12 Omitted Variable Bias Simulation

In this classwork, you’ll do another simulation. This time, you’ll:

• Generate a simulated data set with variables ability, education, and earnings
• Let education depend somewhat on a person’s underlying ability
• Let earnings depend both on education and ability.
• Then you’ll estimate the model earnings ~ education to see if we get omitted variable bias when we omit ability.
• Finally, evaluate: does omitted variable bias disappear when we let the sample size grow?

library(tidyverse)

# 1) In the data set below, what is the true value of beta 1
# (the effect of education on earnings)?

tibble(
  ability = runif(n = 20, min = 0, max = 100),
  education = 0.5 * ability + 0.5 * runif(n = 20, min = 0, max = 100),
  earnings = 0.3 * education + 0.5 * ability + rnorm(n = 20, mean = 50, sd = 30)
)

# 2) Pipe the data into lm() and estimate the model 
# `earnings ~ education`.
# What is your estimate for beta 1?
# When you run the code repeatedly, do the estimates 
# seem to be correct on average?

tibble(
  ability = runif(n = 20, min = 0, max = 100),
  education = 0.5 * ability + 0.5 * runif(n = 20, min = 0, max = 100),
  earnings = 0.3 * education + 0.5 * ability + rnorm(n = 20, mean = 50, sd = 30)
) %>%
  ___ %>%
  broom::tidy()

# 3) Use slice and select to get the estimate of beta 1 only.

tibble(
  ability = runif(n = 20, min = 0, max = 100),
  education = 0.5 * ability + 0.5 * runif(n = 20, min = 0, max = 100),
  earnings = 0.3 * education + 0.5 * ability + rnorm(n = 20, mean = 50, sd = 30)
) %>%
  ___ %>%
  broom::tidy() %>%
  ___ %>%
  ___

# 4) Use map() to run the simulation 100 times, storing the estimate
# for beta 1 every time. Then pipe the simulation results into 
# a ggplot to create a histogram of estimates of beta 1. 
# What value are the estimates centered around? Add a red vertical
# line to represent the true value for beta 1.

map(
  .x = ___,
  .f = function(x) {
    tibble(
      ability = runif(n = 20, min = 0, max = 100),
      education = 0.5 * ability + 0.5 * runif(n = 20, min = 0, max = 100),
      earnings = 0.5 * ability + 0.3 * education + rnorm(n = 20, mean = 50, sd = 30)
      ) %>%
      ___ %>%
      broom::tidy() %>%
      ___ %>%
      ___
    }
  ) %>%
  bind_rows() %>%
  ggplot(___) +
  geom_histogram() +
  geom_vline(___)

# 5) Write a function beta_1_estimate that takes a samplesize,
# runs the simulation above 100 times with map() under the
# samplesize given by the function's user, and then returns a 
# 100x2 tibble of beta1 estimates along with the samplesize.

beta_1_estimate <- function(samplesize) {
  map(
  .x = ___,
  .f = function(x) {
    tibble(
      ability = runif(n = samplesize, min = 0, max = 100),
      education = 0.5 * ability + 0.5 * runif(n = samplesize, min = 0, max = 100),
      earnings = 0.5 * ability + 0.3 * education + rnorm(n = samplesize, mean = 50, sd = 30)
      ) %>%
      ___ %>%
      broom::tidy() %>%
      ___ %>%
      ___
    }
  ) %>%
    bind_rows() %>%
    mutate(n = factor(samplesize))
}

# 6) Test that beta_1_estimate works. When you call it on a samplesize
# of 20, what is the minimum and maximum estimates you get?
# Compare: run beta_1_estimate(1000): what are the minimum and maximum
# estimates there?

beta_1_estimate(20) %>%
  ggplot(___) +
  geom_histogram()

beta_1_estimate(1000) %>%
  ggplot(___) +
  geom_histogram()

# 7) Use `map()` to call `beta_1_estimate` on samplesizes
# c(20, 50, 200, 1000), and plot the distribution of estimates
# with geom_density, where samplesize is represented by fill.
# Add a red vertical line to represent the true value of beta 1.

map(
  .x = c(20, 50, 200, 1000),
  .f = ___
) %>%
  bind_rows() %>%
  ggplot(___) +
  geom_density(alpha = .5) +
  geom_vline(___)

# 8) Interpret: does omitted variable bias disappear as the sample
# size increases?
# 
#
#

Download this assignment

Here’s a link to download this assignment.