UCLA Academic Technology Services HomeServicesClassesContactJobs
Search

SAS Paper Examples
How to Deal with "The Language-as-Fixed Effect Fallacy": Common Misconceptions and Alternative Solutions
By Raaijmakers, Schrijnemakers and Gremmen

This article appeared in Journal of Memory and Language (1999, vol. 41) and it can be accessed directly from a UCLA IP address. Every example here has been done both in proc glm and proc mixed to show the difference and connections between the two procedures. In most of the examples, we have shown two or more different styles of proc mixed syntax. One is from random effect ANOVA approach and the others are from multilevel model approach. To see more on the differences, visit SAS FAQ page on reproducing proc glm analysis with proc mixed. Also notice that the all the examples here are done with balanced data. The SAS code shown here may not work on unbalanced data.


Table 2, the data set for Table 3 and the computation that follows.

data m1;
input ID  ITEM  Y  TREAT @@;
datalines;
 1      1     546      0      1      2     567      0
 1      3     547      0      1      4     566      0
 1      5     554      1      1      6     545      1
 1      7     594      1      1      8     522      1
 2      1     566      0      2      2     556      0
 2      3     538      0      2      4     566      0
 2      5     512      1      2      6     523      1
 2      7     569      1      2      8     524      1
 3      1     567      0      3      2     598      0
 3      3     568      0      3      4     584      0
 3      5     536      1      3      6     539      1
 3      7     589      1      3      8     521      1
 4      1     556      0      4      2     565      0
 4      3     536      0      4      4     550      0
 4      5     516      1      4      6     522      1
 4      7     560      1      4      8     486      1
 5      1     595      0      5      2     609      0
 5      3     585      0      5      4     588      0
 5      5     578      1      5      6     540      1
 5      7     615      1      5      8     546      1
 6      1     569      0      6      2     578      0
 6      3     560      0      6      4     583      0
 6      5     501      1      6      6     535      1
 6      7     568      1      6      8     514      1
 7      1     527      0      7      2     554      0
 7      3     535      0      7      4     527      0
 7      5     480      1      7      6     467      1
 7      7     540      1      7      8     473      1
 8      1     551      0      8      2     575      0
 8      3     558      0      8      4     556      0
 8      5     588      1      8      6     563      1
 8      7     631      1      8      8     558      1
 ;
 run;

Table 3 and the calculation of F' on page 418 using the data above.

There are two forms of quasi F-ratio. One can find the definition for both in Kirk's Experimental Design (page 406-408). For example, for a CRF-pqr design, assuming a random effects model, F' and F'' for testing treatments B and C have the following form:

F' = MSB/(MSAB + MSBC - MSABC)                F'' = (MSB + MSABC)/(MSAB + MSBC)

F' = MSC(MSAC + MSBC - MSABC)                F'' =(MSA + MSABC)/(MSAC + MSBC)

One problem with F' is that it can be possibly negative since the denominator can be negative. F'' is a modification of F' to get around this problem.

 In this article, the authors actually used the second form of quasi F-ratio, that is F'' in the definition above. Both proc glm and proc mixed will give F' in the definition above instead. This is why we will see slight discrepancy between the article and the output from SAS.

proc glm data= m1;
  class id item treat;
  model y = treat item(treat) id treat*id /ss3;
run;
quit;
Dependent Variable: Y
                                        Sum of
Source                      DF         Squares     Mean Square    F Value    Pr > F
Model                       21     64045.45313      3049.78348      30.43    <.0001
Error                       42      4208.78125       100.20908
Corrected Total             63     68254.23438
R-Square     Coeff Var      Root MSE        Y Mean
0.938337      1.813128      10.01045      552.1094
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      8032.64063      8032.64063      80.16    <.0001
ITEM(TREAT)                  6     22174.46875      3695.74479      36.88    <.0001
ID                           7     26251.60938      3750.22991      37.42    <.0001
ID*TREAT                     7      7586.73437      1083.81920      10.82    <.0001

Page 419 left column, subject analysis. Notice that we have two different syntax for proc mixed. They produce identical results. The first one is conceptualized as random effect ANOVA model and the second one is conceptualized as 2-level multilevel model.

