Chapter 7, Section 7.6

 

Exercise Therapy Trial

Unstructured Covariance (REML Estimation)

 

use "exercise.dta"

 

reshape long y, i(id) j(day)

 

drop if (day==2 | day==10)

 

xtmixed y i.group##i.day || id: , noconst /// 

     residuals(unstructured, t(day)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -446.73626  (not concave)

Iteration 1:   log restricted-likelihood = -336.60468  (not concave)

Iteration 2:   log restricted-likelihood = -316.61872  (not concave)

Iteration 3:   log restricted-likelihood = -308.25196  

Iteration 4:   log restricted-likelihood = -304.16332  

Iteration 5:   log restricted-likelihood = -300.16614  

Iteration 6:   log restricted-likelihood = -298.69424  

Iteration 7:   log restricted-likelihood = -298.66997  

Iteration 8:   log restricted-likelihood = -298.66995  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       173

Group variable: id                              Number of groups   =        37

 

                                                Obs per group: min =         3

                                                               avg =       4.7

                                                               max =         5

 

 

                                                Wald chi2(9)       =     43.31

Log restricted-likelihood = -298.66995          Prob > chi2        =    0.0000

 

------------------------------------------------------------------------------

           y |      Coef.   StdErr.      z    P>|z|     [95% Conf. Interval]

-------------+----------------------------------------------------------------

     2.group |   1.360119   1.031825     1.32   0.187    -.6622198    3.382458

             |

         day |

          4  |      1.125    .341683     3.29   0.001     .4553136    1.794686

          6  |   1.360171   .3873741     3.51   0.000      .600932    2.119411

          8  |   1.584576   .5048459     3.14   0.002     .5950964    2.574056

         12  |   1.623562   .5537523     2.93   0.003     .5382271    2.708896

             |

   group#day |

       2  4  |   -.169159   .4548987    -0.37   0.710    -1.060744     .722426

       2  6  |   .2112572   .5123693     0.41   0.680    -.7929682    1.215483

       2  8  |  -.1309186   .6714468    -0.19   0.845     -1.44693    1.185093

       2 12  |    .320518   .7532375     0.43   0.670      -1.1558    1.796836

             |

       _cons |    79.6875   .7773465   102.51   0.000     78.16393    81.21107

------------------------------------------------------------------------------

 

------------------------------------------------------------------------------

  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]

-----------------------------+------------------------------------------------

id:                  (empty) |

-----------------------------+------------------------------------------------

Residual: Unstructured       |

                      sd(e0) |   3.109386    .371632      2.460025    3.930155

                      sd(e4) |   3.542606    .423616      2.802452     4.47824

                      sd(e6) |   3.262157   .3913439       2.57864    4.126854

                      sd(e8) |   3.740391   .4528304        2.9503    4.742067

                     sd(e12) |   3.734224   .4619653        2.9302    4.758867

                 corr(e0,e4) |     .92373   .0248559      .8569743    .9599994

                 corr(e0,e6) |    .884732   .0370517      .7867599    .9392225

                 corr(e0,e8) |   .8437012   .0495937       .714624    .9172148

                corr(e0,e12) |    .810174   .0610429      .6523348    .9006626

                 corr(e4,e6) |   .9597348   .0135117      .9227051    .9792164

                 corr(e4,e8) |   .9493895   .0173758      .9015372    .9743002

                corr(e4,e12) |   .9016954   .0346425      .8068786    .9512169

                 corr(e6,e8) |   .9577076   .0162689      .9108241    .9801979

                corr(e6,e12) |   .9112646   .0304676      .8283414    .9551135

                corr(e8,e12) |    .939416   .0223507      .8764453    .9707927

------------------------------------------------------------------------------

LR test vs. linear regression:      chi2(14) =   296.13   Prob > chi2 = 0.0000

 

Note: The reported degrees of freedom assumes the null hypothesis is not on the

      boundary of the parameter space.  If this is not true, then the reported

      test is conservative.

