Key to assignment 11 

1.(a) dose <- 1:6
      death <- c(28, ......)
      SF ~ cbind(death, alive=250-death)
      glm1 <- glm(SF ~ dose, family=binomial)
      summary(glm1)

      Fitted model: log(p/(1-p)) = -2.64 + 0.67*dose = a + b*dose
      
      p is probability of death. If dose increases, log of `odds' of death, 
      hence odds of death, increases (since b is positive).

      We may write, p = e^(-2.64 + 0.67*dose) / (1 + e^(-2.64 + 0.67*dose))
      
  (b) predict(glm1, data.frame(dose=3.5), type="response", se.fit=T)   

      This will give fitted probability (p) for dose=3.5 and standard error.
      
      CI:   0.4293 - 1.96*0.0147,  0.4293 + 1.96*0.0147  

  (c) library(MASS)
      dose.p(glm1, p=0.5)
     
      median lethal dose, LD50 = 3.92