How to create proper dummy variable for Difference in Difference in do.file

[Chul-Kyoo Jung]

Hi my intervention period is both 2000 and 2001 in the difference-in-differences model. So I want to create a proper dummy variable for Post-intervention period.
In that case, I wonder how to assign no value to the year 2000 and 2001. In other words, I assign the value 0 for all the years before 2000 and 1 for all the years after 2001. Then in stata do file, how could I assign no value for those intervention years of 2000 and 2001.

Initially, I tried the following, but I am not sure if `NA` would be proper way to create the dummy for the DiD estimation.

gen post_intervention = 0
replace post_intervention = 1 if year > 2001
replace post_intervention = NA if year == 2000 & 2001

[Jared Greathouse]

What I usually do is something like

g treat = cond(id==3 & date => 2000, 1,0)

[Chul-Kyoo Jung]

Thanks Jared, do you mean I don't have to separately assign no values for 2000 and 2001? What if I just want to designate post-intervention period starts from 2002?

[Nick Cox]

Stata is just going to throw an error message at the code in #1, unless exceptionally NA is the name of an numeric variable or scalar. You're thinking there of some other software we will call X.

The use of & is also wrong as now explained in detail at https://journals.sagepub.com/doi/pdf...6867X231162009 (published last week).

See https://journals.sagepub.com/doi/pdf...36867X19830921 if you want a fairly comprehensive discussion of indicator variables, including a preference for that term.

The question is still puzzling. A missing value for 2000 and 2001 will just exclude those years from almost any later analysis.

Perhaps you seek separate indicators for being pre- and post- which have in common being 0 for 2000 and 2001.

[Jared Greathouse]

Before we continue, we must clarify a few things. Is more than one unit treated? If yes, are they treated at different times? I don't really see how the intervention can have two start points, unless one period of treatment is a sort of "phrase in" period. I'm not understanding, otherwise.

[Chul-Kyoo Jung]

Thanks Nick Cox for your wonderful reference. Also thanks for Jared Greathouse for the clarification. The following is the detail which I presented in another post. My dataset covers from 1990 to 2010 and the policy was implemented in 2000 and continued until 2001. The treated units were a group of firms. Some firms were treated in 2000 and others were treated 2001. The problem is I cannot identify which firms were which. So I just designated all of them as a single treated group and the treatment period is 2000 and 2001. Could you let me know how to code the treated group and treated period in this case?

[Jared Greathouse]

The treated units were a group of firms. Some firms were treated in 2000 and others were treated 2001. The problem is I cannot identify which firms were which.

Do you mean you don't know in the sense of LITERALLY don't know which firms were treated (like, the list of treated units is unavailable to you/is private), or do you mean you don't know how to identify in code which units were and weren't treated? I can help you, but you need to show us your data using the dataex command.

Once I'm looking at some of your data, then we can have a better, or at least more productive, discussion, since I'll be able to see your data as you see it.

[Chul-Kyoo Jung]

I literally don't know which firms were treated in 2000 and which were in 2001, so I just put both groups together as a single treated group. In sum, treated period is both 2000 and 2001 and treated group is lbg30.

