UCLA Academic Technology Services HomeServicesClassesContactJobs
Search

Stata Textbook Examples
Multilevel Analysis Techniques and Applications by Joop Hox
Chapter 5: Analyzing Longitudinal Data

This chapter uses data file gpach5.dta and vocagrwt.dta.

Table 5.1 on page 78.
use http://www.ats.ucla.edu/stat/stata/examples/mlm_ma_hox/gpach5, clear
keep student sex highgpa gpa time
reshape wide gpa, i(student) j(time)

manova gpa0 gpa1 gpa2 gpa3 gpa4 gpa5 = sex highgpa, cont(highgpa) nocons

mat ymat=(1,0,0,0,0,-1\0,1,0,0,0,-1\0,0,1,0,0,-1\0,0,0,1,0,-1\0,0,0,0,1,-1)
*GPA as categorical
mat xcat = (.5, .5, 0)
manovatest, test(xcat) ytransform(ymat)

 Transformations of the dependent variables
 (1)    gpa0 - gpa5
 (2)    gpa1 - gpa5
 (3)    gpa2 - gpa5
 (4)    gpa3 - gpa5
 (5)    gpa4 - gpa5

 Test constraint
 (1)    .5 sex[1] + .5 sex[2] = 0

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
              manovatest | W   0.8950      1     5.0   193.0     4.53 0.0006 e
                         | P   0.1050            5.0   193.0     4.53 0.0006 e
                         | L   0.1173            5.0   193.0     4.53 0.0006 e
                         | R   0.1173            5.0   193.0     4.53 0.0006 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F
*linear trend
mat ylin =(-5,-3,-1,1,3,5)
manovatest, test(xcat) ytransform(ylin)

 Transformation of the dependent variables
 (1)    -5 gpa0 - 3 gpa1 - gpa2 + gpa3 + 3 gpa4 + 5 gpa5

 Test constraint
 (1)    .5 sex[1] + .5 sex[2] = 0

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
              manovatest | W   0.9391      1     1.0   197.0    12.77 0.0004 e
                         | P   0.0609            1.0   197.0    12.77 0.0004 e
                         | L   0.0648            1.0   197.0    12.77 0.0004 e
                         | R   0.0648            1.0   197.0    12.77 0.0004 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F

*gpa*highgpa
manovatest highgpa, ytransform(ymat)

 Transformations of the dependent variables
 (1)    gpa0 - gpa5
 (2)    gpa1 - gpa5
 (3)    gpa2 - gpa5
 (4)    gpa3 - gpa5
 (5)    gpa4 - gpa5

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
                 highgpa | W   0.9781      1     5.0   193.0     0.87 0.5053 e
                         | P   0.0219            5.0   193.0     0.87 0.5053 e
                         | L   0.0224            5.0   193.0     0.87 0.5053 e
                         | R   0.0224            5.0   193.0     0.87 0.5053 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F
*gpa*gender
mat xsex = (1, -1, 0)
manovatest, test(xsex) ytransform(ymat) 

 Transformations of the dependent variables
 (1)    gpa0 - gpa5
 (2)    gpa1 - gpa5
 (3)    gpa2 - gpa5
 (4)    gpa3 - gpa5
 (5)    gpa4 - gpa5

 Test constraint
 (1)    sex[1] - sex[2] = 0

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
              manovatest | W   0.9646      1     5.0   193.0     1.42 0.2196 e
                         | P   0.0354            5.0   193.0     1.42 0.2196 e
                         | L   0.0367            5.0   193.0     1.42 0.2196 e
                         | R   0.0367            5.0   193.0     1.42 0.2196 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F
*test highgpa
mat ym = 1/6*(1,1,1,1,1,1)
manovatest highgpa, ytransform(ym)

 Transformation of the dependent variables
 (1)    .1666667 gpa0 + .1666667 gpa1 + .1666667 gpa2 + .1666667 gpa3 +
        .1666667 gpa4 + .1666667 gpa5

