fake farmer data.R

# Characteristics of the data set to be generated:
# 
# N(mu, sigma) error term,
# 500 treated units, 500 control units,
# Predictors: treatment (binary variable), and 
# hectares (continuous variable--use a uniform distribution), 
# pre-treatment productivity (continuous variable--use a uniform distribution).
# Outcome variable: post-treatment productivity: (continuous variable).

# Avg pre-treatment productivity for tmt group 
# should be double the average pre-treatment 
# productivity for the control group.  Average 
# hectares should actually be a little higher 
# for the tmt group than for the control group.

# Data generating process for Y should be simply 
# the pre-treatment productivity + error term.

set.seed(9876)
treat <- factor(c(rep(1, 500), rep(0, 500)))                    # 500 treat, 500 control
hectares <- c(runif(500, 50, 150), runif(500, 35, 135))         # Average hectares is higher for tmt
mean(hectares[1:500]) - mean(hectares[501:1000])                # 16.3 mean difference
pre_productivity <- 
  c(runif(500, 1000, 2000), runif(500, 500, 1000))
mean(pre_productivity[1:500])/mean(pre_productivity[501:1000])  # 1.98 mean ratio
post_productivity <- pre_productivity + rnorm(1000, 0, 100)     # err = N(0, 100)
df <- data.frame(post_productivity, pre_productivity, hectares, treat)
str(df)

library(ggplot2)
ggplot(data = df, aes(x = `treat`, y = `post_productivity`)) +
  geom_boxplot(alpha = 0.3, fill = c("blue", "red")) +
  ggtitle("Post-Treatment Productivity Vs.Treatment Status") +
  xlab("Treatment Status") +
  ylab("Post-Treatment Productivity")