"treat" not found

[wilsonhernandezbrena]

Hello Marcelo,

Thanks for your work on the very promising tripleddd package!

I have this issue that I cannot fully understand. I moved from the did package to the triplediff to find heterogeneous effects. The did run correctly. However, I got this error after running the tripleddd:

res <- ddd(yname = "street_index", tname = "week", idname = "id", gname = "treat_week", pname = "partition", xformla = ~ 1, control_group = "notyettreated", weightsname = "Shape__Are", boot = T, nboot = 1000, base_period = "universal", cband = T, cluster = "dis", data = test )

Error en [.data.table(small_groups, , treat):
j (el segundo argumento dentro de [...]) es un símbolo único pero no se encuentra el nombre de la columna 'treat'. Si tenía la intención de seleccionar columnas usando una variable al llamar al ámbito, intente con DT[, ..treat]. El prefijo .. transmite un nivel superior similar a la ruta de un sistema de archivos.

This data is collapsed at the sub_district level and has only 390 observations. When I collapsed it at the district level (N=1590), the error was that subgroup 2 may have insufficient data.

Would the error I first pointed out also be related to the sample size? I have approximately 30 periods for more than 10 treated groups (all yet-to-be-treated) with heterogeneous treatment timing. I attach the data in case it is useful.

test.csv

[marcelortizv]

Hi @wilsonhernandezbrena ,

Thanks for reaching out about triplediff! As you mentioned in your Issue, there was a bug in some of the data checks that the package ran before the estimation process. I’ve fixed that now.

I just wanted to let you know that running your example with the data provided will still be problematic because of the overlap. If I count how many units you have in each (gname, pname) bucket, it seems you have zero units in some groups, which makes the estimation unfeasible. I think this has to do with how you’re constructing your data, so please be aware of the overlap condition. Otherwise, the package will raise an error because there’s not enough data.