5 How well the new teaching method works From the data, we are 95% confident that the new teaching method is (1.73, 157) times effective than the old method. > library(faraway) > data(spector) > help (spector) A data frame with 32 observations on the following 4 variables. grade 1 = exam grades improved, 0 = not improved psi 1 = student exposed to PSI (a new teach method), 0 = not exposed tuce a measure of ability when entering the class gpa grade point average > head(spector) grade psi tuce 1 0 0 20 2 0 0 22 3 0 0 24 4 0 0 12 5 1 0 21 6 0 0 17 gpa 2.66 2.89 3.28 2.92 4.00 2.86 > library(survival) > cmod<-clogit(grade~tuce+gpa+strata(psi),spector) > summary(cmod) Call: coxph(formula = Surv(rep(1, 32), grade) ~ tuce + gpa + strata(psi), data = spector, method = "exact") n= 32 coef exp(coef) se(coef) z p tuce 0.0883 1.09 0.136 0.651 0.51 gpa 2.5850 13.26 1.189 2.174 0.03 tuce gpa exp(coef) exp(-coef) lower .95 upper .95 1.09 0.9155 0.837 1.42 13.26 0.0754 1.290 136.37 Rsquare= 0.242 (max possible= 0.601 ) Likelihood ratio test= 8.85 on 2 df, p=0.012 Wald test = 6 on 2 df, p=0.0499 Score (logrank) test = 8.31 on 2 df, p=0.0157 Drop tuce as the p-value is large (0.51) > cmod2<-clogit(grade~gpa+strata(psi),spector) > summary(cmod2) Call: coxph(formula = Surv(rep(1, 32), grade) ~ gpa + strata(psi), data = spector, method = "exact") n= 32 coef exp(coef) se(coef) z p gpa 2.80 16.5 1.15 2.44 0.015 gpa exp(coef) exp(-coef) lower .95 upper .95 16.5 0.0606 1.73 157 Rsquare= 0.231 (max possible= 0.601 ) Likelihood ratio test= 8.4 on 1 df, Wald test = 5.94 on 1 df, Score (logrank) test = 7.9 on 1 df, p=0.00375 p=0.0148 p=0.00493 From the data, we are 95% confident that the new teaching method is (1.73, 157) times effective than the old method. halfnorm(residuals(cmod2))