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
                 highgpa | W   0.9556      1     1.0   197.0     9.16 0.0028 e
                         | P   0.0444            1.0   197.0     9.16 0.0028 e
                         | L   0.0465            1.0   197.0     9.16 0.0028 e
                         | R   0.0465            1.0   197.0     9.16 0.0028 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F

*test sex
manovatest, test(xsex) ytransform(ym)

 Transformation of the dependent variables
 (1)    .1666667 gpa0 + .1666667 gpa1 + .1666667 gpa2 + .1666667 gpa3 +
        .1666667 gpa4 + .1666667 gpa5

 Test constraint
 (1)    sex[1] - sex[2] = 0

                           W = Wilks' lambda      L = Lawley-Hotelling trace
                           P = Pillai's trace     R = Roy's largest root

                  Source |  Statistic     df   F(df1,    df2) =   F   Prob>F
              -----------+--------------------------------------------------
              manovatest | W   0.9147      1     1.0   197.0    18.37 0.0000 e
                         | P   0.0853            1.0   197.0    18.37 0.0000 e
                         | L   0.0933            1.0   197.0    18.37 0.0000 e
                         | R   0.0933            1.0   197.0    18.37 0.0000 e
                         |--------------------------------------------------
                Residual |               197
              --------------------------------------------------------------
                           e = exact, a = approximate, u = upper bound on F

Table 5.2 on page 79.
use http://www.ats.ucla.edu/stat/stata/examples/mlm_ma_hox/gpach5, clear
collapse (mean) gpa, by(time sex)
table sex time, contents(mean gpa) col row format(%3.1f)
-----------------------------------------------------------
student   |                      time                      
gender    |     0      1      2      3      4      5  Total
----------+------------------------------------------------
     male |   2.6    2.7    2.7    2.8    2.9    3.0    2.8
   female |   2.6    2.8    2.9    3.0    3.1    3.2    2.9
          | 
    Total |   2.6    2.7    2.8    2.9    3.0    3.1    2.9
-----------------------------------------------------------

Figure 5.3 on page 80.

use http://www.ats.ucla.edu/stat/stata/examples/mlm_ma_hox/gpach5, clear
histogram gpa, normal frequency width(.25) start(1.625) xlabel(1.75(.25)4)

Table 5.3 on page 81.
Part 1: Null model.
xtmixed gpa ||student:, var ml

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -456.72811  
Iteration 1:   log likelihood = -456.72811  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(0)       =         .
Log likelihood = -456.72811                     Prob > chi2        =         .

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |      2.865   .0191093   149.93   0.000     2.827546    2.902454
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Identity            |
                  var(_cons) |   .0567683   .0073395      .0440612    .0731402
-----------------------------+------------------------------------------------
               var(Residual) |     .09759   .0043644      .0894002    .1065301
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =   249.82 Prob >= chibar2 = 0.0000


estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .   -456.7281      3     919.4562    934.7265
------------------------------------------------------------------------------
Part 2: With additional variable time.
xtmixed gpa time ||student:, var ml

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -196.82458  
Iteration 1:   log likelihood = -196.82458  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(1)       =    681.70
Log likelihood = -196.82458                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1063143   .0040719    26.11   0.000     .0983336     .114295
       _cons |   2.599214   .0216516   120.05   0.000     2.556778    2.641651
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Identity            |
                  var(_cons) |   .0633616   .0073161       .050529    .0794532
-----------------------------+------------------------------------------------
               var(Residual) |   .0580305   .0025952      .0531605    .0633465
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =   481.33 Prob >= chibar2 = 0.0000


estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .   -196.8246      4     401.6492    422.0095
------------------------------------------------------------------------------
Part 3: The variable job is added.
xtmixed gpa time job ||student:, var ml