proc glm data = m1;
  class id item treat;
  model y = treat id id*treat ;
  random id id*treat /test;
run;
quit;
The GLM Procedure
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1     8032.640625     8032.640625       7.41    0.0297
ID                           7           26252     3750.229911       3.46    0.0618
Error: MS(ID*TREAT)          7     7586.734375     1083.819196
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID*TREAT                     7     7586.734375     1083.819196       1.97    0.0787
Error: MS(Error)            48           26383      549.651042
proc mixed data = m1  ;
  class id item treat;
  model y = treat ;
  random  id id*treat;
run;
proc mixed data = m1  ;
  class id item treat;
  model y = treat ;
  random  intercept treat /subject=id ;
run;
 Covariance Parameter Estimates
Cov Parm      Subject    Estimate
Intercept     ID           333.30
TREAT         ID           133.54
Residual                   549.65
           Fit Statistics
-2 Res Log Likelihood           592.3
AIC (smaller is better)         598.3
AICC (smaller is better)        598.7
BIC (smaller is better)         598.5
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       7       7.41    0.0297

Page 419 left column, item analysis. Notice that we have two different syntax for proc mixed. They produce identical results. The first one is conceptualized as random effect ANOVA model and the second one is conceptualized as 2-level multilevel model.

proc glm data = m1  ;
  class id item treat;
  model y = treat item*treat;
  random item*treat /test ;
run;
quit;
The GLM Procedure
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1     8032.640625     8032.640625       2.17    0.1908
Error                        6           22174     3695.744792
Error: MS(ITEM*TREAT)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ITEM*TREAT                   6           22174     3695.744792       5.44    0.0002
Error: MS(Error)            56           38047      679.412946
proc mixed data = m1  ;
  class id item treat;
  model y = treat ;
  random item*treat  ;
run;
proc mixed data = m1 ;
  class id item treat;
  model y = treat ;
  random treat /subject=item ;
run;
The Mixed Procedure
 Covariance Parameter
       Estimates
Cov Parm       Estimate
ITEM*TREAT       377.04
Residual         679.41
           Fit Statistics
-2 Res Log Likelihood           597.4
AIC (smaller is better)         601.4
AICC (smaller is better)        601.6
BIC (smaller is better)         601.5
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       6       2.17    0.1908

Page 419 left column, combined analysis using proc mixed or proc glm. Notice that we have a couple of different syntax for proc mixed. They produce identical results.

proc glm data = m1 ;
  class id item treat;
  model y = treat  item(treat) id id*treat / e3;
  random item(treat) id id*treat  /test;
run;
quit;
The GLM Procedure
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1     8032.640625     8032.640625       1.72    0.2227
Error                   8.9575           41915     4679.354911
Error: MS(ITEM(TREAT)) + MS(ID*TREAT) - MS(Error)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ITEM(TREAT)                  6           22174     3695.744792      36.88    <.0001
ID*TREAT                     7     7586.734375     1083.819196      10.82    <.0001
Error: MS(Error)            42     4208.781250      100.209077
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID                           7           26252     3750.229911       3.46    0.0618
Error: MS(ID*TREAT)          7     7586.734375     1083.819196
proc mixed data = m1 method=type3;
  class id item treat;
  model y = treat  ;
  random item(treat) id id*treat ;
run;
The Mixed Procedure
                               Type 3 Analysis of Variance
                                                               Error
Source       Error Term                                           DF    F Value    Pr > F
TREAT        MS(ITEM(TREAT)) + MS(ID*TREAT) - MS(Residual)    8.9575       1.72    0.2227
ITEM(TREAT)  MS(Residual)                                         42      36.88    <.0001
ID           MS(ID*TREAT)                                          7       3.46    0.0618
ID*TREAT     MS(Residual)                                         42      10.82    <.0001
Residual     .                                                     .        .       .
  Covariance Parameter
       Estimates
Cov Parm        Estimate
ITEM(TREAT)       449.44
ID                333.30
ID*TREAT          245.90
Residual          100.21
           Fit Statistics
-2 Res Log Likelihood           532.2
AIC (smaller is better)         540.2
AICC (smaller is better)        540.9
BIC (smaller is better)         540.5
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       6       1.72    0.2381
proc mixed data = m1 ;
  class id item treat;
  model y = treat  / solution DDFM=KENWARDROGER ; /*kenwardroger is required to get the right degrees of freedom*/
  random treat intercept /subject=id;
  repeated id /subject=item type=cs;
run;
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1    8.96       1.72    0.2227
proc mixed data = m1 ;
  class id item treat;
  model y = treat  / solution DDFM=KENWARDROGER ;
  random treat intercept /subject=id;
  random treat /subject=item ;
run;
Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1    8.96       1.72    0.2227

