SAS Textbook Examples Applied Linear Statistical Models by Neter, Kutner, et. al. Chapter 24: Random and Mixed Effects Models

Inputting the Apex Enterprises data, table 24.1, p. 964.
data ratings;
input rating officer candidate;
cards;
76  1  1
65  1  2
85  1  3
74  1  4
59  2  1
75  2  2
81  2  3
67  2  4
49  3  1
63  3  2
61  3  3
46  3  4
74  4  1
71  4  2
85  4  3
89  4  4
66  5  1
84  5  2
80  5  3
79  5  4
;
run;
Officer factor means and grand mean, table 24.1, p. 964.
Note: The grand mean is the first observation that has a missing value for the level of officer.
proc means data = ratings mean noprint;
class officer;
var rating;
output out=temp mean=mout;
run;
proc print data=temp;
var officer mout;
run;
Obs    officer     mout

1        .       71.45
2        1       75.00
3        2       70.50
4        3       54.75
5        4       79.75
6        5       77.25
Fig. 24.2, p. 964.
goptions reset=all;
symbol c=blue v=dot h=.8;
axis1 label=(a=90 'Treatment');
proc gplot data=ratings;
plot officer*rating / vaxis=axis1;
run;
quit;
Table 24.2, p. 965.
proc glm data=ratings;
class officer;
model rating= officer;
random officer;
run;
quit;
The GLM Procedure

Class Level Information

Class         Levels    Values
officer            5    1 2 3 4 5

Number of observations    20
The GLM Procedure
Dependent Variable: rating
Sum of
Source                      DF         Squares     Mean Square    F Value    Pr > F
Model                        4     1579.700000      394.925000       5.39    0.0068
Error                       15     1099.250000       73.283333
Corrected Total             19     2678.950000
R-Square     Coeff Var      Root MSE    rating Mean
0.589671      11.98120      8.560569       71.45000
Source                      DF       Type I SS     Mean Square    F Value    Pr > F
officer                      4     1579.700000      394.925000       5.39    0.0068
Source                      DF     Type III SS     Mean Square    F Value    Pr > F
officer                      4     1579.700000      394.925000       5.39    0.0068

The GLM Procedure

Source                  Type III Expected Mean Square
officer                 Var(Error) + 4 Var(officer)
Estimating the overall mean and calculating the 90% confidence limit, p. 967.
Note: There does not appear to be an build in function for this in SAS. So, we will first run the proc glm again and save the Ybar, the MSTR and the df that we need as macro variables. Then we calculate the appropriate t-value and finally we compute the confidence limits.
ods listing close;
ods output  OverallANOVA=anova   FitStatistics=fits;
proc glm data=ratings;
class officer;
model rating= officer;
random officer;
run;
quit;
ods listing;
data _null_;
set anova;
if source = 'Model' then call symput( 'mstr', ms);
if source = 'Corrected Total' then call symput('df', df);
run;
%put here are the macro vars: &mstr and &df;
data _null_;
set fits;
call symput('mean',  DepMean);
run;
%put here is the last value: &mean;
data temp;
t = tinv( .95, 4);
df = &df+1;
s2 = &mstr/(&df +1);
Lower = &mean - t*sqrt(s2);
Upper = &mean + t*sqrt(s2);
run;
proc print data=temp;
run;
Obs       t       df       s2       Lower      Upper

1     2.13185    20    19.7463    61.9768    80.9232
Estimating the 90% confidence limit of simga_mu2/(sigma_mu2 + sigma2), p. 968-969.
ods listing close;
ods output overallanova=anova;
proc glm data=ratings;
class officer;
model rating = officer;
run;
quit;
ods listing;
data _null_;
set anova;
if source='Model' then call symput('mstr', ms);
if source='Error' then call symput('mse', ms);
if source='Model' then call symput('dfmodel', df);
if source='Error' then call symput('dferr', df);
run;
%put here are the macro variables:      394.925, &mse, &dfmodel and &dferr;
data temp;
lower_f = finv( .95, &dfmodel, &dferr);
upper_f = finv( .05, &dfmodel, &dferr);
mstr =      394.925;
mse = &mse;
L = (1/4)*( (     394.925/&mse)*(1/lower_f) - 1 );
U = (1/4)*( (     394.925/&mse)*(1/upper_f) - 1 );
L_star = L/(1+L);
U_star = U/(1+U);
run;
proc print data=temp;
run;
Obs    lower_f    upper_f      mstr       mse         L          U        L_star     U_star

1     3.05557    0.17071    394.925    73.2833    0.19092    7.64195    0.16031    0.88429
Estimating sigma2 and the 90% confidence interval, p. 970.
Note: We are using the same numbers from the ANOVA table of the proc glm so we will be using the macro variables that were created when we were estimating the 90% confidence limit of simga_mu2/(sigma_mu2 + sigma2) on p. 968.
%put here are the macro variables we will be using in this procedure: &mse and &dferr;
data temp;
mse = &mse;
dfden = &dferr;
lower_chi = cinv( .95, &dferr);
upper_chi = cinv( .05, &dferr);
lower2 = (&dferr*&mse )/lower_chi;
upper2 = (&dferr*&mse )/upper_chi;
lower = sqrt(lower2);
upper = sqrt(upper2);
run;
proc print data=temp;
run;
                            lower_     upper_
Obs      mse      dfden      chi        chi       lower2     upper2     lower      upper