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood =  -160.1304  
Iteration 1:   log likelihood =  -160.1304  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(2)       =    788.20
Log likelihood =  -160.1304                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |    .103166   .0040021    25.78   0.000     .0953219      .11101
         job |  -.1608636    .018356    -8.76   0.000    -.1968408   -.1248864
       _cons |   2.945837    .044462    66.26   0.000     2.858693    3.032981
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Identity            |
                  var(_cons) |   .0533597   .0063712      .0422258    .0674291
-----------------------------+------------------------------------------------
               var(Residual) |   .0556077    .002494      .0509282    .0607171
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =   388.27 Prob >= chibar2 = 0.0000

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .   -160.1304      5     330.2608    355.7112
------------------------------------------------------------------------------
Part 4: The variable highgpa and sex are added to the model.
xtmixed gpa time job highgpa sex ||student:, var ml

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -147.27246  
Iteration 1:   log likelihood = -147.27246  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(4)       =    820.13
Log likelihood = -147.27246                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1031482   .0040024    25.77   0.000     .0953037    .1109926
         job |  -.1617732   .0183036    -8.84   0.000    -.1976476   -.1258988
     highgpa |   .0858379   .0279769     3.07   0.002     .0310042    .1406717
         sex |   .1483967   .0333128     4.45   0.000     .0831048    .2136886
       _cons |   2.613448   .0982039    26.61   0.000     2.420972    2.805924
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Identity            |
                  var(_cons) |   .0457555   .0056063      .0359872    .0581754
-----------------------------+------------------------------------------------
               var(Residual) |    .055617   .0024948      .0509361    .0607281
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =   333.27 Prob >= chibar2 = 0.0000


estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .   -147.2725      7     308.5449    344.1755
------------------------------------------------------------------------------

Table 5.4 on page 83.

Part 1: The variable time is included as a random effect.
xtmixed gpa time job highgpa sex ||student: time, var ml cov(un)

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -90.102483  
Iteration 1:   log likelihood = -90.102468  
Iteration 2:   log likelihood = -90.102468  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(4)       =    431.48
Log likelihood = -90.102468                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1039731   .0056223    18.49   0.000     .0929537    .1149925
         job |  -.1196211   .0174584    -6.85   0.000    -.1538388   -.0854033
     highgpa |   .0898354   .0264746     3.39   0.001     .0379462    .1417246
         sex |   .1167606   .0315324     3.70   0.000     .0549583     .178563
       _cons |   2.527287   .0926005    27.29   0.000     2.345793     2.70878
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                   var(time) |   .0039121   .0006455       .002831    .0054059
                  var(_cons) |   .0389692   .0062212      .0284991    .0532858
             cov(time,_cons) |  -.0025635   .0015582     -.0056175    .0004905
-----------------------------+------------------------------------------------
               var(Residual) |   .0417646   .0020993      .0378462    .0460887
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   447.61   Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .   -90.10247      9     198.2049    244.0156
------------------------------------------------------------------------------
Part 2: Cross level interaction of variable time and sex is included. We first created the interaction term.
gen sxtime= sex*time
xtmixed gpa time job highgpa sex sxtime ||student: time, var ml cov(un)

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -86.602037  
Iteration 1:   log likelihood = -86.602002  
Iteration 2:   log likelihood = -86.602002  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(5)       =    452.34
Log likelihood = -86.602002                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .0884798   .0080081    11.05   0.000     .0727841    .1041754
         job |  -.1206129   .0174249    -6.92   0.000    -.1547651   -.0864606
     highgpa |   .0898068   .0264668     3.39   0.001     .0379327    .1416808
         sex |   .0767958   .0348988     2.20   0.028     .0083954    .1451962
      sxtime |   .0294742   .0110399     2.67   0.008     .0078363     .051112
       _cons |   2.550491   .0928935    27.46   0.000     2.368423    2.732559
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                   var(time) |   .0036912   .0006242      .0026499    .0051418
                  var(_cons) |   .0385556   .0061494      .0282049    .0527048
             cov(time,_cons) |   -.002273   .0015061     -.0052248    .0006789
-----------------------------+------------------------------------------------
               var(Residual) |   .0417767   .0021004      .0378562    .0461031
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   446.52   Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |   1200           .     -86.602     10      193.204    244.1048
------------------------------------------------------------------------------

