File to ping Github's API, get the commit numbers for applicants to our predoc program, and run a simple DiD on their daily commits before and after the program.

Predocs_tracking_and_DiD.R

# A simple script to grab github commits for a list of users and plot
# Group average commits per day for two groups
# To use this script, you'll need to set up a yaml file with credentials
# for github, and set up a google sheet with columns `Github handle` and 
#`Attended` (with values "Yes" or "No")

# Author: Jim Savage, Schmidt Futures

# Load libraries
library(tidyverse); library(httr); library(yaml);library(jsonlite);
library(googlesheets); library(RCurl); library(lubridate); library(ggthemes)
library(lfe)

# Load credentials
creds <- read_yaml("creds.yaml")
gs_auth()

# Load the google sheet
the_sheet <- gs_url(x = creds$googlesheets$usernames) %>% 
  gs_read(ws = 1)

# Filter for program applicants that have a Github profile
only_GH_members <- the_sheet %>% 
  filter(!is.na(`Github handle`))

# Function to get commits data for each users
get_commits <- function(user) {
  base_api <- "https://api.github.com/"
  events <- GET(paste0(base_api, "users/",user), authenticate(creds$github$username, creds$github$pw))
  repos <- parse_json(GET(paste0(parse_json(events)$repos_url), authenticate(creds$github$username, creds$github$pw)))
  commits <- lapply(repos, function(y) {
    commit_info <- parse_json(GET(paste0(base_api, "repos/", user, "/", y$name, "/commits"), authenticate(creds$github$username, creds$github$pw)))
    if(length(unlist(commit_info))>2) {
      if(TRUE){
      commit_info <- commit_info[unlist(lapply(commit_info, function(x) x$committer$login == user))]
      lapply(commit_info, function(x) {
        tibble(author = x$commit$committer$name, 
               repo = y$name,
               `Github handle` = user, 
               date = as.POSIXct(x$commit$author$date),
               comment = x$commit$message)
      }) %>% bind_rows()
      } else(tibble())
    } else {
      tibble()
    }
  }) %>% bind_rows
  commits
}

# Get commits for all applicants with github accounts
get_all_commits <- lapply(only_GH_members$`Github handle`, get_commits)

# Bind them all together and 
class_commits <- bind_rows(get_all_commits) %>% 
  left_join(only_GH_members)

daily_commits <- class_commits %>% 
  group_by(Attended) %>%
  group_by(`Github handle`, Month = as.Date(date), Attended) %>% 
  summarise(n = n()) %>%
  right_join(crossing(`Github handle` = unique(.$`Github handle`), Month = seq(from = as.Date("2019-01-01"), to = Sys.Date(), by = "day"))) %>% 
  group_by(`Github handle`) %>%
  mutate(Attended = first(Attended[!is.na(Attended)]),
         Attended = ifelse(is.na(Attended), "No", Attended)) %>%
  mutate(n = ifelse(is.na(n), 0, n)) %>% 
  mutate(period = ifelse(between(Month, as.Date("2019-01-01"), as.Date("2019-09-19")), "Pre",
                         ifelse(Month > as.Date("2019-09-22"), "Post", "During")))

daily_commits %>% 
  group_by(Month, Attended) %>% 
  summarise(`Average commits`= mean(n)) %>% 
  filter(Month > as.Date("2019-01-01")) %>%
  ggplot(aes(x = Month, y = `Average commits`, colour = Attended)) +
  geom_line() +
  theme_hc() +
  labs(y = "Average daily commits",
       x = "Date",
       title = "Github commits per day",
       subtitle = "NYU predoc attendees and non-attendee applicants")

daily_commits %>% 
  filter(period %in% c("Pre", "Post")) %>% 
  group_by(`Github handle`) %>% 
  mutate(demeaned_commits = n - mean(n)) %>% 
  group_by(period, Attended) %>% 
  summarise(m = mean(demeaned_commits),
            se = sd(demeaned_commits)/sqrt(n()),
            n = n()) %>% 
  ungroup %>%
  mutate(period = relevel(as.factor(period), "Pre")) %>%
  ggplot(aes(x = period, y = m, colour = Attended)) + 
  geom_linerange(aes(ymin = m - 1.96*se, ymax = m+1.96*se)) +
  geom_point() + 
  theme_hc()

# Simple regression analysis. Fixed effects linear regression on commit counts 
# with dummies for treated RAs during and after the training. 
linear_fit <- felm(n ~ I(period=="During" & Attended == "Yes")+ 
                     I(period=="Post" & Attended == "Yes") | 
                     `Github handle` | 0 | 0, data = daily_commits)
summary(linear_fit)