Chapter 6, Section 6.5

 

Vlagtwedde-Vlaardingen Study

Linear Trend Model (REML Estimation) 

 

 

use "smoking.dta"

(771 observations read)

 

xtmixed fev1 1.smoker c.time 1.smoker#c.time || id: , ///

     noconst residuals(unstructured, t(time)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -659.26555  (not concave)

Iteration 1:   log restricted-likelihood =  -222.6972  

Iteration 2:   log restricted-likelihood = -161.53664  

Iteration 3:   log restricted-likelihood = -136.44052  

Iteration 4:   log restricted-likelihood = -133.29021  

Iteration 5:   log restricted-likelihood = -133.02603  

Iteration 6:   log restricted-likelihood = -133.02532  

Iteration 7:   log restricted-likelihood = -133.02532  

 

Computing standard errors:

 

Mixed-effects REML regression                   Number of obs      =       771

Group variable: id                              Number of groups   =       133

 

                                                Obs per group: min =         1

                                                               avg =       5.8

                                                               max =         7

 

 

                                                Wald chi2(3)       =    607.67

Log restricted-likelihood = -133.02532          Prob > chi2        =    0.0000

 

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

        fev1 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

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

    1.smoker |  -.2616996      .1151    -2.27   0.023    -.4872913   -.0361078

        time |  -.0332243   .0030663   -10.84   0.000    -.0392342   -.0272144

             |

      smoker#|

      c.time |

          1  |  -.0049984   .0035254    -1.42   0.156     -.011908    .0019112

             |

       _cons |   3.507313   .1003784    34.94   0.000     3.310575    3.704051

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e0) |   .5959408    .039439      .5234452     .678477

                      sd(e3) |   .5828185   .0368568       .514878    .6597241

                      sd(e6) |   .5667042   .0356325      .5009976    .6410283

                      sd(e9) |   .5691593   .0364624      .5019989    .6453048

                     sd(e12) |   .5594311   .0352272      .4944776    .6329167

                     sd(e15) |   .5768604   .0376883      .5075266     .655666

                     sd(e19) |   .5481524   .0357412      .4823924    .6228768

                 corr(e0,e3) |   .8628108   .0247147      .8057155     .904019

                 corr(e0,e6) |   .8457978   .0275915      .7822737     .891912

                 corr(e0,e9) |   .8378473   .0286658      .7720474    .8858768

                corr(e0,e12) |   .8553072   .0258347      .7957719    .8984684

                corr(e0,e15) |   .8389877   .0329113      .7615506    .8928072

                corr(e0,e19) |   .8308471   .0309197      .7595637    .8824091

                 corr(e3,e6) |    .887808   .0201635      .8410899    .9213779

                 corr(e3,e9) |   .8329119    .029272      .7658603    .8820443

                corr(e3,e12) |   .8619242   .0239013      .8070688    .9020284

                corr(e3,e15) |   .8740405   .0225603       .821894    .9116596

                corr(e3,e19) |   .8237995   .0301766      .7549978    .8746602

                 corr(e6,e9) |   .8333251   .0284496      .7684131    .8812644

                corr(e6,e12) |   .8900114   .0197849      .8441506     .922941

                corr(e6,e15) |   .8677585   .0235932      .8133048    .9071429

                corr(e6,e19) |   .8336594   .0294319      .7661302    .8829817

                corr(e9,e12) |    .885758   .0199613      .8397366    .9191447

                corr(e9,e15) |   .8762306   .0216853      .8262728    .9125104

                corr(e9,e19) |   .8396783   .0283297      .7746383    .8871413

               corr(e12,e15) |   .9320125   .0135812      .8997387    .9541482

               corr(e12,e19) |   .8576585   .0252635      .7994772    .8998977

               corr(e15,e19) |    .892769   .0197004      .8469221    .9254391

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

LR test vs. linear regression:      chi2(27) =  1052.48   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.

 

 

 

 

 

Linear Trend Model (ML Estimation) 

 

 

xtmixed fev1 1.smoker c.time 1.smoker#c.time || id: , noconst /// 

     residuals(unstructured, t(time))

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log likelihood = -645.39358  (not concave)

Iteration 1:   log likelihood = -210.49385  

Iteration 2:   log likelihood = -208.15603  (not concave)

Iteration 3:   log likelihood =  -162.1537  (not concave)

Iteration 4:   log likelihood = -127.84853  

Iteration 5:   log likelihood = -120.88432  

Iteration 6:   log likelihood = -119.24672  

Iteration 7:   log likelihood = -119.23047  

Iteration 8:   log likelihood = -119.23046  

Iteration 9:   log likelihood = -119.23046  

 

Computing standard errors:

 

Mixed-effects ML regression                     Number of obs      =       771

Group variable: id                              Number of groups   =       133

 