1     73.2833      15     24.9958    7.26094    43.9774    151.392    6.63155    12.3042
Point estimate and 90% confidence interval for sigma_mu2, p. 972-973.
Note: We are using the same numbers from the ANOVA table of the proc glm so we will be using the macro variables that were created when we were estimating the 90% confidence limit of simga_mu2/(sigma_mu2 + sigma2) on p. 968.
%put here are the macro variables that we are using in this calculation: &mstr and &mse;
data temp;
s_mu2 = ( &mstr - &mse)/4;
df = round( (4*s_mu2)**2/( (&mstr**2)/(5-1) + (&mse**2)/( 5*(4-1) ) ), 1);
lower_chi = cinv( .95, df);
upper_chi = cinv( .05, df);
lower2 = (df*s_mu2)/lower_chi;
upper2 = (df*s_mu2)/upper_chi;
lower = sqrt(lower2);
upper = sqrt(upper2);
run;
proc print data=temp;
run;
                         lower_     upper_
Obs     s_mu2     df      chi        chi       lower2     upper2     lower      upper

1     80.4104     3    7.81473    0.35185    30.8688    685.615    5.55597    26.1843
Estimating the 90% confidence interval of s_mu2 using the MLS procedure, p. 974-5.
Note: We are using the same numbers from the ANOVA table of the proc glm so we will be using the macro variables that were created when we were estimating the 90% confidence limit of simga_mu2/(sigma_mu2 + sigma2) on p. 968.
Note: When calculating the value of the F distribution SAS won't allow you to use infinity but instead we use 1000000000 as an approximation of infinity.
data temp;
c1 = 1/4;
c2 = -1/4;
ms1 = &mstr;
ms2 = &mse;
df1= &dfmodel;
df2= &dferr;
L = ( &mstr - &mse)/4;
f1 = finv(.95, df1, 1000000000 );
f2 = finv(.95, df2, 1000000000);
f3 = finv(.95, 1000000000, df1);
f4 = finv(.95, 1000000000, df2);
f5 = finv(.95, df1, df2);
f6 = finv(.95, df2, df1);
g1 = 1 - 1/f1;
g2 = 1 - 1/f2;
g3 = ( (f5 - 1)**2 - (g1*f5)**2 - (f4-1)**2 )/f5;
g4 = f6*( ( (f6-1)/f6 )**2 - ( (f3-1)/f6 )**2 - g2**2 );
H_lower = sqrt( (g1*c1*ms1)**2 + ( (f4-1)*c2*ms2 )**2 - g3*c1*c2*ms1*ms2 ) ;
H_upper = sqrt( ( (f3-1)*c1*ms1 )**2 + (g2*c2*ms2)**2 - g4*c1*c2*ms1*ms2 ) ;
lower2 = L - H_lower;
upper2 = L + H_upper;
lower = sqrt(lower2);
upper = sqrt(upper2);
run;
proc print data=temp;
run;
Obs  c1    c2    ms1     ms2   df1 df2    L       f1      f2      f3      f4      f5      f6

1  0.25 -0.25 394.925 73.2833  4   15 80.4104 2.37193 1.66639 5.62807 2.06585 3.05557 5.85781

Obs    g1       g2        g3        g4     H_lower  H_upper   lower2   upper2   lower    upper

1  0.57840  0.39990  -0.011190  -0.56475  60.1848  455.875  20.2256  536.285  4.49729  23.1578
The training example p. 982-983, 987-988 and 989 was omitted because the data is not available.
Inputting Sheffield Foods Co. data, table 24.11, p. 995.
data Sheffield;
input fat method lab rep ;
cards;
5.19  1  1  1
5.09  1  1  2
4.09  1  2  1
3.99  1  2  2
3.75  1  2  3
4.04  1  2  4
4.06  1  2  5
4.62  1  3  1
4.32  1  3  2
4.35  1  3  3
4.59  1  3  4
3.71  1  4  1
3.86  1  4  2
3.79  1  4  3
3.63  1  4  4
3.26  2  1  1
3.48  2  1  2
3.24  2  1  3
3.41  2  1  4
3.35  2  1  5
3.04  2  1  6
3.02  2  2  1
3.32  2  2  2
2.83  2  2  3
2.96  2  2  4
3.23  2  2  5
3.07  2  2  6
3.08  2  3  1
2.95  2  3  2
2.98  2  3  3
2.74  2  3  4
3.07  2  3  5
2.70  2  3  6
2.98  2  4  1
2.89  2  4  2
2.75  2  4  3
3.04  2  4  4
2.88  2  4  5
3.20  2  4  6
;
run;
Fig. 24.3, p. 995.
goptions reset=all;
symbol v=dot h=.8 c=blue;
axis1 order=(0 to 5 by 1) label=(a=90 'Laboratory');
axis2 order=(2.5 to 5.5 by 1);
proc gplot data=sheffield;
by method;
plot lab*fat/ vaxis=axis1 haxis=axis2;
run;
quit;
Fig. 24.4, p. 996.
Note: First we need to calculate the means using proc sql and then we create the graph.
proc sql;
create table plot as
select method, lab, mean(fat) as mfat
from sheffield
where method=1
group by lab;
quit;
proc sql;
create table plot1 as
select method, lab, mean(fat) as mfat1
from sheffield
where method=2
group by lab;
quit;
data combo;
set plot plot1;
run;
symbol1 c=blue v=dot h=.8 i=join;
symbol2 c=red v=square h=.8 i=join;
axis1 label=(a=90 'Mean');
legend1 label=none value=(height=1 font=swiss 'Method 1' 'Method 2' )
position=(bottom right inside) mode=share cborder=black;
proc gplot data=combo;
plot (mfat mfat1)*lab / vaxis=axis1 overlay legend=legend1;
run;
quit; 
Table 24.13-24.14 and fig. 24.5 were not generated, p. 1000-1001.

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.