Table 4 on page 422. This data set is used in the following sections for multiple examples.

data m2;
 input   ID    TREAT    BLOCK     Y @@;
datalines;
  1      1        1      493  1      1        2      519
  1      1        3      513  1      1        4      542
  1      2        1      499  1      2        2      525
  1      2        3      502  1      2        4      557
  2      1        1      562  2      1        2      552
  2      1        3      565  2      1        4      591
  2      2        1      544  2      2        2      536
  2      2        3      533  2      2        4      563
  3      1        1      519  3      1        2      558
  3      1        3      555  3      1        4      567
  3      2        1      575  3      2        2      582
  3      2        3      551  3      2        4      587
  4      1        1      518  4      1        2      523
  4      1        3      514  4      1        4      563
  4      2        1      523  4      2        2      565
  4      2        3      539  4      2        4      597
  5      1        1      567  5      1        2      562
  5      1        3      577  5      1        4      595
  5      2        1      521  5      2        2      563
  5      2        3      559  5      2        4      575
  6      1        1      520  6      1        2      534
  6      1        3      527  6      1        4      568
  6      2        1      512  6      2        2      541
  6      2        3      531  6      2        4      559
  7      1        1      516  7      1        2      544
  7      1        3      513  7      1        4      575
  7      2        1      555  7      2        2      569
  7      2        3      550  7      2        4      601
  8      1        1      525  8      1        2      528
  8      1        3      528  8      1        4      559
  8      2        1      551  8      2        2      542
  8      2        3      529  8      2        4      578
;
run;

Table 6 using data in Table 4.

proc glm data = m2;
  class id block treat;
  model y = treat block id treat*block treat*id block*id /ss3;
run;
quit;
Dependent Variable: Y
                                        Sum of
Source                      DF         Squares     Mean Square    F Value    Pr > F
Model                       42     40110.00000       955.00000       9.35    <.0001
Error                       21      2143.93750       102.09226
Corrected Total             63     42253.93750
R-Square     Coeff Var      Root MSE        Y Mean
0.949261      1.847285      10.10407      546.9688
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1       770.06250       770.06250       7.54    0.0121
BLOCK                        3     16985.31250      5661.77083      55.46    <.0001
ID                           7     12758.18750      1822.59821      17.85    <.0001
BLOCK*TREAT                  3       321.31250       107.10417       1.05    0.3916
ID*TREAT                     7      6254.68750       893.52679       8.75    <.0001
ID*BLOCK                    21      3020.43750       143.83036       1.41    0.2194

Page 422 right column, subject analysis..

proc glm data = m2;
  class id block treat;
  model y = treat id id*treat ;
  random id id*treat /test;
run;
quit;
The GLM Procedure
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      770.062500      770.062500       0.86    0.3841
ID                           7           12758     1822.598214       2.04    0.1838
Error: MS(ID*TREAT)          7     6254.687500      893.526786
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID*TREAT                     7     6254.687500      893.526786       1.91    0.0888
Error: MS(Error)            48           22471      468.145833
proc mixed data = m2  ;
  class id block treat;
  model y = treat ;
  random  id  id*treat ;
run;
proc mixed data = m2  ;
  class id block treat;
  model y = treat  ;
  random  intercept treat /subject=id;
run;
Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       7       0.86    0.3841

Page 422 right column, item analysis with matching.