                                                Obs per group: min =         1

                                                               avg =       5.8

                                                               max =         7

 

 

                                                Wald chi2(3)       =    617.75

Log likelihood = -119.23046                     Prob > chi2        =    0.0000

 

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

        fev1 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

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

    1.smoker |  -.2617388   .1142359    -2.29   0.022     -.485637   -.0378405

        time |  -.0332263    .003042   -10.92   0.000    -.0391884   -.0272641

             |

      smoker#|

      c.time |

          1  |  -.0050053   .0034973    -1.43   0.152    -.0118598    .0018493

             |

       _cons |    3.50742   .0996254    35.21   0.000     3.312157    3.702682

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e0) |   .5919617   .0389915      .5202673    .6735359

                      sd(e3) |   .5789156    .036411      .5117746    .6548649

                      sd(e6) |   .5628814   .0351886      .4979707    .6362531

                      sd(e9) |   .5653843   .0360434      .4989758    .6406312

                     sd(e12) |    .555653   .0347946      .4914757    .6282107

                     sd(e15) |   .5731412   .0372955      .5045127    .6511052

                     sd(e19) |   .5439858   .0352392      .4791229    .6176297

                 corr(e0,e3) |   .8610213    .024942      .8034508    .9026394

                 corr(e0,e6) |   .8440068   .0277964      .7800691    .8905002

                 corr(e0,e9) |   .8362326   .0288426      .7700804    .8845926

                corr(e0,e12) |   .8542686   .0259267      .7945642    .8976115

                corr(e0,e15) |   .8386578   .0329311      .7611986     .892525

                corr(e0,e19) |   .8309296   .0307681      .7600454    .8822747

                 corr(e3,e6) |   .8863249   .0203588      .8391913    .9202426

                 corr(e3,e9) |   .8309423   .0295061      .7634115    .8805036

                corr(e3,e12) |   .8606053   .0240416      .8054691    .9009713

                corr(e3,e15) |   .8733229   .0226091      .8210989    .9110465

                corr(e3,e19) |   .8230869   .0301631      .7543687    .8739615

                 corr(e6,e9) |   .8311253   .0287065      .7656864    .8795354

                corr(e6,e12) |   .8888206    .019931      .8426548    .9220144

                corr(e6,e15) |   .8665699   .0237259      .8118484    .9062002

                corr(e6,e19) |   .8324892   .0295209      .7648086    .8819958

                corr(e9,e12) |   .8842779   .0201471      .8378642    .9179978

                corr(e9,e15) |    .874868   .0218493      .8245692    .9114453

                corr(e9,e19) |   .8381261   .0284945      .7727609    .8858992

               corr(e12,e15) |   .9311458   .0137308      .8985317    .9535336

               corr(e12,e19) |   .8559657   .0254601      .7973831    .8985661

               corr(e15,e19) |   .8914829    .019863      .8452951    .9244458

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

LR test vs. linear regression:      chi2(27) =  1052.33   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.

 

 

Quadratic Trend Model (ML Estimation) 

 

 

 

xtmixed fev1 1.smoker c.time##c.time 1.smoker#(c.time##c.time) || id: , ///

     noconst residuals(unstructured, t(time))

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log likelihood = -645.04414  (not concave)

Iteration 1:   log likelihood = -210.20495  

Iteration 2:   log likelihood = -153.11538  

Iteration 3:   log likelihood =  -121.4061  

Iteration 4:   log likelihood = -118.64164  

Iteration 5:   log likelihood = -118.59129  

Iteration 6:   log likelihood =  -118.5912  

Iteration 7:   log likelihood =  -118.5912  

 

Computing standard errors:

 

Mixed-effects ML regression                     Number of obs      =       771

Group variable: id                              Number of groups   =       133

 

                                                Obs per group: min =         1

                                                               avg =       5.8

                                                               max =         7

 

 

                                                Wald chi2(5)       =    617.54

Log likelihood =  -118.5912                     Prob > chi2        =    0.0000

 

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

        fev1 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

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

    1.smoker |  -.2977847   .1183755    -2.52   0.012    -.5297965    -.065773

        time |  -.0413393   .0100488    -4.11   0.000    -.0610346    -.021644

             |

      c.time#|

      c.time |   .0004106   .0004852     0.85   0.397    -.0005403    .0013615

             |

      smoker#|

      c.time |

          1  |     .00755   .0114822     0.66   0.511    -.0149548    .0300547

             |

      smoker#|

      c.time#|

      c.time |

          1  |  -.0006422   .0005591    -1.15   0.251    -.0017381    .0004537

             |

       _cons |   3.531199   .1034022    34.15   0.000     3.328534    3.733863

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e0) |   .5914796   .0388846      .5199728    .6728199