 

 

 

Autoregressive Covariance (REML Estimation)

 

gen visit=day

 

recode visit (0=1) (4=2) (6=3) (8=4) (12=5)

(visit: 185 changes made)

 

xtmixed y i.group##i.day || id: , noconst residuals(ar 1, t(visit)) reml

Note: time gaps exist in the estimation data

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -446.73626  (not concave)

Iteration 1:   log restricted-likelihood = -320.11973  

Iteration 2:   log restricted-likelihood = -310.59396  

Iteration 3:   log restricted-likelihood = -310.53595  

Iteration 4:   log restricted-likelihood = -310.53589  

Iteration 5:   log restricted-likelihood = -310.53589  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       173

Group variable: id                              Number of groups   =        37

 

                                                Obs per group: min =         3

                                                               avg =       4.7

                                                               max =         5

 

 

                                                Wald chi2(9)       =     39.73

Log restricted-likelihood = -310.53589          Prob > chi2        =    0.0000

 

------------------------------------------------------------------------------

           y |      Coef.   StdErr.      z    P>|z|     [95% Conf. Interval]

-------------+----------------------------------------------------------------

     2.group |   1.360119   1.143159     1.19   0.234    -.8804318     3.60067

             |

         day |

          4  |      1.125   .2978976     3.78   0.000     .5411315    1.708869

          6  |   1.311971    .418601     3.13   0.002     .4915281    2.132414

          8  |   1.534932   .5036625     3.05   0.002      .547772    2.522093

         12  |   1.615939   .5745615     2.81   0.005     .4898188    2.742059

             |

   group#day |

       2  4  |  -.2143965   .3976147    -0.54   0.590    -.9937071     .564914

       2  6  |   .2594575   .5535425     0.47   0.639    -.8254658    1.344381

       2  8  |  -.0477374    .670704    -0.07   0.943    -1.362293    1.266818

       2 12  |   .3589018   .7788361     0.46   0.645    -1.167589    1.885392

             |

       _cons |    79.6875   .8612227    92.53   0.000     77.99953    81.37547

------------------------------------------------------------------------------

 

------------------------------------------------------------------------------

  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]

-----------------------------+------------------------------------------------

id:                  (empty) |

-----------------------------+------------------------------------------------

Residual: AR(1)              |

                         rho |   .9401763   .0145141      .9041454    .9629272

                       sd(e) |   3.444891   .3703569      2.790382    4.252921

------------------------------------------------------------------------------

LR test vs. linear regression:       chi2(1) =   272.40   Prob > chi2 = 0.0000

 

Note: The reported degrees of freedom assumes the null hypothesis is not on the

      boundary of the parameter space.  If this is not true, then the reported

      test is conservative.

 

 

 

Exponential Covariance (REML Estimation)

 

xtmixed y i.group##i.day || id: , noconst residuals(exponential, t(day)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -428.20643  (not concave)

Iteration 1:   log restricted-likelihood = -317.28453  

Iteration 2:   log restricted-likelihood = -309.28173  

Iteration 3:   log restricted-likelihood = -309.27294  

Iteration 4:   log restricted-likelihood = -309.27293  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       173

Group variable: id                              Number of groups   =        37

 

                                                Obs per group: min =         3

                                                               avg =       4.7

                                                               max =         5

 

 

                                                Wald chi2(9)       =     34.63

Log restricted-likelihood = -309.27293          Prob > chi2        =    0.0001

 

------------------------------------------------------------------------------

           y |      Coef.   StdErr.      z    P>|z|     [95% Conf. Interval]

-------------+----------------------------------------------------------------

     2.group |   1.360119   1.143527     1.19   0.234    -.8811522     3.60139

             |

         day |

          4  |      1.125   .3511182     3.20   0.001      .436821    1.813179

          6  |   1.310419   .4279933     3.06   0.002     .4715678    2.149271

          8  |   1.571879   .4890739     3.21   0.001     .6133114    2.530446

         12  |    1.60783   .5898073     2.73   0.006     .4518286    2.763831

             |

   group#day |

       2  4  |   -.210826   .4678108    -0.45   0.652    -1.127718    .7060664

       2  6  |   .2610093    .566663     0.46   0.645    -.8496297    1.371648

       2  8  |  -.0735649   .6507037    -0.11   0.910    -1.348921    1.201791

       2 12  |   .3498045   .8031588     0.44   0.663    -1.224358    1.923967

             |

       _cons |    79.6875   .8614997    92.50   0.000     77.99899    81.37601

------------------------------------------------------------------------------

 

------------------------------------------------------------------------------

  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]

-----------------------------+------------------------------------------------

id:                  (empty) |

-----------------------------+------------------------------------------------

Residual: Exponential        |

                         rho |   .9785563   .0052913      .9653241    .9868081

                       sd(e) |   3.445999   .3689702      2.793671    4.250647

------------------------------------------------------------------------------

LR test vs. linear regression:       chi2(1) =   274.93   Prob > chi2 = 0.0000

 

Note: The reported degrees of freedom assumes the null hypothesis is not on the

      boundary of the parameter space.  If this is not true, then the reported

      test is conservative.