UCLA Academic Technology Services HomeServicesClassesContactJobs
Search

SAS Textbook Examples
Modeling Longitudinal Data by Robert Weiss
Chapter 12: Missing Data

The data files can be downloaded from http://rem.ph.ucla.edu/~rob/mld/data.html .

Table 12.1, page 367.

*Complete data;
proc mixed data = missingdata1 method = reml noinfo noclprint noitprint;
 class idno;
 model weight = week/ solution notest;
 random intercept/ subject = idno type = un;
run;

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      idno         714.01
Residual                 10.1683

           Fit Statistics
-2 Res Log Likelihood           735.4
AIC (smaller is better)         739.4
AICC (smaller is better)        739.5
BIC (smaller is better)         741.4

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     1        384.48          <.0001

                   Solution for Fixed Effects
                         Standard
Effect       Estimate       Error      DF    t Value    Pr > |t|
Intercept      196.46      5.9972      19      32.76      <.0001
week           0.1414     0.08522      99       1.66      0.1002

*MAR;
proc mixed data = missingdata2 method = reml noclprint noitprint noinfo;
 class idno;
 model weight = week/ solution notest;
 random intercept/ subject = idno type =un;
run;

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      idno         700.56
Residual                  7.9755

           Fit Statistics
-2 Res Log Likelihood           560.5
AIC (smaller is better)         564.5
AICC (smaller is better)        564.6
BIC (smaller is better)         566.5

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     1        301.54          <.0001

                   Solution for Fixed Effects
                         Standard
Effect       Estimate       Error      DF    t Value    Pr > |t|
Intercept      196.84      5.9377      19      33.15      <.0001
week         -0.05619     0.09671      70      -0.58      0.5631

*MNAR;
proc mixed data = missingdata3 method = reml noclprint noitprint noinfo;
 class idno;
 model weight = week/ solution notest;
 random intercept/ subject = idno type =un;
run;

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      idno         707.88
Residual                  6.0963

           Fit Statistics
-2 Res Log Likelihood           490.2
AIC (smaller is better)         494.2
AICC (smaller is better)        494.4
BIC (smaller is better)         496.2

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     1        280.88          <.0001

                   Solution for Fixed Effects
                         Standard
Effect       Estimate       Error      DF    t Value    Pr > |t|
Intercept      196.29      5.9666      19      32.90      <.0001
week          -0.2007     0.08905      60      -2.25      0.0279
Table 12.2, page 370.
proc sql;
 create table schiz as
  select *,
  (max(week) ne 6) as drop_out,
  sqrt(week) as sweek
 from schizophrenia
 group by id
 order by drop_out, id;
quit;

*Pooled;
proc mixed data = schiz method = ml noitprint noclprint noinfo;
 class id ;
 model severity = sweek drug sweek*drug / solution notest;
 random intercept sweek/ subject=id type = un;
run;

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      ID           0.3687
UN(2,1)      ID          0.02085
UN(2,2)      ID           0.2420
Residual                  0.5778

           Fit Statistics
-2 Log Likelihood              4649.0
AIC (smaller is better)        4665.0
AICC (smaller is better)       4665.1
BIC (smaller is better)        4697.6

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     3        542.43          <.0001

                   Solution for Fixed Effects
                          Standard
Effect        Estimate       Error      DF    t Value    Pr > |t|
Intercept       5.3480     0.08790     435      60.84      <.0001
sweek          -0.3361     0.06794     435      -4.95      <.0001
drug           0.04634      0.1011     729       0.46      0.6469
sweek*drug     -0.6405     0.07752     729      -8.26      <.0001

*By dropout;
proc mixed data = schiz method = ml noitprint noclprint noinfo;
 by drop_out;
 class id ;
 model severity = sweek drug sweek*drug / solution notest;
 random intercept sweek/ subject=id type = un;
run;

drop_out=0

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      ID           0.3984
UN(2,1)      ID         -0.01110
UN(2,2)      ID           0.2048
Residual                  0.5600

           Fit Statistics
-2 Log Likelihood              3782.1
AIC (smaller is better)        3798.1
AICC (smaller is better)       3798.2
BIC (smaller is better)        3828.6

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     3        468.38          <.0001

                   Solution for Fixed Effects
                          Standard
Effect        Estimate       Error      DF    t Value    Pr > |t|
Intercept       5.2212      0.1091     333      47.84      <.0001
sweek          -0.3934     0.07345     333      -5.36      <.0001
drug            0.2016      0.1226     655       1.64      0.1008
sweek*drug     -0.5386     0.08255     655      -6.52      <.0001


drop_out=1

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      ID           0.2858
UN(2,1)      ID          0.02045
UN(2,2)      ID           0.6921
Residual                  0.5147

           Fit Statistics
-2 Log Likelihood               822.9
AIC (smaller is better)         838.9
AICC (smaller is better)        839.5
BIC (smaller is better)         859.9

