Error: "No pre-treatment periods to test"

[azev77]

I created a simple fake dataset: 5 years: 1993-1997 6 groups: 1-6 30 observations = 5 * 6 2 groups (group # 3 & 4) are treated in 1994 2 groups (group # 5 & 6) are treated in 1996

I thought my data has the same structure as in your example.

When I run

out <- att_gt(yname = "hpg",
              gname = "first_treat",
              idname = "fips11",
              tname = "year",
              xformla = ~1,         #xformla = ~X,
              data = Book1,
              est_method = "reg"
)

I get the message:

But that's wrong bc 1993 is the first year in my data & has no treatment.

How can I fix this? please help

[pedrohcgs]

You need more than 1 pre-treatment period to pre-test.

It is true we cant pre-test parallel trends for the g=1994, but we should be able to do it or the others, I think.

On Fri, Jan 28, 2022 at 18:23 azev77 @.***> wrote:

I created a simple fake dataset: 5 years: 1993-1997 6 groups: 1-6 30 observations = 5 * 6 2 groups (group # 3 & 4) are treated in 1994 2 groups (group # 5 & 6) are treated in 1996

[image: image] https://user-images.githubusercontent.com/7883904/151643491-1db8c7cc-7443-4eda-a2c0-454ca6090165.png

I thought my data has the same structure as in your example.

When I run

out <- att_gt(yname = "hpg", gname = "first_treat", idname = "fips11", tname = "year", xformla = ~1, #xformla = ~X, data = Book1, est_method = "reg" )

I get the message: [image: image] https://user-images.githubusercontent.com/7883904/151643595-344cbfaf-ad70-469a-aa47-c6ca5fc35083.png

But that's wrong bc 1993 is the first year in my data & has no treatment.

How can I fix this? please help

— Reply to this email directly, view it on GitHub https://github.com/bcallaway11/did/issues/108, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABE7343OK22DITUXACT2CGTUYNFRXANCNFSM5NCIV4CA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

--

Pedro H. C. Sant'Anna Department of Economics Vanderbilt University 615-875-8448 (phone) @.*** https://pedrohcgs.github.io

[pedrohcgs]

But it seems that you have just 2 units in the g=1996. Is that the case? With teo observations in a group, it is very hard hard to use any sampling-based procedure to do inference.

On Fri, Jan 28, 2022 at 19:22 Pedro H. C. Sant'Anna @.***> wrote:

You need more than 1 pre-treatment period to pre-test.

It is true we cant pre-test parallel trends for the g=1994, but we should be able to do it or the others, I think.

On Fri, Jan 28, 2022 at 18:23 azev77 @.***> wrote:

I created a simple fake dataset: 5 years: 1993-1997 6 groups: 1-6 30 observations = 5 * 6 2 groups (group # 3 & 4) are treated in 1994 2 groups (group # 5 & 6) are treated in 1996

[image: image] https://user-images.githubusercontent.com/7883904/151643491-1db8c7cc-7443-4eda-a2c0-454ca6090165.png

I thought my data has the same structure as in your example.

When I run

out <- att_gt(yname = "hpg", gname = "first_treat", idname = "fips11", tname = "year", xformla = ~1, #xformla = ~X, data = Book1, est_method = "reg" )

I get the message: [image: image] https://user-images.githubusercontent.com/7883904/151643595-344cbfaf-ad70-469a-aa47-c6ca5fc35083.png

But that's wrong bc 1993 is the first year in my data & has no treatment.

How can I fix this? please help

— Reply to this email directly, view it on GitHub https://github.com/bcallaway11/did/issues/108, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABE7343OK22DITUXACT2CGTUYNFRXANCNFSM5NCIV4CA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

--

Pedro H. C. Sant'Anna Department of Economics Vanderbilt University 615-875-8448 (phone) @.*** https://pedrohcgs.github.io

--

Pedro H. C. Sant'Anna Department of Economics Vanderbilt University 615-875-8448 (phone) @.*** https://pedrohcgs.github.io

[azev77]

I'm confused.

1. in the example data(mpdta) the first year is 2003 and the first treatment is 2004 hence there is also only 1 pre-treatment time period...
2. I tried w/ a dataset w/ > 150k observations but got the same message "No pre-treatment periods to test"
3. eventually would it make sense to nest the 2*2 case as a special case?
4. I can't get the other example in the docs to work

The following modification works but it's not in the docs:

dta <- build_sim_dataset(reset.sim())

[arizagusti]

Hi, I would like to ask about a similar problem that I face. I have a large dataset with around 3 million observations and I am trying to run your estimator with the never treated as the control. I am using the following code:

num_acc_bal_2017_nev_treat <- att_gt(yname = "num_acc", tname = "year", idname = "id_estab_num", gname = "year_inspect", control = "nevertreated", allow_unbalanced_panel = FALSE, data = df_2017_never_treated, base_period = "universal", panel = TRUE )

and am getting the following error "No pre-treatment periods to test"

I don't believe that I have no pre-treatment periods. I cross tabulated the number of observations by year of treatment and year (shown in the following table) and as you can see, I do have pre-treatment periods to test and there are not a few of them. I also tried running the command csdid on stata, and it does work. Would you be able to help me out with this? Thanks in advance for the help!