F test overall joint significance

[Tariq Abdullah]

In the attached image I have the time_to_treat variable where if time_to_treat = -1 after that every value is the lead and before -1 everything is lag. This is a way of doing event study.

Could anyone tell me how can I do the joint significance of some selected values of time_to_treat variable like a F-test for coefficients? Whenever I'm trying to do the joint significance it's providing me an error.

In the following section I'm providing my data so that it would be easier to give me a feedback on this.

. dataex time_to_treat coef se

----------------------- copy starting from the next line -----------------------
[CODE]
* Example generated by -dataex-. For more info, type help dataex
clear
input float(time_to_treat coef se)
-24 .008508598 .005053634
-23 .0022498965 .0045150574
-22 .0013960946 .004846417
-21 .002219392 .004902562
-20 .0020763692 .004785649
-19 .006928295 .004210821
-18 .0027256764 .004483498
-17 -.00321555 .005539487
-16 -.008326484 .005001508
-15 -.01394333 .007769032
-14 -.0017484284 .0043266085
-13 .00001703563 .006014496
-12 .0008115402 .003706033
-11 -.012659636 .00606336
-10 -.01776172 .006103916
-9 -.015083795 .0070582
-8 -.013199518 .007084396
-7 -.011645027 .005474801
-6 -.009051482 .003864478
-5 -.0006222086 .0034474526
-4 .0011750647 .0024080195
-3 .0014252485 .002636254
-2 .002093018 .0017095497
-1 0 0
0 -.0009728096 .0012391168
1 -.0019464534 .0019846538
2 -.000770941 .002230628
3 -.001301857 .002529583
4 .0017778093 .003131443
5 -.00698993 .002825085
6 -.005874823 .0034410385
7 -.009891002 .005276658
8 -.0135705 .005358401
9 -.0199844 .007074628
10 -.02002133 .007729002
11 -.0178371 .006769186
12 -.006262023 .00446658
13 -.0088388445 .0045080045
14 -.00482809 .004252179
15 -.0044788974 .005194928
16 -.00049108505 .005439484
17 -.004579766 .0046234056
18 -.00701394 .0042848717
19 -.006032526 .004182411
20 -.003769428 .004684277
21 -.00958967 .005299903
22 .0003095985 .004741376

[Rich Goldstein]

your question is not clear to me - and you don't provide the code you tried but I think the following is the place to start

help test

you might be particularly interested in the testparm command in that help file

[Tariq Abdullah]

This is the code I tried in the beginning and got the following error message :

test time_to_treat
time_to_treat not found
r(111);

After your kind suggestion also tested out the

testparm time_to_treat
no such variables;
the specified varlist does not identify any testable coefficients
r(111);

I don't know why its saying that there is no testable coefficient. I want to the same what Damian Clarke ( 2020 ) did the same his leads and lags - testing overall joint significance. My leads and lags are given in time_to_treat variable ( as in the value after -1 are leads and before -1 are lags).

[Richard Williams]

Always show your commands and output, including the error message. You could be getting an error because of a typo, or it could be there is some fundamental flaw in your approach.

Incidentally dataex is not being used correctly. Indeed it gets cut off because of the image you provide covers up the last several lines.

[Richard Williams]

What was the command before the test command though? The dataex file is incomplete and we can't see all the commands you gave (or should have given). Try to post an example from start to finish (dataex output followed by other stata commands) so we can replicate exactly what you are doing.

[Tariq Abdullah]