Figure 5.4 on page 84 based on the previous model.

gen p = 2.55 + .088*time + .0768*sex + .029*sxtime
sort sex time
label variable p "Predicted GPA"
twoway scatter p time, connect(L) ylabel(2.5(.1)3.3)

Table 5.5 on page 85.
tab time
       time |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |        200       16.67       16.67
          1 |        200       16.67       33.33
          2 |        200       16.67       50.00
          3 |        200       16.67       66.67
          4 |        200       16.67       83.33
          5 |        200       16.67      100.00
------------+-----------------------------------
      Total |      1,200      100.00
Part 1: 1st occasion = 0, same as the first part of Table 5.4.
xtmixed gpa time job highgpa sex ||student: time, var ml cov(un)


------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1039731   .0056223    18.49   0.000     .0929537    .1149925
         job |  -.1196211   .0174584    -6.85   0.000    -.1538388   -.0854033
     highgpa |   .0898354   .0264746     3.39   0.001     .0379462    .1417246
         sex |   .1167606   .0315324     3.70   0.000     .0549583     .178563
       _cons |   2.527287   .0926005    27.29   0.000     2.345793     2.70878
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                   var(time) |   .0039121   .0006455       .002831    .0054059
                  var(_cons) |   .0389692   .0062212      .0284991    .0532858
             cov(time,_cons) |  -.0025635   .0015582     -.0056175    .0004905
-----------------------------+------------------------------------------------
               var(Residual) |   .0417646   .0020993      .0378462    .0460887
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   447.61   Prob > chi2 = 0.0000
 

Part 2: The variable time has been recoded as -5, ...,-1, 0, ...with the last occasion coded as zero. We first recode variable time into time1.

gen time1 = time -5
xtmixed gpa time1 job highgpa sex ||student: time1, var ml cov(un)
Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -90.102485  
Iteration 1:   log likelihood = -90.102468  
Iteration 2:   log likelihood = -90.102468  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(4)       =    431.48
Log likelihood = -90.102468                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       time1 |   .1039731   .0056223    18.49   0.000     .0929537    .1149926
         job |  -.1196211   .0174584    -6.85   0.000    -.1538388   -.0854033
     highgpa |   .0898354   .0264746     3.39   0.001     .0379462    .1417246
         sex |   .1167606   .0315324     3.70   0.000     .0549583     .178563
       _cons |   3.047152   .0938291    32.48   0.000      2.86325    3.231054
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                  var(time1) |   .0039121   .0006455       .002831    .0054059
                  var(_cons) |   .1111359   .0137347      .0872286    .1415956
            cov(time1,_cons) |   .0169968    .002676      .0117519    .0222417
-----------------------------+------------------------------------------------
               var(Residual) |   .0417646   .0020993      .0378462    .0460887
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   447.61   Prob > chi2 = 0.0000

Part 3: The variable time is recoded centered around its mean and is included as a fixed effect.

gen timec = time - 2.5
xtmixed gpa timec job highgpa sex ||student: timec, var ml cov(un)

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -90.102706  
Iteration 1:   log likelihood = -90.102468  
Iteration 2:   log likelihood = -90.102468  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =      1200
Group variable: student                         Number of groups   =       200

                                                Obs per group: min =         6
                                                               avg =       6.0
                                                               max =         6


