     .0328821    .1262217
                  U |   .9944192   .0076434   130.10   0.000     .9794385      1.0094
              _cons |   1.875981   .1410543    13.30   0.000     1.599519    2.152442
  ------------------+----------------------------------------------------------------
  dg02i3            |
            aid1003 |   .1072014   .0273746     3.92   0.000     .0535481    .1608546
                  U |   .9758766   .0108004    90.36   0.000     .9547082     .997045
              _cons |   1.879241   .1417078    13.26   0.000     1.601499    2.156983
  ------------------+----------------------------------------------------------------
  dg02i4            |
            aid1004 |    .161862   .0319861     5.06   0.000     .0991705    .2245536
                  U |   .9593964   .0161795    59.30   0.000     .9276851    .9911077
              _cons |    1.97645   .1467579    13.47   0.000      1.68881     2.26409
--------------------+----------------------------------------------------------------
       mean(aid1001)|   .3257185   .0867281     3.76   0.000     .1557346    .4957024
       mean(aid1002)|   .3670659   .0873159     4.20   0.000     .1959299    .5382018
       mean(aid1003)|   .4761129   .0891765     5.34   0.000     .3013302    .6508955
       mean(aid1004)|   .3741062   .0874225     4.28   0.000     .2027612    .5454512
--------------------+----------------------------------------------------------------
       var(e.dg02i1)|   .0316635   .0068772                      .0206862    .0484659
       var(e.dg02i2)|   .0244043   .0061581                      .0148826    .0400179
       var(e.dg02i3)|   .0602049   .0116014                      .0412673     .087833
       var(e.dg02i4)|   .1005082    .017815                      .0710112    .1422578
        var(aid1001)|          1          .                             .           .
        var(aid1002)|          1          .                             .           .
        var(aid1003)|          1          .                             .           .
        var(aid1004)|          1          .                             .           .
              var(U)|          1          .                             .           .
--------------------+----------------------------------------------------------------
cov(aid1001,aid1002)|   .8779208   .0193756    45.31   0.000     .8399454    .9158963
cov(aid1001,aid1003)|    .618358   .0521995    11.85   0.000     .5160487    .7206672
cov(aid1001,aid1004)|    .298639   .0769779     3.88   0.000     .1477651    .4495129
      cov(aid1001,U)|  -.0762641    .086005    -0.89   0.375    -.2448307    .0923026
cov(aid1002,aid1003)|    .650063   .0488007    13.32   0.000     .5544153    .7457107
cov(aid1002,aid1004)|   .3680503   .0730669     5.04   0.000     .2248418    .5112587
      cov(aid1002,U)|  -.1238961   .0852441    -1.45   0.146    -.2909715    .0431793
cov(aid1003,aid1004)|   .7115269   .0417278    17.05   0.000      .629742    .7933118
      cov(aid1003,U)|  -.1148602   .0848663    -1.35   0.176    -.2811951    .0514746
      cov(aid1004,U)|  -.1518099   .0834334    -1.82   0.069    -.3153363    .0117165
-------------------------------------------------------------------------------------
LR test of model vs. saturated: chi2(10) = 75.37                 Prob > chi2 = 0.0000

要約:

  • このモデルは、時間とともに変化しない観測されない異質性を表す潜在的なU項を持つ。
  • このUは、独立変数と従属変数と相関している。この相関は時間的に一定である。
  • 我々は、独立変数からの効果を各時間で一定に設定する。
  • また、従属変数の分散も各時刻で一定に設定する。

6.ランダム効果SEM

Random Effect は、U項が独立変数と相関しないことを仮定している。したがって、我々の前のモデルから、それらの共分散を取り除き、REモデルを持つことができる。

. sem (U -> dg02i1, ) (U -> dg02i2, ) (U -> dg02i3, ) (U -> dg02i4, ) (aid1001 -> dg02i1, ) (aid1002 -> dg02i2, ) (aid1003 -> dg02i3, ) (aid1004 -> dg02i4, ), covstruct(lexogenous, diagonal) cov(lexogenous_oexogenous@0) standardized latent(U ) cov( aid1001aid1002 aid1002aid1003 aid1003aid1001 aid1003aid1004 aid1004aid1001 aid1004*aid1002) nocapslatent