                      sd(e3) |   .5783574   .0363607      .5113075    .6541999

                      sd(e6) |   .5619766   .0350744      .4972704    .6351026

                      sd(e9) |   .5652513   .0360951      .4987543    .6406142

                     sd(e12) |   .5555015   .0347735      .4913617    .6280137

                     sd(e15) |   .5732584   .0373328      .5045646    .6513044

                     sd(e19) |    .543566   .0351889      .4787933    .6171014

                 corr(e0,e3) |    .860158    .025138      .8021264    .9020949

                 corr(e0,e6) |    .847731   .0274739      .7843756    .8935814

                 corr(e0,e9) |   .8336349   .0295184      .7658776    .8830809

                corr(e0,e12) |   .8549655   .0258526      .7954075    .8981692

                corr(e0,e15) |   .8357665   .0337025      .7564526     .890856

                corr(e0,e19) |   .8314704   .0306323      .7609075    .8825966

                 corr(e3,e6) |   .8876317   .0201856      .8408673    .9212414

                 corr(e3,e9) |   .8299552   .0297678      .7618028    .8799369

                corr(e3,e12) |   .8613552   .0239414      .8064328    .9015433

                corr(e3,e15) |    .873009   .0226647      .8206589    .9108264

                corr(e3,e19) |   .8222123   .0303305      .7531145    .8733681

                 corr(e6,e9) |   .8303305   .0288483      .7645729    .8789808

                corr(e6,e12) |   .8879501   .0201156      .8413509    .9214458

                corr(e6,e15) |   .8670327   .0236408      .8125054    .9065202

                corr(e6,e19) |    .835163   .0290713      .7684814    .8838998

                corr(e9,e12) |   .8842953   .0201423      .8378929    .9180077

                corr(e9,e15) |   .8743164   .0219771      .8237153    .9111009

                corr(e9,e19) |   .8356393   .0292867      .7683506    .8846571

               corr(e12,e15) |   .9318293    .013622      .8994575    .9540305

               corr(e12,e19) |   .8565223   .0253863      .7980964     .898991

               corr(e15,e19) |   .8905087   .0200509       .843886    .9237833

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

LR test vs. linear regression:      chi2(27) =  1052.91   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.

 

 

 

 

 

Treatment of Lead Exposed Children (TLC) Trial

Piecewise Linear Model (REML Estimation)

 

 

use "tlc.dta"

 

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 time_1=max(time-1, 0)

 

xtmixed y c.time c.time_1 1.trt#c.time 1.trt#c.time_1 || id: , /// 

     noconst residuals(unstructured, t(time)) reml

 

Obtaining starting values by EM: 

 

Performing gradient-based optimization: 

 

Iteration 0:   log restricted-likelihood = -1321.7656  (not concave)

Iteration 1:   log restricted-likelihood = -1247.7923  

Iteration 2:   log restricted-likelihood = -1235.3283  

Iteration 3:   log restricted-likelihood = -1220.0722  

Iteration 4:   log restricted-likelihood = -1218.7419  

Iteration 5:   log restricted-likelihood = -1218.7262  

Iteration 6:   log restricted-likelihood = -1218.7262  

 

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(4)       =    285.30

Log restricted-likelihood = -1218.7262          Prob > chi2        =    0.0000

 

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

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

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

        time |  -1.629603   .7817556    -2.08   0.037    -3.161816   -.0973901

      time_1 |   1.430495   .8777538     1.63   0.103     -.289871    3.150861

             |

  trt#c.time |

          1  |     -11.25   1.092447   -10.30   0.000    -13.39116   -9.108842

             |

trt#c.time_1 |

          1  |   12.58226   1.227847    10.25   0.000     10.17573     14.9888

             |

       _cons |   26.34221   .4991175    52.78   0.000     25.36396    27.32046

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

 

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

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

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

id:                  (empty) |

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

Residual: Unstructured       |

                      sd(e0) |   4.999233   .3553283      4.349136    5.746506

                      sd(e1) |   6.648487   .4736068      5.782122    7.644664

                      sd(e4) |    6.92402   .4981448      6.013386    7.972555

                      sd(e6) |    7.71917   .5567123      6.701643     8.89119

                 corr(e0,e1) |   .5693051   .0680191      .4212831    .6877835

                 corr(e0,e4) |   .5597697   .0696783       .408335    .6812185

                 corr(e0,e6) |   .5735463   .0678934      .4255836    .6916397

                 corr(e1,e4) |   .7676784   .0420108      .6718727    .8382179

                 corr(e1,e6) |   .5760139   .0682007      .4271723    .6944607

                 corr(e4,e6) |   .5527423    .070034      .4008575     .675055

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

LR test vs. linear regression:       chi2(9) =   206.08   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.