Chapter 7, Section 7.6

 

Exercise Therapy Trial

Unstructured Covariance (REML Estimation)

 

 

library(foreign)

ds <- read.dta("c:/exercise.dta")

exlong <- reshape(ds, idvar="id", 

+    varying=c("y0","y2","y4","y6","y8","y10","y12"), v.names="y", 

+    timevar="time", time=0:6, direction="long")

exlong <- subset(exlong, time!=1 & time!=5)

attach(exlong)

day <- time*2

day.f <- factor(day, c(0,4,6,8,12))

group.f <- factor(group, c(1,2))

 

newtime <- time

newtime[time==0] <- 1

newtime[time==2] <- 2

newtime[time==3] <- 3

newtime[time==4] <- 4

newtime[time==6] <- 5 

 

library(nlme)

> model1 <- gls(y ~ group.f*day.fna.action=na.omit

+    corr=corSymm(, form= ~ newtime | id), 

+    weights = varIdent(form = ~ 1 | newtime))

summary(model1)

 

Generalized least squares fit by REML

  Model: y ~ group.f * day.f 

  Data: NULL 

       AIC      BIC    logLik

  647.3399 724.6837 -298.6699

 

Correlation Structure: General

 Formula: ~newtime | id 

 Parameter estimate(s):

 Correlation: 

  1     2     3     4    

2 0.924                  

3 0.885 0.960            

4 0.844 0.949 0.958      

5 0.810 0.902 0.911 0.939

Variance function:

 Structure: Different standard deviations per stratum

 Formula: ~1 | newtime 

 Parameter estimates:

       1        2        3        4        5 

1.000000 1.139327 1.049132 1.202937 1.200952 

 

Coefficients:

                    Value Std.Error   t-value p-value

(Intercept)      79.68750 0.7773470 102.51214  0.0000

group.f2          1.36012 1.0318252   1.31817  0.1893

day.f4            1.12500 0.3416830   3.29253  0.0012

day.f6            1.36017 0.3873739   3.51126  0.0006

day.f8            1.58458 0.5048462   3.13873  0.0020

day.f12           1.62356 0.5537518   2.93193  0.0039

group.f2:day.f4  -0.16916 0.4548986  -0.37186  0.7105

group.f2:day.f6   0.21126 0.5123690   0.41231  0.6807

group.f2:day.f8  -0.13092 0.6714472  -0.19498  0.8457

group.f2:day.f12  0.32052 0.7532369   0.42552  0.6710

 

 Correlation: 

                 (Intr) grp.f2 day.f4 day.f6 day.f8 dy.f12 g.2:.4 g.2:.6 g.2:.8

group.f2         -0.753                                                        

day.f4            0.119 -0.090                                                 

day.f6           -0.144  0.109  0.760                                          

day.f8            0.023 -0.017  0.819  0.820                                   

day.f12          -0.038  0.029  0.665  0.696  0.798                            

group.f2:day.f4  -0.090  0.119 -0.751 -0.571 -0.615 -0.500                     

group.f2:day.f6   0.109 -0.145 -0.575 -0.756 -0.620 -0.526  0.761              

group.f2:day.f8  -0.017  0.023 -0.616 -0.617 -0.752 -0.600  0.815  0.822       

group.f2:day.f12  0.028 -0.037 -0.489 -0.511 -0.586 -0.735  0.647  0.681  0.779

 

Standardized residuals:

        Min          Q1         Med          Q3         Max 

-2.26656151 -0.65852801 -0.01531460  0.59847743  2.53952400 

 

Residual standard error: 3.109388 

Degrees of freedom: 173 total; 163 residual

 

 

 

Autoregressive Covariance (REML Estimation)

 

> model2 <- gls(y ~ group.f*day.fna.action=na.omit

+    corr=corAR1(, form= ~ newtime | id))

summary(model2)

 

Generalized least squares fit by REML

  Model: y ~ group.f * day.f 

  Data: NULL 

       AIC      BIC    logLik

  645.0718 682.1968 -310.5359

 

Correlation Structure: ARMA(1,0)

 Formula: ~newtime | id 

 Parameter estimate(s):

     Phi1 

0.9401763 

 

Coefficients:

                    Value Std.Error  t-value p-value

(Intercept)      79.68750 0.8612227 92.52833  0.0000

group.f2          1.36012 1.1431591  1.18979  0.2359

day.f4            1.12500 0.2978976  3.77647  0.0002

day.f6            1.31197 0.4186010  3.13418  0.0020

day.f8            1.53493 0.5036625  3.04754  0.0027

day.f12           1.61594 0.5745615  2.81247  0.0055

group.f2:day.f4  -0.21440 0.3976147 -0.53921  0.5905

group.f2:day.f6   0.25946 0.5535425  0.46872  0.6399

group.f2:day.f8  -0.04774 0.6707040 -0.07118  0.9433

group.f2:day.f12  0.35890 0.7788361  0.46082  0.6455

 

