Chapter 5, Section 5.7

 

Treatment of Lead Exposed Children Trial (TLC)

Analysis of Response Profiles assuming Equal Mean Response at Baseline 

 

 

use "tlc.dta"

 

gen baseline = y0

 

reshape long y, i(id) j(time)

(note: j = 0 1 4 6)

 

Data                               wide   ->   long

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

Number of obs.                      100   ->     400

Number of variables                   7   ->       5

j variable (4 values)                     ->   time

xij variables:

                           y0 y1 ... y6   ->   y

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

 

 

gen change = y - baseline

 

gen t1=1.time

 

gen t4=4.time

 

gen t6=6.time

 

xtmixed y  t1 t4 t6 1.trt#(c.t1 c.t4 c.t6) || id: , noconst /// 

     residuals(unstructured, t(time)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -1314.3479  (not concave)

Iteration 1:   log restricted-likelihood = -1242.5977  

Iteration 2:   log restricted-likelihood = -1232.4333  

Iteration 3:   log restricted-likelihood = -1212.0086  

Iteration 4:   log restricted-likelihood = -1209.0474  

Iteration 5:   log restricted-likelihood = -1208.9948  

Iteration 6:   log restricted-likelihood = -1208.9948  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       400

Group variable: id                              Number of groups   =       100

 

                                                Obs per group: min =         4

                                                               avg =       4.0

                                                               max =         4

 

 

                                                Wald chi2(6)       =    296.51

Log restricted-likelihood = -1208.9948          Prob > chi2        =    0.0000

 

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

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

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

          t1 |  -1.644501   .7824025    -2.10   0.036    -3.177981   -.1110198

          t4 |  -2.231356   .8073858    -2.76   0.006    -3.813803   -.6489084

          t6 |  -2.642064   .8864596    -2.98   0.003    -4.379493   -.9046351

             |

    trt#c.t1 |

          1  |    -11.341   1.093118   -10.37   0.000    -13.48347   -9.198527

             |

    trt#c.t4 |

          1  |  -8.765289   1.131264    -7.75   0.000    -10.98253   -6.548052

             |

    trt#c.t6 |

          1  |  -3.119872   1.250775    -2.49   0.013    -5.571346   -.6683972

             |

       _cons |     26.406     .49989    52.82   0.000     25.42623    27.38577

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e0) |     4.9989   .3552555      4.348929    5.746012

                      sd(e1) |   6.649058   .4737002      5.782529     7.64544

                      sd(e4) |   6.872662    .489636      5.976983    7.902564

                      sd(e6) |   7.646416   .5447313      6.649948      8.7922

                 corr(e0,e1) |   .5694734   .0679828      .4215286    .6878887

                 corr(e0,e4) |   .5680163   .0681493      .4197565     .686752

                 corr(e0,e6) |    .575384    .067303      .4287286    .6924937

                 corr(e1,e4) |   .7745661   .0403532      .6825851    .8423895

                 corr(e1,e6) |   .5805742   .0668605      .4346822    .6967734

                 corr(e4,e6) |   .5795825   .0669768      .4334686    .6960039

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

LR test vs. linear regression:       chi2(9) =   210.71   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.

 

test 1.trt#c.t1 1.trt#c.t4 1.trt#c.t6

 

 ( 1)  [y]1.trt#c.t1 = 0

 ( 2)  [y]1.trt#c.t4 = 0

 ( 3)  [y]1.trt#c.t6 = 0

 

           chi2(  3) =  111.94

         Prob > chi2 =    0.0000

 

 

 

 

 

 

 

Analysis of Response Profiles of Changes in Response from Baseline

 

 

 

drop if time==0

(100 observations deleted)

 

 

xtmixed change i.trt##i.time || id: , noconst /// 

     residuals(unstructured, t(time)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -950.18241  

Iteration 1:   log restricted-likelihood =  -928.1055  (backed up)

Iteration 2:   log restricted-likelihood = -916.24057  

Iteration 3:   log restricted-likelihood =  -909.5036  

Iteration 4:   log restricted-likelihood = -909.45736  

Iteration 5:   log restricted-likelihood = -909.45729  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       300

