README.md

In MatthieuStigler/multiDiff: Multi-year diff-diff and FE decompositions

multiDiff

This R package implements the multi-period diff-diff as discussed in Imai and Kim (2019) and Chaisemartin and Hautfeuille (2019).

Installation

remotes::install_github("MatthieuStigler/multiDiff", upgrade = "never")

Main functions:

• DD(): period-by-period diff-diff (DiD)
• DD_manu(): DiD the manual way: showing the four means
• DD_manu_many(): period-by-period manual DiD
• FE_decompo(): FE decomposition following Gibbons et al
• mDid_weights_CH(): Weights from Chaisemartin and Hautefeuille (2020)

Example

Main function is DD() (name might change!), with following arguments

• y_var The name of the y variable
• treat treatment The name of the treatment variable
• time.index The name of the time dimension variable
• unit.index The name of the unit dimension variable
• data The dataset

library(multiDiff)
suppressMessages(library(tidyverse))
data <- sim_dat(N=100)
DD_out <- DD(data=data)

## Main output is:
knitr::kable(DD_out[1:5,])

| time | DiD | treat | control | n_treat | n_control | n_min | miss_data | estimate | std.error | statistic | p.value | conf.low | conf.high | D_var | n_vals | | ---: | --: | :---- | :------ | -------: | ---------: | -----: | :--------- | ----------: | --------: | ----------: | --------: | ----------: | --------: | --------: | ------: | | 2 | 1 | 0_1 | 0_0 | 21 | 50 | 21 | FALSE | 0.6462758 | 0.3832034 | 1.6865085 | 0.0962147 | -0.1181941 | 1.410746 | 0.1269104 | 142 | | 2 | 2 | 0_1 | 1_1 | 21 | 5 | 5 | FALSE | -0.1444697 | 0.7759808 | -0.1861769 | 0.8538707 | -1.7460153 | 1.457076 | 0.2454751 | 52 | | 2 | 3 | 1_0 | 0_0 | 24 | 50 | 24 | FALSE | 1.2720072 | 0.3211001 | 3.9614042 | 0.0001734 | 0.6319059 | 1.912108 | 0.1367899 | 148 | | 2 | 4 | 1_0 | 1_1 | 24 | 5 | 5 | FALSE | 2.0627527 | 0.5086024 | 4.0557271 | 0.0003818 | 1.0191867 | 3.106319 | 0.2468240 | 58 | | 3 | 1 | 0_1 | 0_0 | 19 | 55 | 19 | FALSE | 0.6212886 | 0.3327167 | 1.8673205 | 0.0659275 | -0.0419699 | 1.284547 | 0.1126586 | 148 |

Plot the year-by-year treatment:

DD_out %>% 
  filter(DiD %in% c(1,4)) %>% 
  mutate(case = paste("Treat: ", treat, ", Control: ", control, sep = "")) %>% 
  ggplot(aes(x=time, y = estimate, color = case)) +
  geom_ribbon(aes(ymin = .data$conf.low, 
                    ymax = .data$conf.high,
                  group = case),
              fill = "grey", alpha = I(0.4),
              linetype = 2) +
  geom_line(size = 1) +
  theme(legend.position = "bottom") +
  geom_hline(yintercept =1, linetype = 2) +
  ggtitle("Diff-diff by year and identification case")

Aggregate the results over time:

DiD_aggreg(x=DD_out, by_DiD = FALSE)
#> # A tibble: 1 x 1
#>   estimate
#>      <dbl>
#> 1       NA
DiD_aggreg(x=DD_out, by_DiD = TRUE)
#> # A tibble: 2 x 4
#>     DiD treat control estimate
#>   <int> <chr> <chr>      <dbl>
#> 1     1 0_1   0_0        0.757
#> 2     4 1_0   1_1       NA

MatthieuStigler/multiDiff documentation built on Oct. 21, 2023, 8:13 a.m.