proc glm data = m2  ;
  class id block treat;
  model y = treat block block*treat;
  random block*treat block /test;
run;
quit;
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      770.062500      770.062500       7.19    0.0750
BLOCK                        3           16985     5661.770833      52.86    0.0043
Error                        3      321.312500      107.104167
Error: MS(BLOCK*TREAT)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
BLOCK*TREAT                  3      321.312500      107.104167       0.25    0.8624
Error: MS(Error)            56           24177      431.736607
proc mixed data = m2 method=type3 ;
  class id block treat;
  model y = treat /solution ;
  random block  block*treat;
run;
                          Type 3 Analysis of Variance
                                                     Error
Source       Error Term                                 DF    F Value    Pr > F
TREAT        MS(BLOCK*TREAT)                             3       7.19    0.0750
BLOCK        MS(BLOCK*TREAT)                             3      52.86    0.0043
BLOCK*TREAT  MS(Residual)                               56       0.25    0.8624
Residual     .                                           .        .       .
  Covariance Parameter
       Estimates
Cov Parm        Estimate
BLOCK             347.17
BLOCK*TREAT     -40.5791
Residual          431.74
           Fit Statistics
-2 Res Log Likelihood           562.6
AIC (smaller is better)         568.6
AICC (smaller is better)        569.0
BIC (smaller is better)         566.8
                       Solution for Fixed Effects
                                  Standard
Effect       TREAT    Estimate       Error      DF    t Value    Pr > |t|
Intercept               550.44      9.4941       3      57.98      <.0001
TREAT        1         -6.9375      2.5873       3      -2.68      0.0750
TREAT        2               0           .       .        .         .
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       3       7.19    0.0750
proc mixed data = m2;
  class id block treat ;
  model y = treat /solution ddfm=kenwardroger;
  random  intercept  / subject = block;
  repeated treat /subject=treat(block) type=cs;
run;
    Covariance Parameter Estimates
Cov Parm      Subject         Estimate
Intercept     BLOCK             347.17
CS            TREAT(BLOCK)    -40.5791
Residual                        431.74
           Fit Statistics
-2 Res Log Likelihood           562.6
AIC (smaller is better)         568.6
AICC (smaller is better)        569.0
BIC (smaller is better)         566.8
                       Solution for Fixed Effects
                                  Standard
Effect       TREAT    Estimate       Error      DF    t Value    Pr > |t|
Intercept               550.44      9.4941    3.11      57.98      <.0001
TREAT        1         -6.9375      2.5873       3      -2.68      0.0750
TREAT        2               0           .       .        .         .
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       3       7.19    0.0750

Page 422 right column, item analysis without matching.

proc glm data = m2  ;
  class id block treat;
  model y = treat block*treat;
  random block*treat /test;
run;
quit;
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      770.062500      770.062500       0.27    0.6239
Error                        6           17307     2884.437500
Error: MS(BLOCK*TREAT)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
BLOCK*TREAT                  6           17307     2884.437500       6.68    <.0001
Error: MS(Error)            56           24177      431.736607
proc mixed data = m2  ;
  class id block treat;
  model y = treat  ;
  random block*treat;
run;

proc mixed data = m2  ;
  class id block treat;
  model y = treat  ;
  random treat /subject=block ;
run;
 Covariance Parameter Estimates
Cov Parm     Subject    Estimate
TREAT        BLOCK        306.59
Residual                  431.74
           Fit Statistics
-2 Res Log Likelihood           570.5
AIC (smaller is better)         574.5
AICC (smaller is better)        574.7
BIC (smaller is better)         573.3
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1       6       0.27    0.6239

Page 422 right column, combined analysis without matching.