                                                Wald chi2(4)       =    431.48
Log likelihood = -90.102468                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       timec |   .1039731   .0056223    18.49   0.000     .0929537    .1149926
         job |  -.1196211   .0174584    -6.85   0.000    -.1538388   -.0854033
     highgpa |   .0898354   .0264746     3.39   0.001     .0379462    .1417246
         sex |   .1167606   .0315324     3.70   0.000     .0549583     .178563
       _cons |   2.787219    .092151    30.25   0.000     2.606607    2.967832
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                  var(timec) |   .0039121   .0006455       .002831    .0054059
                  var(_cons) |   .0506022   .0058846      .0402885     .063556
            cov(timec,_cons) |   .0072167   .0014799      .0043162    .0101172
-----------------------------+------------------------------------------------
               var(Residual) |   .0417646   .0020993      .0378462    .0460887
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   447.61   Prob > chi2 = 0.0000

Table 5.6 using data file vocagrwt.dta.

use http://www.ats.ucla.edu/stat/stata/examples/mlm_ma_hox/vocagrwt, clear

table age, contents (count child)

----------------------
age in    |
months    |   N(child)
----------+-----------
       12 |         22
       14 |          5
       16 |         22
       18 |         11
       20 |         22
       22 |         11
       24 |         22
       26 |         11
----------------------

Table 5.7 on page 89. We have to recode the variable study as follows.
Part 1: Intercept only model.
gen study1 = study - .5
xtmixed vocab study1 ||child:, var ml cov(un)

Note: single-variable random-effects specification; covariance structure set to
      identity

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -834.45287  
Iteration 1:   log likelihood = -834.45002  
Iteration 2:   log likelihood = -834.45001  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(1)       =      8.08
Log likelihood = -834.45001                     Prob > chi2        =    0.0045

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |   -110.912    39.0169    -2.84   0.004    -187.3838   -34.44031
       _cons |   132.0697   19.50845     6.77   0.000     93.83379    170.3055
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   2400.597    2148.08      415.5804    13867.03
-----------------------------+------------------------------------------------
               var(Residual) |   31075.49   4201.565      23841.37    40504.64
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =     1.94 Prob >= chibar2 = 0.0821
Part 2: The variable age is grand mean centered and is included as a fixed effect.
sum age

    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
         age |       126    18.88889    4.578598         12         26

gen agec=age-r(mean)
xtmixed vocab study1 agec ||child:, var ml cov(un)

Note: single-variable random-effects specification; covariance structure set to
      identity

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -775.46764  
Iteration 1:   log likelihood = -775.46764  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(2)       =    225.19
Log likelihood = -775.46764                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -69.79256   35.47381    -1.97   0.049    -139.3199   -.2651797
        agec |   29.55533   2.013816    14.68   0.000     25.60832    33.50233
       _cons |   137.7813   17.68648     7.79   0.000     103.1165    172.4462
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   4882.778   1937.101      2243.765    10625.68
-----------------------------+------------------------------------------------
               var(Residual) |   10377.41   1421.189      7934.449    13572.54
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =    26.44 Prob >= chibar2 = 0.0000
Part 3: The squared term of agec is included as a fixed effect.
gen agec2 = agec*agec
xtmixed vocab study1 agec agec2 ||child:, var ml cov(un)

Note: single-variable random-effects specification; covariance structure set to
      identity

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -759.74092  
Iteration 1:   log likelihood = -759.74092  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(3)       =    336.38
Log likelihood = -759.74092                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -68.25089   34.85923    -1.96   0.050    -136.5737    .0719371
        agec |   30.62189   1.746006    17.54   0.000     27.19978    34.04399
       agec2 |   2.557313    .422708     6.05   0.000     1.728821    3.385806
       _cons |   84.79167   19.46981     4.36   0.000     46.63154    122.9518
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   5166.262   1902.714      2510.049    10633.36
-----------------------------+------------------------------------------------
               var(Residual) |   7721.242   1060.154      5899.485    10105.55
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =    38.85 Prob >= chibar2 = 0.0000

Part 4: The centered variable agec is included as a random effect.

xtmixed vocab study1 agec agec2 ||child: agec, var ml cov(un)