Endogenous variables
  Observed: dg02i1 dg02i2 dg02i3 dg02i4

Exogenous variables
  Observed: aid1001 aid1002 aid1003 aid1004
  Latent:   U

Fitting conditional model:
Iteration 0:   log likelihood = -3010.3817  (not concave)
Iteration 1:   log likelihood = -2784.3438  (not concave)
Iteration 2:   log likelihood = -2756.3185  (not concave)
Iteration 3:   log likelihood = -2731.7585  (not concave)
Iteration 4:   log likelihood = -2722.3319  (not concave)
Iteration 5:   log likelihood = -2709.4781  (not concave)
Iteration 6:   log likelihood =  -2703.332  (not concave)
Iteration 7:   log likelihood = -2700.5239  (not concave)
Iteration 8:   log likelihood = -2699.3366  (not concave)
Iteration 9:   log likelihood =  -2698.098  (not concave)
Iteration 10:  log likelihood = -2696.6677  (not concave)
Iteration 11:  log likelihood =  -2693.605  (not concave)
Iteration 12:  log likelihood = -2691.2325  (not concave)
Iteration 13:  log likelihood = -2689.1527  (not concave)
Iteration 14:  log likelihood = -2687.1086  (not concave)
Iteration 15:  log likelihood = -2661.4224  (not concave)
Iteration 16:  log likelihood = -2641.8011  (not concave)
Iteration 17:  log likelihood = -2625.6766  (not concave)
Iteration 18:  log likelihood = -2617.3964  (not concave)
Iteration 19:  log likelihood =   -2589.71  (not concave)
Iteration 20:  log likelihood = -2567.8006  
Iteration 21:  log likelihood = -2557.3664  
Iteration 22:  log likelihood = -2552.2769  
Iteration 23:  log likelihood = -2550.3008  
Iteration 24:  log likelihood = -2550.2726  
Iteration 25:  log likelihood = -2550.2725  

Fitting target model:
Iteration 0:   log likelihood = -2550.2725  
Iteration 1:   log likelihood = -2550.2725  

Structural equation model                                  Number of obs = 140
Estimation method: ml

Log likelihood = -2550.2725

 ( 1)  [dg02i1]U = 1
-------------------------------------------------------------------------------------
                    |                 OIM
       Standardized | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
--------------------+----------------------------------------------------------------
Structural          |
  dg02i1            |
            aid1001 |   .0889541   .0240784     3.69   0.000     .0417612     .136147
                  U |    .980069   .0042394   231.18   0.000       .97176    .9883781
              _cons |   1.771049   .1353193    13.09   0.000     1.505828     2.03627
  ------------------+----------------------------------------------------------------
  dg02i2            |
            aid1002 |   .0732052   .0228726     3.20   0.001     .0283757    .1180347
                  U |   .9851812   .0036744   268.12   0.000     .9779795    .9923829
              _cons |   1.861995   .1400489    13.30   0.000     1.587504    2.136485
  ------------------+----------------------------------------------------------------
  dg02i3            |
            aid1003 |   .1014453   .0263491     3.85   0.000      .049802    .1530886
                  U |   .9650206    .006648   145.16   0.000     .9519908    .9780504
              _cons |   1.861056   .1402512    13.27   0.000     1.586169    2.135944
  ------------------+----------------------------------------------------------------
  dg02i4            |
            aid1004 |   .1546387   .0303409     5.10   0.000     .0951717    .2141057
                  U |   .9380669   .0107013    87.66   0.000     .9170927    .9590411
              _cons |   1.934959   .1430286    13.53   0.000     1.654628     2.21529
--------------------+----------------------------------------------------------------
       mean(aid1001)|   .3257185   .0867281     3.76   0.000     .1557346    .4957024
       mean(aid1002)|   .3670659   .0873159     4.20   0.000     .1959299    .5382018
       mean(aid1003)|   .4761129   .0891765     5.34   0.000     .3013302    .6508955
       mean(aid1004)|   .3741062   .0874225     4.28   0.000     .2027612    .5454512
--------------------+----------------------------------------------------------------
       var(e.dg02i1)|   .0315519   .0068498                      .0206174    .0482855
       var(e.dg02i2)|    .024059   .0060736                      .0146688    .0394603
       var(e.dg02i3)|   .0584441   .0113073                      .0399997    .0853935
       var(e.dg02i4)|   .0961174   .0170179                      .0679351     .135991
        var(aid1001)|          1          .                             .           .
        var(aid1002)|          1          .                             .           .
        var(aid1003)|          1          .                             .           .
        var(aid1004)|          1          .                             .           .
              var(U)|          1          .                             .           .
--------------------+----------------------------------------------------------------
cov(aid1001,aid1002)|   .8779208   .0193756    45.31   0.000     .8399454    .9158963
cov(aid1001,aid1003)|    .618358   .0521995    11.85   0.000     .5160487    .7206672
cov(aid1001,aid1004)|    .298639   .0769779     3.88   0.000     .1477651    .4495129
cov(aid1002,aid1003)|    .650063   .0488007    13.32   0.000     .5544153    .7457107
cov(aid1002,aid1004)|   .3680503   .0730669     5.04   0.000     .2248418    .5112587
cov(aid1003,aid1004)|   .7115269   .0417278    17.05   0.000      .629742    .7933118
-------------------------------------------------------------------------------------
LR test of model vs. saturated: chi2(14) = 79.94                 Prob > chi2 = 0.0000