###############################################################################

# IHGA bootstrap analysis

# unbuffer the output (if relevant)

# change to an appropriate directory

# define the raw data set

ihga.raw.data <- list( 

  C = c( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 
    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 
    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 
    3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 7, 7 ), 

  E = c( 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 
    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 
    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 
    3, 3, 3, 3, 3, 3, 4, 4, 4, 5, 6 ), 

  n.C = 287, n.E = 285

)

# extract the control and experimental data vectors

C <- ihga.raw.data$C

E <- ihga.raw.data$E

# set the number of bootstrap samples and the random number seed

M <- 100000

seed <- 20815034

set.seed( seed )

multiplicative.effect <- rep( NA, M )

system.time( 

  for ( m in 1:M ) {

    C.star <- sample( C, replace = T )

    C.star.mean <- mean( C.star )

    E.star <- sample( E, replace = T )

    E.star.mean <- mean( E.star )

    multiplicative.effect[ m ] <- E.star.mean / C.star.mean

  }

)

#   user  system elapsed 
#   3.29    0.06    3.36

hist( multiplicative.effect, prob = T, breaks = 100,
  xlab = 'multiplicative effect', 
  main = 'bootstrap approximation to posterior distribution' )

posterior.probability.of.improvement <-
  mean( multiplicative.effect < 1 )

c( mean( multiplicative.effect ), sd( multiplicative.effect ),
  posterior.probability.of.improvement )

# [1] 0.8188057 0.0903853 0.9694300

#                 multiplicative     
#                     effect         
                                 posterior probability
#   method          mean    SD         of improvement

# Bayesian REPR    0.8253 0.09191          0.963
# bootstrap        0.8188 0.09039          0.969             

# the bootstrap, which is both a frequentist and a bayesian
# nonparametric method, gets almost exactly the same results
# as the bayesian REPR model fitting, 

# --> without having to do any parametric model specification

# the only assumptions made by the bootstrap, which are
# implied by context, are:

#   exchangeability of C_1, C_2, ...
#   exchangeability of E_1, E_2, ...

###############################################################################