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -651.93737  
Iteration 1:   log likelihood = -651.40036  
Iteration 2:   log likelihood = -651.37911  
Iteration 3:   log likelihood = -651.37642  
Iteration 4:   log likelihood = -651.37628  
Iteration 5:   log likelihood = -651.37626  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(3)       =    288.74
Log likelihood = -651.37626                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -4.971305   8.003023    -0.62   0.534    -20.65694    10.71433
        agec |   28.13162   3.399128     8.28   0.000     21.46945    34.79379
       agec2 |   2.175139   .1454073    14.96   0.000     1.890146    2.460132
       _cons |     87.982   17.52353     5.02   0.000     53.63651    122.3275
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Unstructured          |
                   var(agec) |   245.7646   76.12805      133.9218    451.0114
                  var(_cons) |   6399.886   1995.074      3473.939    11790.23
             cov(agec,_cons) |   1254.139   384.1707      501.1785      2007.1
-----------------------------+------------------------------------------------
               var(Residual) |   862.6386   119.6392      657.3185    1132.092
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   255.58   Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference

Table 5.8 on page 91.
Part 1: Intercept only model. This is Part 1 from Table 5.7. We only show the output here.
------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |   -110.912    39.0169    -2.84   0.004    -187.3838   -34.44031
       _cons |   132.0697   19.50845     6.77   0.000     93.83379    170.3055
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   2400.597    2148.08      415.5804    13867.03
-----------------------------+------------------------------------------------
               var(Residual) |   31075.49   4201.565      23841.37    40504.64
------------------------------------------------------------------------------

Part 2: The variable age is centered on 12 months and is included as a fixed effect.

xtmixed vocab study1 age12 ||child: , var ml cov(un)

Note: single-variable random-effects specification; covariance structure set to
      identity

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -775.46764  
Iteration 1:   log likelihood = -775.46764  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(2)       =    225.19
Log likelihood = -775.46764                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -69.79256   35.47381    -1.97   0.049    -139.3199   -.2651795
       age12 |   29.55533   2.013816    14.68   0.000     25.60832    33.50233
       _cons |  -65.82204    22.2324    -2.96   0.003    -109.3967   -22.24734
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   4882.778   1937.101      2243.765    10625.68
-----------------------------+------------------------------------------------
               var(Residual) |   10377.41   1421.189      7934.449    13572.54
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =    26.44 Prob >= chibar2 = 0.0000

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |    126           .   -775.4676      5     1560.935    1575.117
------------------------------------------------------------------------------
Part 3: The variable age12sq is included as a fixed effect.
xtmixed vocab study1 age12 age12sq ||child: , var ml cov(un)

Note: single-variable random-effects specification; covariance structure set to
      identity

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood = -759.74092  
Iteration 1:   log likelihood = -759.74092  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(3)       =    336.38
Log likelihood = -759.74092                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -68.25089   34.85923    -1.96   0.050    -136.5737    .0719375
       age12 |  -4.612211   5.910617    -0.78   0.435    -16.19681    6.972386
     age12sq |   2.557313    .422708     6.05   0.000     1.728821    3.385806
       _cons |  -4.797205   23.22992    -0.21   0.836    -50.32702    40.73261
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Identity              |
                  var(_cons) |   5166.262   1902.714      2510.049    10633.36
-----------------------------+------------------------------------------------
               var(Residual) |   7721.242   1060.154      5899.485    10105.55
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =    38.85 Prob >= chibar2 = 0.0000

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |    126           .   -759.7409      6     1531.482      1548.5
------------------------------------------------------------------------------
Part 4: The variable age12 is a random effect.
xtmixed vocab study1 age12 age12sq ||child: age12, var ml cov(un)

Performing EM optimization: 

Performing gradient-based optimization: 