proc glm data = m2;
   class block id treat;
   model y = treat id treat*id treat*block;
   random  id treat*id treat*block /test;
run;
quit;
Tests of Hypotheses for Mixed Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      770.062500      770.062500       0.21    0.6572
Error                   8.8996           32528     3655.002976
Error: MS(ID*TREAT) + MS(BLOCK*TREAT) - MS(Error)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID                           7           12758     1822.598214       2.04    0.1838
Error: MS(ID*TREAT)          7     6254.687500      893.526786
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID*TREAT                     7     6254.687500      893.526786       7.27    <.0001
BLOCK*TREAT                  6           17307     2884.437500      23.46    <.0001
Error: MS(Error)            42     5164.375000      122.961310
proc mixed data = m2;
   class block id treat;
   model y = treat / ddfm = kenwardroger;
   random  id treat*id treat*block;
run;
Covariance Parameter
       Estimates
Cov Parm        Estimate
ID                116.13
ID*TREAT          192.64
BLOCK*TREAT       345.18
Residual          122.96
           Fit Statistics
-2 Res Log Likelihood           532.9
AIC (smaller is better)         540.9
AICC (smaller is better)        541.6
BIC (smaller is better)         541.2
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1     8.9       0.21    0.6572
proc mixed data = t2;
   class block id treat;
   model y = treat /solution ddfm = kenwardroger;
   random  intercept treat /subject=id ;
   random treat /subject=block ;
run;
Covariance Parameter Estimates
Cov Parm      Subject    Estimate
Intercept     ID           116.13
TREAT         ID           192.64
TREAT         BLOCK        345.18
Residual                   122.96
           Fit Statistics
-2 Res Log Likelihood           532.9
AIC (smaller is better)         540.9
AICC (smaller is better)        541.6
BIC (smaller is better)         532.9
                       Solution for Fixed Effects
                                  Standard
Effect       TREAT    Estimate       Error      DF    t Value    Pr > |t|
Intercept               550.44     11.3462    11.1      48.51      <.0001
TREAT        1         -6.9375     15.1142     8.9      -0.46      0.6572
TREAT        2               0           .       .        .         .
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1     8.9       0.21    0.6572
proc mixed data = m2;
   class block id treat;
   model y = treat /solution ddfm = kenwardroger;
   random  intercept treat /subject=id;
   random treat*block;
run;
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1     8.9       0.21    0.6572

Page 422 right column, combined analysis with matching.

proc glm data = m2;
   class block id treat;
   model y = treat id treat*id treat*block block|id;
   random treat|id  treat|block block|id /test;
run;
quit;
Tests of Hypotheses for Random Model Analysis of Variance
Dependent Variable: Y
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
TREAT                        1      770.062500      770.062500       0.86    0.3862
Error                   6.8204     6128.403230      898.538690
Error: MS(ID*TREAT) + MS(BLOCK*TREAT) - MS(Error)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID                           7           12758     1822.598214       1.95    0.1907
Error                   7.5709     7080.797825      935.264881
Error: MS(ID*TREAT) + MS(BLOCK*ID) - MS(Error)
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
ID*TREAT                     7     6254.687500      893.526786       8.75    <.0001
BLOCK*TREAT                  3      321.312500      107.104167       1.05    0.3916
BLOCK*ID                    21     3020.437500      143.830357       1.41    0.2194
Error: MS(Error)            21     2143.937500      102.092262
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
BLOCK                        3           16985     5661.770833      38.04    0.0017
Error                   4.1759      621.551813      148.842262
Error: MS(BLOCK*TREAT) + MS(BLOCK*ID) - MS(Error)
proc mixed data = m2;
  class id block treat ;
  model y = treat /solution ddfm= kenwardroger;
  random intercept treat block /subject = id ;  
  random  intercept  / subject = block;
  repeated treat /subject=treat(block) type=cs;
run;
The Mixed Procedure
    Covariance Parameter Estimates
Cov Parm      Subject         Estimate
Intercept     ID                110.92
TREAT         ID                197.86
BLOCK         ID               20.8690
Intercept     BLOCK             344.56
CS            TREAT(BLOCK)      0.6265
Residual                        102.09
           Fit Statistics
-2 Res Log Likelihood           524.4
AIC (smaller is better)         536.4
AICC (smaller is better)        538.0
BIC (smaller is better)         524.4
                       Solution for Fixed Effects
                                  Standard
Effect       TREAT    Estimate       Error      DF    t Value    Pr > |t|
Intercept               550.44     11.3462    6.02      48.51      <.0001
TREAT        1         -6.9375      7.4939    6.82      -0.93      0.3862
TREAT        2               0           .       .        .         .
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
TREAT           1    6.82       0.86    0.3862

Table 8 on page 425.

data m3;
  input  id t g l y @@;
cards;
 1    1    1    1    511  2    1    1    1     522
 3    1    1    1    588  4    1    1    1     554
 1    2    1    2    514  2    2    1    2     550
 3    2    1    2    595  4    2    1    2     567
 1    3    1    3    496  2    3    1    3     514
 3    3    1    3    563  4    3    1    3     544
 5    1    2    3    571  6    1    2    3     476
 7    1    2    3    489  8    1    2    3     553
 5    2    2    1    591  6    2    2    1     483
 7    2    2    1    514  8    2    2    1     560
 5    3    2    2    596  6    3    2    2     492
 7    3    2    2    519  8    3    2    2     565
 9    1    3    2    536 10    1    3    2     600
11    1    3    2    511 12    1    3    2     490
 9    2    3    3    517 10    2    3    3     569
11    2    3    3    498 12    2    3    3     476
 9    3    3    1    534 10    3    3    1     592
11    3    3    1    509 12    3    3    1     483
;
run;