* Example generated by -dataex-. For more info, type help dataex
clear
input float year str6 firmid byte ksic2c float(k_productivity hhi_sales lbg30 top30sharei lshare)
1992 "000020" 21   588529984  .05352982 0          0  .4364
1993 "000020" 21   744310016  .05338813 0          0  .4415
1994 "000020" 21   608419968   .0517649 0          0  .4982
1995 "000020" 21   7.021e+08   .0512402 0          0  .4797
1996 "000020" 21   779320000  .05364646 0          0  .4948
1997 "000020" 21   9.117e+08 .071214095 0          0  .5006
1998 "000020" 21   991529984  .07117989 0          0  .4273
1999 "000020" 21  1964070016  .07088037 0          0  .2614
2000 "000020" 21   502310016  .06271941 0          0  .8708
2001 "000020" 21  1313059968    .055039 0          0   .556
2002 "000020" 21  1786019968  .05649804 0          0  .5746
2003 "000020" 21  1744419968  .05231318 0   .0277531  .6055
2004 "000020" 21  1746249984  .05077323 0  .02407099  .6867
2005 "000020" 21  1473859968  .05311356 0  .02611842  .5684
2006 "000020" 21  1287030016  .05217143 0  .02535325  .6056
2007 "000020" 21    1.07e+09  .05387806 0 .022888433  .6112
2008 "000020" 21 27293939712  .05611051 0  .02041181  .4473
2009 "000020" 21  2445459968  .05811165 0  .01744266  .4235
2010 "000020" 21   217370000   .0560264 0  .01730872  .5767
2011 "000020" 21   210430000  .05059672 0   .0147009  .5075
1992 "000040" 31   723260032  .48241925 0    .875504  .5819
1993 "000040" 31   820640000  .48116615 0   .9667003  .5178
1994 "000040" 31  1063180032  .50073665 0   .9780892  .6232
1996 "000040" 31   657430016  .49643224 0   .9655888   .414
1998 "000040" 31   138540000   .4526372 0   .9864518   .621
2001 "000040" 31   127340000   .2890171 0   .8169259  .8521
2003 "000040" 31   362630016  .27689195 0   .9959713  .4164
2005 "000040" 31   120380000  .27040976 0   .9956971  .2975
2006 "000040" 31    51940000  .27439922 0   .9961551 1.1796
2007 "000040" 31   257520000  .27405494 0   .9964164  .2167
2010 "000040" 31   157460000   .2815831 0   .9973839   .163
2011 "000040" 31    8.60e+07   .3801102 0    .969853  .2409
1992 "000050" 47   199160000  .34448385 0  .10926908  .6793
1993 "000050" 47   195150000   .3559043 0  .11468977  .7178
1994 "000050" 47   301750016   .3546242 0  .10875304  .5764
1995 "000050" 47   442369984   .3592656 0  .10488386  .5844
1996 "000050" 47   327750016   .3725042 0   .0869855  .6432
1997 "000050" 47   422470016   .3836862 0   .0825581  .5627
1998 "000050" 47   640760000   .3867261 0  .08987462  .4604
1999 "000050" 47   929169984   .3996449 0  .08584384  .5143
2000 "000050" 47   200840000  .33890685 0   .8212672  .5777
2001 "000050" 47   195320000  .35462224 0   .9863203  .6374
2002 "000050" 47   280409984   .3785505 0   .9879443   .551
2003 "000050" 47   3.366e+08   .3876458 0   .9831518  .6876
2005 "000050" 47   4.352e+08   .4127032 0   .9907051  .1305
2007 "000050" 47  1138919936   .4149574 0   .9924192  .0447
2009 "000050" 47    94220000   .4066013 0   .9926313  .1152
2010 "000050" 47   188150000   .4055971 0   .9901092  .0569
2011 "000050" 47    87650000  .56482244 0    .913202   .086
1992 "000070" 71   268120000   .0786362 0   .6988313  .4794
1993 "000070" 71   1.973e+08  .07177163 0    .712829  .4826
1994 "000070" 71    95210000  .07273845 0   .7131559  .5245
1995 "000070" 71   125360000  .07342871 0   .7148253  .4993
1996 "000070" 71   1.202e+08 .072934516 0   .7074937  .5574
1997 "000070" 71   124480000   .0781921 0   .7431347  .4979
1998 "000070" 71    90530000  .07776387 0   .7463828  .3591
1999 "000070" 71   1.061e+08  .07633055 1   .7860447  .3336
2000 "000070" 71   1.213e+08  .07649403 0   .7330863  .3717
2001 "000070" 71   223430000 .068164214 0   .6619592  .3247
