---
title: "Session 2 — Resume experiment"
subtitle: "QSS 2.1 | Probability and Statistics (MIR)"
author: "Prof. Dae-Jin Lee (daelee@faculty.ie.edu)"
date: "Fall 2026"
output:
  html_document:
    toc: true
    toc_float: true
    theme: flatly
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE)
library(tidyverse)
library(qss)
```

Randomized race-sounding names on resumes; outcome = callback (0/1).

```{r}
data("resume", package = "qss")

dim(resume)
head(resume)
summary(resume)

resume <- as_tibble(resume)
```

# Callback rates by race

```{r}
resume |>
  count(race, call) |>
  group_by(race) |>
  mutate(share = n / sum(n))

resume |>
  group_by(race) |>
  summarise(
    n = n(),
    callback_rate = mean(call),
    .groups = "drop"
  )

mean(resume$call)
```

# Racial gap

```{r}
rates <- resume |>
  group_by(race) |>
  summarise(callback_rate = mean(call), .groups = "drop")

rates
rates$callback_rate[rates$race == "white"] -
  rates$callback_rate[rates$race == "black"]
```

**Interpret:** About a 3.2 percentage-point higher callback rate for white-sounding names under random assignment of names.

# Potential outcomes (TCWD language)

For each employer contact:

- Treatment: race-sounding name on the resume
- Potential outcomes: \(Y(\text{black})\), \(Y(\text{white})\)
- We observe only the callback for the assigned name

The difference in mean(`call`) by race estimates the average effect of race-sounding names under random assignment.