Table 10 on page 426 and analyses on page 425, first model without pooling.

proc glm data = m3;
  class g t l id;
  model y = g id(g) t l t*l;
  test h = t e = t*l;
run;
quit;
The GLM Procedure
Dependent Variable: y
                                        Sum of
Source                      DF         Squares     Mean Square    F Value    Pr > F
Model                       17     52130.16667      3066.48039     104.37    <.0001
Error                       18       528.83333        29.37963
Corrected Total             35     52659.00000
R-Square     Coeff Var      Root MSE        y Mean
0.989957      1.014088      5.420298      534.5000
Source                      DF       Type I SS     Mean Square    F Value    Pr > F
g                            2      1720.16667       860.08333      29.27    <.0001
id(g)                        9     47206.16667      5245.12963     178.53    <.0001
t                            2        51.50000        25.75000       0.88    0.4333
l                            2      3106.16667      1553.08333      52.86    <.0001
t*l                          2        46.16667        23.08333       0.79    0.4708
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
g                            0         0.00000          .             .       .
id(g)                        9     47206.16667      5245.12963     178.53    <.0001
t                            2        51.50000        25.75000       0.88    0.4333
l                            2      3106.16667      1553.08333      52.86    <.0001
t*l                          2        46.16667        23.08333       0.79    0.4708
        Tests of Hypotheses Using the Type III MS for t*l as an Error Term
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
t                            2     51.50000000     25.75000000       1.12    0.4727

Table 10 on page 426 and analyses on page 425, first model with pooling.

proc glm data = m3;
  class g t l id;
  model y =  g id(g) t l ;
  test h = g e = id(g);
  run;
quit;
                                        Sum of
Source                      DF         Squares     Mean Square    F Value    Pr > F
Model                       15     52084.00000      3472.26667     120.77    <.0001
Error                       20       575.00000        28.75000
Corrected Total             35     52659.00000
R-Square     Coeff Var      Root MSE        y Mean
0.989081      1.003162      5.361903      534.5000
Source                      DF       Type I SS     Mean Square    F Value    Pr > F
g                            2      1720.16667       860.08333      29.92    <.0001
id(g)                        9     47206.16667      5245.12963     182.44    <.0001
t                            2        51.50000        25.75000       0.90    0.4241
l                            2      3106.16667      1553.08333      54.02    <.0001
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
g                            2      1720.16667       860.08333      29.92    <.0001
id(g)                        9     47206.16667      5245.12963     182.44    <.0001
t                            2        51.50000        25.75000       0.90    0.4241
l                            2      3106.16667      1553.08333      54.02    <.0001
       Tests of Hypotheses Using the Type III MS for id(g) as an Error Term
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
g                            2     1720.166667      860.083333       0.16    0.8512

proc mixed data = m3;
  class g t l id;
  model y = t /solution;
  random  id(g) l ;
run;
proc mixed data = m3;
  class g t l id;
  model y = t /solution ;
  random intercept  /subject=id(g);
  random intercept /subject=l;
run;
The Mixed Procedure
 Covariance Parameter Estimates
Cov Parm      Subject    Estimate
Intercept     id(g)       1473.03
Intercept     l            127.03
Residual                  28.7500
           Fit Statistics
-2 Res Log Likelihood           275.4
AIC (smaller is better)         281.4
AICC (smaller is better)        282.2
BIC (smaller is better)         275.4
                     Solution for Fixed Effects
                              Standard
Effect       t    Estimate       Error      DF    t Value    Pr > |t|
Intercept           533.92     12.9418    12.1      41.26      <.0001
t            1     -0.5000      2.1890      20      -0.23      0.8216
t            2      2.2500      2.1890      20       1.03      0.3163
t            3           0           .       .        .         .
        Type 3 Tests of Fixed Effects
              Num     Den
Effect         DF      DF    F Value    Pr > F
t               2      20       0.90    0.4241

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