2002 "000070" 71   179140000 .067267805 0   .6489187  .3011
2003 "000070" 71   160610000 .073163316 0   .6581174  .3953
2004 "000070" 71   163840000   .0745528 0   .5721435  .2468
2005 "000070" 71   181240000 .071113676 0  .54938686  .2641
2006 "000070" 71   133430000  .07161281 0   .5551959  .3372
2007 "000070" 71   108510000  .07922857 0  .50252575  .3959
2009 "000070" 71   228260000    .104891 0   .4726001   .232
2010 "000070" 71   274460000  .09211212 0  .50564134  .1768
1992 "000080" 11   666329984   .5063352 1   .9237077  .5111
1993 "000080" 11   4.961e+08   .4772263 1   .9184917  .4148
1994 "000080" 11   371140000    .484674 1   .9256746  .3457
1995 "000080" 11   379240000   .4526424 1   .9295023  .4394
1996 "000080" 11   3.686e+08   .4169289 1   .8886818  .4654
1997 "000080" 11   342590016   .4172299 1    .893319  .3858
1998 "000080" 11   394180000    .406155 1   .8832415  .2884
1999 "000080" 11   5.059e+08   .3698842 1   .8357728  .2843
2001 "000080" 11   348460000   .4423804 0  .59209883   .454
2002 "000080" 11   211420000   .4422717 0  .58359474  .6478
2009 "000080" 11   342220000   .4075448 0   .5522526  .2534
2010 "000080" 11   345830016   .4097762 0   .5573384  .3394
2011 "000080" 11    3.58e+07    .444746 0   .5610578  .3654
1992 "000100" 21   838990016  .05352982 0          0  .6757
1993 "000100" 21   894609984  .05338813 0          0  .6775
1994 "000100" 21   961889984   .0517649 0          0   .701
1995 "000100" 21  1.0516e+09   .0512402 0          0  .6769
1996 "000100" 21   938689984  .05364646 0          0  .7363
1997 "000100" 21  1217949952 .071214095 0          0  .6192
1998 "000100" 21  1954390016  .07117989 0          0  .3821
1999 "000100" 21  2420780032  .07088037 0          0  .3312
2000 "000100" 21  1609560064  .06271941 0          0  .3875
2001 "000100" 21  1816489984    .055039 0          0  .4049
2002 "000100" 21  1453410048  .05649804 0          0  .3436
2003 "000100" 21  1243570048  .05231318 0   .0277531  .3247
2004 "000100" 21  1069459968  .05077323 0  .02407099  .2913
2005 "000100" 21  1691129984  .05311356 0  .02611842  .2975
2006 "000100" 21   560849984  .05217143 0  .02535325  .3071
2007 "000100" 21   897340032  .05387806 0 .022888433  .3254
2008 "000100" 21  1039760000  .05611051 0  .02041181  .3194
2009 "000100" 21  1197410048  .05811165 0  .01744266  .3519
2010 "000100" 21  1.7069e+09   .0560264 0  .01730872  .3253
2011 "000100" 21  1098310016  .05059672 0   .0147009  .5041
end

[Clyde Schechter]

The fact that it is impossible to discern the current treatment status of any observation in 2000 and 2001 is deeply problematic. I think the best solution is to simply drop all observations from those years. While this is, in a sense, throwing away data, the data being thrown away is more like disinformation than information. If you include it, no matter what you do with your pre-post variable, you will end up with biased estimation of the treatment effect. I think those observations have to go.

[Chul-Kyoo Jung]

Clyde Schechter Thanks so much. I am still trying to figure out the exact treated units. Now I know for sure that suppose I would get those information, I can go with two way fixed effects model as you suggested in another post. Now just one last question for some hypothetical situation: suppose the policy was implemented for the every entities in the treated group but still I could not identify them, then can I just drop all those 2000, 2001 observations and run a simple DiD with pre-intervention "0" for 1990-1999 and post-intervention "1" for 2002-2010? Thanks.

[Clyde Schechter]

See #9 at https://www.statalist.org/forums/for...nce-in-do-file.

[Chul-Kyoo Jung]

Clyde Schechter Thanks I got it!