Better compatibility with other event study estimators

[kylebutts]

From making did2s::event_study I noticed most functions use g = 0 as the never-treated group, but you all default to g = Inf. An easy fix is to allow for both by changing line 874 to:

df[, "g"] <-  df[,g] 
df[df$g == 0, "g"] <- Inf

staggered/R/compute_efficient_estimator_and_se.R

Lines 829 to 881 in c84eb40

	processDF <- function(df, i, g, t, y){
	#This function processes the df inputted to staggered (or staggered_cs/sa)
	# It checks that the columns in the user-inputted values of i,g,t,y are actually in the data
	# It also renames these columns to "i", "g", "t", "y"
	# Let's make sure we have columns with name i, t, y and g
	colnames_df <- colnames(df)
	if(!i %in% colnames_df){
	stop(paste0("There is no column ", i, " in the data. Thus, we are not able to find the unit identifier variable."))
	}
	if(!t %in% colnames_df){
	stop(paste0("There is no column ", t, " in the data. Thus, we are not able to find the time identifier variable."))
	}
	if(!g %in% colnames_df){
	stop(paste0("There is no column ", g, " in the data. Thus, we are not able to find the group identifier variable."))
	}
	if(!y %in% colnames_df){
	stop(paste0("There is no column ", y, " in the data. Thus, we are not able to find the outcome variable."))
	}
	# Sanity checks
	if(i %in% c("g", "t", "y" )){
	stop(paste0("Unit identifier cannot be labeled g, t, or y"))
	}
	if(t %in% c("i","y", "g" )){
	stop(paste0("Time identifier cannot be labeled i, g, or y"))
	}
	if(g %in% c("i", "t" ,"y" )){
	stop(paste0("Group identifier cannot be labeled i, t, or y"))
	}
	# Re-label i, t, g, y
	if(i != "i"){
	df[,"i"] <- df[,i]
	}
	if(t != "t"){
	df[, "t"] <- df[,t]
	}
	if(g != "g"){
	df[, "g"] <- df[,g]
	}
	if(y != "y"){
	df[, "y"] <- df[,y]
	}
	return(df)
	}

If you want to be safe, in case y actually has year 0 for whatever reason

if(!any(0 %in% unique(df[, y])) {
    df[df$g == 0, "g"] <- Inf
}

[jonathandroth]

Thanks for the suggestion, @kylebutts!

I think we should add an option neverTreatedValue with default = Inf. Then you use-case would be neverTreatedValue =0. I personally find the convention of using 0 for never treated very odd and do not like hardcoding it. But I'd be happy with a warning

if( 0 %in% unique(df$g) & ! (0 %in% df$y) * neverTreatedValue = 0){
warning("Some observations have g=0. If you meant to use g=0 to denote never treated, set neverTreatedValue = 0")

I don't have bandwidth right now but will try to do this in a few weeks