https://github.com/bcallaway11/twfeweights

TWFE weights

https://github.com/bcallaway11/twfeweights

Repository

TWFE weights

README.Rmd

---
output: github_document
title: README
---

# twfeweights

`twfeweights` is an R package to compute implicit weights from using two-way fixed effects (TWFE) regressions in the context of causal inference with panel data. 

The package currently includes functions to compute implicit weights for:

* TWFE regressions with staggered treatment adoption

* TWFE regressions with staggered treatment adoption that include covariates (in progress)

* Weighted averages into overall average treatment effects and simple overall average treatment effects, as discussed in [@callaway-santanna-2021]

## Installation

```{r}
library(twfeweights)
library(did)
library(fixest)
library(ggplot2)
data(mpdta)
mpdta$post <- 1*( (mpdta$year >= mpdta$first.treat) & (mpdta$treat==1) )

# compute ATT(g,t)
attgt <- did::att_gt(yname="lemp",
                idname="countyreal",
                gname="first.treat",
                tname="year",
                data=mpdta,
                control_group="nevertreated",
                base_period="universal")
summary(attgt)

# compute TWFE estimate
fixest::feols(lemp ~ post | countyreal + year, cluster="countyreal", data=mpdta)

# compute TWFE weights
tw <- twfe_weights(attgt)
tw
twfe_alp <- sum(tw$wTWFEgt * tw$attgt)
twfe_alp

sum(tw$wTWFEgt[tw$post==1] * tw$attgt[tw$post==1])


# drop untreated group
tw <- tw[tw$G != 0,]
tw$post <- as.factor(1*(tw$TP >= tw$G))
sum(tw$wTWFEgt[tw$post==1])

ggplot(data=tw,
       mapping=aes(x=wTWFEgt, y=attgt, color=post)) +
  geom_hline(yintercept=0, linewidth=1.2) +
  geom_vline(xintercept=0, linewidth=1.2) + 
  geom_point(size=3) +
  theme_bw() +
  ylim(c(-.15,.05)) + xlim(c(-.25,.7))

wO <- attO_weights(attgt)
wO <- wO[wO$G != 0,]
wO$post <- as.factor(1*(wO$TP >= wO$G))
sum(wO$wOgt * wO$attgt)

ggplot(data=wO,
       mapping=aes(x=wOgt, y=attgt, color=post)) +
  geom_hline(yintercept=0, linewidth=1.2) +
  geom_vline(xintercept=0, linewidth=1.2) + 
  geom_point(shape=18, size=5) +
  theme_bw() +
  ylim(c(-.15,.05)) + xlim(c(-.25,.7))

# plot the difference between the weights in post-treatment periods
plot_df <- cbind.data.frame(tw, wOgt=wO$wOgt)
plot_df <- plot_df[plot_df$post==1,]
sapply(unique(plot_df$G), function(g) mean(subset(plot_df, G==g)$wTWFEgt))
plot_df$g.t <- as.factor(paste0(plot_df$G,",",plot_df$TP))

ggplot(plot_df, aes(x=wTWFEgt, y=attgt, color=g.t)) +
  geom_point(size=3) +
  theme_bw() +
  ylim(c(-.15,.05)) + xlim(c(-.25,.7)) +
  geom_point(aes(x=wOgt), shape=18, size=5) +
  xlab("weight")
```

Owner

GitHub Events