 Correlation: 

                 (Intr) grp.f2 day.f4 day.f6 day.f8 dy.f12 g.2:.4 g.2:.6 g.2:.8

group.f2         -0.753                                                        

day.f4           -0.173  0.130                                                 

day.f6           -0.239  0.180  0.690                                          

day.f8           -0.289  0.218  0.557  0.792                                   

day.f12          -0.328  0.247  0.475  0.674  0.839                            

group.f2:day.f4   0.130 -0.172 -0.749 -0.517 -0.417 -0.356                     

group.f2:day.f6   0.181 -0.240 -0.522 -0.756 -0.599 -0.510  0.689              

group.f2:day.f8   0.217 -0.288 -0.418 -0.595 -0.751 -0.630  0.552  0.792       

group.f2:day.f12  0.242 -0.321 -0.350 -0.497 -0.619 -0.738  0.462  0.663  0.824

 

Standardized residuals:

        Min          Q1         Med          Q3         Max 

-2.18724319 -0.59439302 -0.01382309  0.58072348  2.62469204 

 

Residual standard error: 3.444891 

Degrees of freedom: 173 total; 163 residual

 

 

 

Exponential Covariance (REML Estimation)

 

> model3 <- gls(y ~ group.f*day.fna.action=na.omit

+    corr=corExp(, form= ~ day | id))

summary(model3)

 

Generalized least squares fit by REML

  Model: y ~ group.f * day.f 

  Data: NULL 

       AIC      BIC    logLik

  642.5459 679.6709 -309.2729

 

Correlation Structure: Exponential spatial correlation

 Formula: ~day | id 

 Parameter estimate(s):

   range 

46.13188 

 

Coefficients:

                    Value Std.Error  t-value p-value

(Intercept)      79.68750 0.8614996 92.49859  0.0000

group.f2          1.36012 1.1435267  1.18941  0.2360

day.f4            1.12500 0.3511182  3.20405  0.0016

day.f6            1.31042 0.4279933  3.06178  0.0026

day.f8            1.57188 0.4890739  3.21399  0.0016

day.f12           1.60783 0.5898073  2.72603  0.0071

group.f2:day.f4  -0.21083 0.4678108 -0.45067  0.6528

group.f2:day.f6   0.26101 0.5666630  0.46061  0.6457

group.f2:day.f8  -0.07356 0.6507037 -0.11305  0.9101

group.f2:day.f12  0.34980 0.8031588  0.43554  0.6637

 

 Correlation: 

                 (Intr) grp.f2 day.f4 day.f6 day.f8 dy.f12 g.2:.4 g.2:.6 g.2:.8

group.f2         -0.753                                                        

day.f4           -0.204  0.154                                                 

day.f6           -0.245  0.185  0.803                                          

day.f8           -0.280  0.211  0.688  0.846                                   

day.f12          -0.335  0.252  0.548  0.673  0.785                            

group.f2:day.f4   0.153 -0.203 -0.751 -0.603 -0.516 -0.411                     

group.f2:day.f6   0.185 -0.246 -0.606 -0.755 -0.639 -0.509  0.802              

group.f2:day.f8   0.211 -0.280 -0.517 -0.636 -0.752 -0.590  0.684  0.846       

group.f2:day.f12  0.246 -0.326 -0.402 -0.495 -0.576 -0.734  0.532  0.658  0.766

 

Standardized residuals:

        Min          Q1         Med          Q3         Max 

-2.18976671 -0.59420196 -0.01381865  0.58098711  2.62281222 

 

Residual standard error: 3.445998 

Degrees of freedom: 173 total; 163 residual

 

 

anova(model1,model2)

 

       Model df      AIC      BIC    logLik   Test  L.Ratio p-value

model1     1 25 647.3399 724.6837 -298.6699                        

model2     2 12 645.0718 682.1968 -310.5359 1 vs 2 23.73189  0.0337

 

anova(model1,model3)

 

       Model df      AIC      BIC    logLik   Test  L.Ratio p-value

model1     1 25 647.3399 724.6837 -298.6699                        

model3     2 12 642.5459 679.6709 -309.2729 1 vs 2 21.20597   0.069

 

anova(model2,model3)

 

       Model df      AIC      BIC    logLik

model2     1 12 645.0718 682.1968 -310.5359

model3     2 12 642.5459 679.6709 -309.2729