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     3         73.32          <.0001

                   Solution for Fixed Effects
                          Standard
Effect        Estimate       Error      DF    t Value    Pr > |t|
Intercept       5.5337      0.1409     100      39.26      <.0001
sweek          -0.1235      0.1773     100      -0.70      0.4877
drug           -0.1939      0.1779      74      -1.09      0.2791
sweek*drug     -1.1674      0.2239      74      -5.21      <.0001

*Covariate and contrasts;
proc mixed data = schiz method = ml noitprint noclprint noinfo;
 class id ;
 model severity = sweek drug sweek*drug drop_out drop_out*sweek 
   drop_out*drug drop_out*sweek*drug / solution notest;
 random intercept sweek/ subject=id type = un;
 estimate 'int placebo nodrop' intercept 1;
 estimate 'int placebo drop'   intercept 1 drop_out 1;
 estimate 'int drug nodrop'    intercept 1 drug 1;
 estimate 'int drug drop'      intercept 1 drug 1 drop_out 1;
 estimate 'slope placebo nodrop' sweek 1;
 estimate 'slope placebo drop' sweek 1 drop_out*sweek 1;
 estimate 'slope drug nodrop' sweek 1 sweek*drug 1;
 estimate 'slope drug drop' sweek 1 sweek*drug 1 drop_out*sweek 1 drop_out*sweek*drug 1;
run;

The Mixed Procedure

 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
UN(1,1)      ID           0.3612
UN(2,1)      ID          0.01175
UN(2,2)      ID           0.2300
Residual                  0.5768

           Fit Statistics
-2 Log Likelihood              4623.3
AIC (smaller is better)        4647.3
AICC (smaller is better)       4647.5
BIC (smaller is better)        4696.2

  Null Model Likelihood Ratio Test
    DF    Chi-Square      Pr > ChiSq
     3        523.49          <.0001

                        Solution for Fixed Effects
                                    Standard
Effect                 Estimate       Error      DF    t Value    Pr > |t|
Intercept                5.2210      0.1075     433      48.55      <.0001
sweek                   -0.3934     0.07635     433      -5.15      <.0001
drug                     0.2017      0.1208     729       1.67      0.0955
sweek*drug              -0.5386     0.08580     729      -6.28      <.0001
drop_out                 0.3203      0.1864     729       1.72      0.0862
sweek*drop_out           0.2517      0.1593     729       1.58      0.1146
drug*drop_out           -0.3987      0.2270     729      -1.76      0.0794
sweek*drug*drop_out     -0.6348      0.1960     729      -3.24      0.0013

                                 Estimates
                                    Standard
Label                   Estimate       Error      DF    t Value    Pr > |t|
int placebo nodrop        5.2210      0.1075     433      48.55      <.0001
int placebo drop          5.5413      0.1523     729      36.39      <.0001
int drug nodrop           5.4227     0.05511     729      98.39      <.0001
int drug drop             5.7430      0.1944     729      29.54      <.0001
slope placebo nodrop     -0.3934     0.07635     433      -5.15      <.0001
slope placebo drop       -0.1417      0.1399     729      -1.01      0.3115
slope drug nodrop        -0.9320     0.03916     729     -23.80      <.0001
slope drug drop          -1.3150      0.1073     729     -12.26      <.0001
Figure 12.2, page 371.
proc mixed data = schiz method = ml noitprint noclprint noinfo;
 class id ;
 model severity = sweek drug sweek*drug drop_out drop_out*sweek 
   drop_out*drug drop_out*sweek*drug / solution notest outpredm=pred_schiz;;
 random intercept sweek/ subject=id type = un;
run;

proc sql;
 create table pred_schiz2 as
  select mean(pred) as pred,
  drug, drop_out, week
  from pred_schiz
  group by week, drop_out, drug
  order by drop_out, drug, week;
 quit;

data pred_schiz2;
 set pred_schiz2;
 group=compress(drug||drop_out);
  if group = "01" then group = "placebo drop";
  if group = "00" then group = "placebo nodrop";
  if group = "11" then group = "drug drop";
  if group = "10" then group = "drug nodrop";
run; 

goptions reset = all;
 symbol1 value=none interpol=join r=4;
 axis1 order =(0 to 6 by 1 ) label=(a=0 'Week') minor=none;
 axis2 order =(2 to 6 by 1) label = (a=90 'Illness severity') minor=none ;

proc gplot data = pred_schiz2;
 plot pred*week=group /haxis=axis1 vaxis = axis2;
run; quit; 

How to cite this page

Report an error on this page

UCLA Researchers are invited to our Statistical Consulting Services
We recommend others to our list of Other Resources for Statistical Computing Help
These pages are Copyrighted (c) by UCLA Academic Technology Services


The content of this web site should not be construed as an endorsement of any particular web site, book, or software product by the University of California