Playing with predicting gender / race from names

salary-explore.r

library(tidyverse)
library(readr)
library(stringr)
library(scales)
library(wru)                            # Predict race
library(gender)                         # Predict gender


## Employees / salary info for Chicago
df_sal <- read_csv("https://data.cityofchicago.org/api/views/xzkq-xp2w/rows.csv?accessType=DOWNLOAD")

## Clean up / recode / format for compatibility with wru / gender packages
df_munged <- df_sal %>%
  mutate(
    firstname = toupper(str_extract(Name, "(?<=,\\s{1,10})\\p{L}+")),
    surname = toupper(str_extract(Name, "^.+(?=,)")),
    county = "031",                     # Mimic voter file
    `Annual Salary` = as.numeric(gsub("\\$", "", `Annual Salary`)),
    pid = 1:n()
  )

gender_pred <- gender(unique(df_munged$firstname), years=c(2017 - 70, 2017 - 18)) %>%
  select(firstname = name, gender) %>%
  mutate(sex = gender == "female")     # Because voterfile is so 1950s

df_munged <- left_join(df_munged, gender_pred, by = "firstname") 

## Get census data and predict race
df_pred_race <- predict_race(df_munged, surname.only = TRUE, sex = TRUE) %>%
  gather(race, prob_race, pred.whi:pred.oth) %>%
  group_by(pid) %>%
  slice(which.max(prob_race)) %>%
  ungroup() %>%
  mutate(race = recode(race,
    pred.asi = "asian",
    pred.bla = "black",
    pred.his = "hispanic",              # an ethnicity, not a race
    pred.oth = "other",
    pred.whi = "white"))

## race vs annual salary
df_pred_race %>%
  filter(`Salary or Hourly` == "Salary" &
           `Full or Part-Time` == "F" &
           race != "other") %>%
  ggplot(aes(x = race, y = `Annual Salary`)) +
  geom_boxplot() +
  scale_y_continuous(labels = dollar)

df_pred_race %>%
  filter(`Salary or Hourly` == "Salary" &
           `Full or Part-Time` == "F" &
           race != "other") %>%
  ggplot(aes(x = `Annual Salary`, fill = race)) +
  geom_density(alpha = 0.4) +
  scale_x_continuous(labels = dollar)

## ditto gender
df_pred_race %>%
  filter(`Salary or Hourly` == "Salary" &
           `Full or Part-Time` == "F" &
           ! is.na(gender)) %>%
  ggplot(aes(x = `Annual Salary`, fill = gender)) +
  geom_density(alpha = .4) +
  scale_x_continuous(labels = dollar)

mod_lin <- lm(
  `Annual Salary` ~ gender + race,
  data = df_pred_race %>%
    filter(`Full or Part-Time` == "F")
)