. dataex time_to_treat coef se

----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. For more info, type help dataex
clear
input float(time_to_treat coef se)
-24    .008508598  .005053634
-23   .0022498965 .0045150574
-22   .0013960946  .004846417
-21    .002219392  .004902562
-20   .0020763692  .004785649
-19    .006928295  .004210821
-18   .0027256764  .004483498
-17    -.00321555  .005539487
-16   -.008326484  .005001508
-15    -.01394333  .007769032
-14  -.0017484284 .0043266085
-13  .00001703563  .006014496
-12   .0008115402  .003706033
-11   -.012659636   .00606336
-10    -.01776172  .006103916
 -9   -.015083795    .0070582
 -8   -.013199518  .007084396
 -7   -.011645027  .005474801
 -6   -.009051482  .003864478
 -5  -.0006222086 .0034474526
 -4   .0011750647 .0024080195
 -3   .0014252485  .002636254
 -2    .002093018 .0017095497
 -1             0           0
  0  -.0009728096 .0012391168
  1  -.0019464534 .0019846538
  2   -.000770941  .002230628
  3   -.001301857  .002529583
  4   .0017778093  .003131443
  5    -.00698993  .002825085
  6   -.005874823 .0034410385
  7   -.009891002  .005276658
  8     -.0135705  .005358401
  9     -.0199844  .007074628
 10    -.02002133  .007729002
 11     -.0178371  .006769186
 12   -.006262023   .00446658
 13  -.0088388445 .0045080045
 14    -.00482809  .004252179
 15  -.0044788974  .005194928
 16 -.00049108505  .005439484
 17   -.004579766 .0046234056
 18    -.00701394 .0042848717
 19   -.006032526  .004182411
 20   -.003769428  .004684277
 21    -.00958967  .005299903
 22   .0003095985  .004741376
end

[Tariq Abdullah]

These are the two following commands and the error messages I'm getting :
test time_to_treat
time_to_treat not found
r(111);

end of do-file

r(111);

testparm time_to_treat
no such variables;
the specified varlist does not identify any testable coefficients
r(111);

end of do-file

r(111);


I deeply apologize for any unintentional inconvenience

[Tariq Abdullah]

For doing the whole replication you can use the data here :

use "http://www.damianclarke.net/stata/bacon_example.dta", clear
gen timeToTreat = year - _nfd

Then from this website you can kindly copy paste the command, I can attach the image but if I try to copy paste it ( from the stata section ) then it becomes very messy. ( https://lost-stats.github.io/Model_E...t_study.html#r )

[Jeff Wooldridge]

You have to do some kind of estimation before using the test command.

[Tariq Abdullah]

Dear Dr. Wooldridge,

This is the first time I'm doing event study and I'm totally out of my depth. The above command is the only one that works for my cross section data ( Which I got from Dr. Nick Huntington-Klein 's website https://lost-stats.github.io/Model_E...t_study.html#r ) .

If you could kindly give me a lead what kind of estimation I can do with my data to go further ahead then that would be very kind.

[Justin Niakamal]

Here's the basic idea

use "http://www.damianclarke.net/stata/bacon_example.dta", clear 
gen timeToTreat = year - _nfd
 
eventdd asmrs pcinc asmrh cases i.year, timevar(timeToTreat) method(fe, cluster(stfips))  lags(10) leads(10) accum

// post estimation for eventdd
estat eventdd

// by hand 
test lag10  lag9 lag8 lag7 lag6 lag5 lag4 lag3 lag2 lag1 lag0

Res:

. estat eventdd
                                       
      Joint significance test for
           leads and lags
----------------------------------------
                LEADS
----------------------------------------
F-stat:                    0.8738
P-value:                   0.5547
Degrees of freedom        (9,48)
----------------------------------------
                 LAGS
----------------------------------------
F-stat:                    1.7641
P-value:                   0.0876
Degrees of freedom        (11,48)
----------------------------------------

. 
. 
. test lag10  lag9 lag8 lag7 lag6 lag5 lag4 lag3 lag2 lag1 lag0

 ( 1)  lag10 = 0
 ( 2)  lag9 = 0
 ( 3)  lag8 = 0
 ( 4)  lag7 = 0
 ( 5)  lag6 = 0
 ( 6)  lag5 = 0
 ( 7)  lag4 = 0
 ( 8)  lag3 = 0
 ( 9)  lag2 = 0
 (10)  lag1 = 0
 (11)  lag0 = 0

       F( 11,    48) =    1.76
            Prob > F =    0.0876

[Tariq Abdullah]

Thanks everyone for putting their time for answering my question. Really appreciate everyone's patience with me !

[Sanjana Ravi]

Hi all - thanks so much for the advice presented here. I'm also using the eventdd package and was wondering if there are other recommended postestimation tests, robustness checks, and/or diagnostics I should run? Any/all thoughts would be much appreciated, thank you!