Iteration 0:   log likelihood =  -652.0315  
Iteration 1:   log likelihood = -651.50087  
Iteration 2:   log likelihood = -651.40437  
Iteration 3:   log likelihood = -651.38281  
Iteration 4:   log likelihood = -651.37783  
Iteration 5:   log likelihood = -651.37663  
Iteration 6:   log likelihood = -651.37632  
Iteration 7:   log likelihood = -651.37627  
Iteration 8:   log likelihood = -651.37626  

Computing standard errors:

Mixed-effects ML regression                     Number of obs      =       126
Group variable: child                           Number of groups   =        22

                                                Obs per group: min =         4
                                                               avg =       5.7
                                                               max =         8


                                                Wald chi2(3)       =    288.73
Log likelihood = -651.37626                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       vocab |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      study1 |  -4.975021   8.002868    -0.62   0.534    -20.66035    10.71031
       age12 |  -1.836976   3.891349    -0.47   0.637     -9.46388    5.789927
     age12sq |    2.17514   .1454086    14.96   0.000     1.890144    2.460136
       _cons |  -2.588417   8.341687    -0.31   0.756    -18.93782    13.76099
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
child: Unstructured          |
                  var(age12) |   245.7569   76.12895      133.9141    451.0091
                  var(_cons) |   784.0285   361.7215       317.409     1936.62
            cov(age12,_cons) |  -438.9538   159.5462     -751.6586   -126.2489
-----------------------------+------------------------------------------------
               var(Residual) |   862.6541   119.6425      657.3286    1132.116
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   255.58   Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference

estat ic

------------------------------------------------------------------------------
       Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
-------------+----------------------------------------------------------------
           . |    126           .   -651.3763      8     1318.753    1341.443
------------------------------------------------------------------------------

Table 5.9 on page 92 is created using HLM. We omit it here.

Table 5.10 on page 101 using the data file gpach5.dta.
use http://www.ats.ucla.edu/stat/stata/examples/mlm_ma_hox/gpach5, clear
Part 1: The variable time is a fixed effect. We have built the model at the beginning of this chapter. We will use it here.
xtmixed gpa time job highgpa sex ||student: , var ml

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1031482   .0040024    25.77   0.000     .0953037    .1109926
         job |  -.1617732   .0183036    -8.84   0.000    -.1976476   -.1258988
     highgpa |   .0858379   .0279769     3.07   0.002     .0310042    .1406717
         sex |   .1483967   .0333128     4.45   0.000     .0831048    .2136886
       _cons |   2.613448   .0982039    26.61   0.000     2.420972    2.805924
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Identity            |
                  var(_cons) |   .0457555   .0056063      .0359872    .0581754
-----------------------------+------------------------------------------------
               var(Residual) |    .055617   .0024948      .0509361    .0607281
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =   333.27 Prob >= chibar2 = 0.0000
Part 2: The variable time is included as random effect. This is Part 1 of table 5.4.
xtmixed gpa time job highgpa sex ||student: time, var ml cov(un)

------------------------------------------------------------------------------
         gpa |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        time |   .1039731   .0056223    18.49   0.000     .0929537    .1149925
         job |  -.1196211   .0174584    -6.85   0.000    -.1538388   -.0854033
     highgpa |   .0898354   .0264746     3.39   0.001     .0379462    .1417246
         sex |   .1167606   .0315324     3.70   0.000     .0549583     .178563
       _cons |   2.527287   .0926005    27.29   0.000     2.345793     2.70878
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
student: Unstructured        |
                   var(time) |   .0039121   .0006455       .002831    .0054059
                  var(_cons) |   .0389692   .0062212      .0284991    .0532858
             cov(time,_cons) |  -.0025635   .0015582     -.0056175    .0004905
-----------------------------+------------------------------------------------
               var(Residual) |   .0417646   .0020993      .0378462    .0460887
------------------------------------------------------------------------------
LR test vs. linear regression:       chi2(3) =   447.61   Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference
Part 3: The variable time is a fixed effect, MANOVA.

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