Group variable: id                              Number of groups   =       100

 

                                                Obs per group: min =         3

                                                               avg =       3.0

                                                               max =         3

 

 

                                                Wald chi2(5)       =    129.99

Log restricted-likelihood = -909.45729          Prob > chi2        =    0.0000

 

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

      change |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

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

       1.trt |    -11.406    1.11994   -10.18   0.000    -13.60104   -9.210958

             |

        time |

          4  |  -.5899999   .6427015    -0.92   0.359    -1.849672    .6696719

          6  |     -1.014     .93431    -1.09   0.278    -2.845214    .8172139

             |

    trt#time |

        1 4  |      2.582   .9089172     2.84   0.005      .800555    4.363445

        1 6  |      8.254   1.321314     6.25   0.000     5.664273    10.84373

             |

       _cons |     -1.612   .7919171    -2.04   0.042    -3.164129   -.0598711

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e1) |   5.599699   .3999787      4.868157    6.441171

                      sd(e4) |   5.762338   .4115954      5.009549    6.628249

                      sd(e6) |   6.282794   .4487709      5.462013    7.226914

                 corr(e1,e4) |   .6803765    .054254      .5593101    .7730249

                 corr(e1,e6) |   .3863277   .0859388      .2064783    .5409252

                 corr(e4,e6) |   .3851844   .0860278      .2051922    .5399742

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

LR test vs. linear regression:       chi2(5) =    81.45   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.

 

test 1.trt 1.trt#4.time 1.trt#6.time

 

 ( 1)  [change]1.trt = 0

 ( 2)  [change]1.trt#4.time = 0

 ( 3)  [change]1.trt#6.time = 0

 

           chi2(  3) =  107.79

         Prob > chi2 =    0.0000

 

 

 

 

Analysis of Response Profiles of Adjusted Changes in Response from Baseline

 

 

gen cbaseline=baseline - 26.406

 

xtmixed change cbaseline i.trt##i.time || id: , noconst ///

     residuals(unstructured, t(time)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -947.88576  

Iteration 1:   log restricted-likelihood =  -921.6877  

Iteration 2:   log restricted-likelihood = -912.10282  

Iteration 3:   log restricted-likelihood = -908.82315  

Iteration 4:   log restricted-likelihood = -908.78288  

Iteration 5:   log restricted-likelihood = -908.78281  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       300

Group variable: id                              Number of groups   =       100

 

                                                Obs per group: min =         3

                                                               avg =       3.0

                                                               max =         3

 

 

                                                Wald chi2(6)       =    138.67

Log restricted-likelihood = -908.78281          Prob > chi2        =    0.0000

 

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

      change |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

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

   cbaseline |  -.1954811   .0938967    -2.08   0.037    -.3795152   -.0114471

       1.trt |  -11.35361   1.098486   -10.34   0.000     -13.5066   -9.200618

             |

        time |

          4  |  -.5899999   .6427016    -0.92   0.359    -1.849672     .669672

          6  |     -1.014   .9343096    -1.09   0.278    -2.845213    .8172133

             |

    trt#time |

        1 4  |      2.582   .9089173     2.84   0.005     .8005548    4.363445

        1 6  |      8.254   1.321313     6.25   0.000     5.664273    10.84373

             |

       _cons |  -1.638195   .7766451    -2.11   0.035    -3.160391   -.1159981

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e1) |   5.490989    .393797      4.770952    6.319696

                      sd(e4) |   5.677179   .4065877      4.933684    6.532717

                      sd(e6) |   6.283115   .4504089      5.459541    7.230926

                 corr(e1,e4) |    .669291   .0558931      .5448435    .7648909

                 corr(e1,e6) |   .3765418   .0868946      .1950458    .5331034

                 corr(e4,e6) |    .377343   .0868397      .1959302    .5337819

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

LR test vs. linear regression:       chi2(5) =    78.21   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.

 

test 1.trt 1.trt#4.time 1.trt#6.time

 

 ( 1)  [change]1.trt = 0

 ( 2)  [change]1.trt#4.time = 0

 ( 3)  [change]1.trt#6.time = 0

 

           chi2(  3) =  111.13

         Prob > chi2 =    0.0000