Issue with cluster parameter in ddd() function - no effect and error with custom data

[xiezheyu]

Dear Ortiz-Villavicencio and Sant’Anna

Hi, first, thank you for developing and maintaining the triplediff package. It is an incredibly valuable tool for the research community. I encountered two issues when using the cluster parameter in the ddd() function:

Issue 1: Cluster parameter has no effect
When comparing results with and without the cluster parameter, I get identical outputs:

Without cluster

att_gt <- ddd(yname = "y", tname = "time", idname = "id", gname = "state",
pname = "partition", xformla = ~cov1 + cov2 + cov3 + cov4,
data = data, control_group = "nevertreated",
base_period = "universal", est_method = "dr")

With cluster

att_gt <- ddd(yname = "y", tname = "time", idname = "id", gname = "state",
pname = "partition", xformla = ~cov1 + cov2 + cov3 + cov4,
data = data, control_group = "nevertreated",
base_period = "universal", est_method = "dr", cluster = "cluster")
Both return the same results, suggesting the cluster parameter might not be working as expected.

Issue 2: Error with custom data
When using my dataset with the cluster parameter, I frequently encounter this error:

Error in as.vector(x, "list"):
cannot coerce type 'closure' to vector of type 'list'
This error doesn't occur when I remove the cluster parameter from the function call.

Could you please help investigate these issues? The cluster parameter seems important for obtaining robust standard errors, but it's currently not functioning properly.

Thank you for your time and this excellent package!

[marcelortizv]

Hi @xiezheyu, thanks for opening this and for the helpful comments.

Issue 1: This looks like a bug. I can reproduce it and will work on a fix ASAP.

Issue 2: Could you share a minimal reproducible example, plus the full error message and steps to reproduce? That will help me diagnose and fix it.

Thanks again!

[xiezheyu]

Hi@marcelortizv, thanks so much for the quick and positive reply!

It's great to hear that you were able to reproduce Issue 1 and are working on a fix. I'm looking forward to it.

Of course, regarding Issue 2, here is the information you requested.

#1. A minimal dataset that causes the error.
`df_minimal_data.zip

#2. Minimal code that causes the error.

library(fect)
library(marginaleffects)

library(fixest)
library(dplyr)
library(bigmemory)
library(ggplot2)
library(triplediff)

sessionInfo()

df_minimal <- readRDS("df_minimal_data.rds")

att_gt <- ddd(yname = "night_light", tname = "year", idname = "pointid", gname = "action",
pname = "Group", xformla = ~ DEM + dis_guanweihui,
data = df_minimal, control_group = "nevertreated",
base_period = "universal", est_method = "dr", cores = 4, cluster = "cluster")
#error as.vector(x, "list"):
cannot coerce type 'closure' to vector of type 'list'

att_gt <- ddd(yname = "night_light", tname = "year", idname = "pointid", gname = "action",
pname = "Group", xformla = ~ DEM + dis_guanweihui,
data = df_minimal, control_group = "nevertreated",
base_period = "universal", est_method = "dr", cores = 4, cluster = "country_code")
#error as.vector(x, "list"):
cannot coerce type 'closure' to vector of type 'list'

summary(att_gt)

es_e <- agg_ddd(att_gt, type = "eventstudy")

summary(es_e)

`
Thanks again for your help!

[xiezheyu]

Thanks so much!

[marcelortizv]

Thanks @xiezheyu for rising this issue. I just merged a branch with the fixes. I've tested it with the data sample you provided, and it seems to work from my end. Please, let me know if you find any other problems and thanks again for using the package.