Generate treatment that is dependent on previous and future years' variable values

[Andrew Yang]

Hi all,

I'm having difficulty generating a treatment and control group for my panel after searching unsuccessfully for a solution on Statalist. I have firm id, year, regis_group (firm's registration type), sales, export, employment.

1) How do I generate treatment in which before year 2002, regis_group can be either 1, 2 or 3 or combination and year 2002 till 2006 the regis_group needs to be 4? (see dataex id "100013968")
2) How do I generate a control in which before year 2002, regis_group can be either 1, 2 or 3 or combination and in year 2002 till 2006, the regis_group needs to be different from the last value in 2001 but not take the value 4 and follow this pattern thereafter? (see dataex id "100019964")

Appreciate your help.

* Example generated by -dataex-. For more info, type help dataex
clear
input str9 id float(year regis_group) double(sales export) long emp
"100011508" 2003 1           0         0    70
"100011516" 1998 1 10810299000 134505000 60530
"100011516" 1999 1  8537410000 107470000 50465
"100011516" 2000 1  8922932000 231721000 52226
"100011516" 2001 1 13978554000 248648000 44609
"100011516" 2003 1 16871809000   8514000 41993
"100011516" 2005 1  1195377000         0  9770
"100013968" 1998 1   561527000  32659000 10645
"100013968" 1999 1   590480000  45579000 10486
"100013968" 2000 1   744993000  66293000  8949
"100013968" 2001 1   780799000  64777000  8877
"100013968" 2002 4   723199000 172516000  8495
"100013968" 2003 4   825809000 183443000  7880
"100013968" 2004 4  1844572000         .  7183
"100013968" 2005 4  1055243000 240015000  6925
"100013968" 2006 4  1061813000 308052000  6727
"100014151" 2006 4    24193000         0    25
"100016069" 1998 1   261016000         0    41
"100016069" 1999 1   189262000         0    63
"100016069" 2004 1    35064000         .    64
"100016069" 2006 1    76030000  23007000   102
"100019964" 1998 3   329570000   6510000  1471
"100019964" 1999 3   384030000   8990000  1403
"100019964" 2000 3   3.801e+08   9880000  1381
"100019964" 2001 3   509510000   8870000  1214
"100019964" 2002 2   985290000   8270000  1205
"100019964" 2003 2  1481020000  28000000  1260
"100019964" 2004 2  1550048000         .  1285
"100019964" 2005 1  1415840000 154870000  1235
"100019964" 2006 1  1537510000 188980000  1185
"100023808" 2001 3    39780000         0   200
"100023808" 2002 3    27743000         0   123
"100023808" 2003 3    27480000         0   257
"100023808" 2004 3    28991000         .   261
"100023808" 2005 3    23443000         0   254
"100023808" 2006 3     6408000         0   117
"100024130" 1999 3    16131000         0    35
"100024130" 2000 3    25651000         0    36
"100024130" 2001 3    10468000         0    36
"100024130" 2002 3    27733000         0    30
"100024130" 2003 3    61086000         0    30
"100024130" 2004 3   143185000         .    35
"100024130" 2005 3   347424000         0    37
"100024130" 2006 3   818107000         0    39
"100024982" 1999 3    38578000         0    78
"100024982" 2000 3   369757000         0   121
"100025037" 1999 3     7436000         0    15
"100026347" 2004 3     8420000         .    42
"100026435" 2005 3   138082000         0    12
"100026435" 2006 3   146080000         0    16
"100026793" 1998 3   253740000         0  1705
"100026793" 1999 3   795443000  49014000  2438
"100026793" 2000 3  1926873000  43737000  2971
"100026793" 2001 3  2837997000  32810000  2585
"100026793" 2002 3  3548779000  63879000  2478
"100026793" 2003 3  4038365000 115249000  2879
"100026793" 2004 3  5242410000         .  2857
"100026793" 2005 3  5314752000  86281000  3101
"100026793" 2006 3  5677846000 170152000  3223
"100026961" 2004 3   302533000         .   818
"100026961" 2005 3   303240000  79650000   677
"100026961" 2006 3   225360000  76870000   638
"100027067" 1999 1      110000         0    19
"100027067" 2000 1       90000         0     6
"100028078" 2001 3    75144000         0   167
"100028078" 2002 3    82364000         0   108
"100028078" 2003 3    82504000         0    86
"100028190" 2001 3     6017000         0     9
"100028190" 2002 3     7598000         0     9
"100028190" 2004 3     9820000         .    45
"100028190" 2005 3    19928000         0    58
"100028190" 2006 3    28218000         0    55
"100116596" 1998 1      229000         0     4
"100132211" 1998 1      898000         0   165
"100140035" 2001 1     6384000   6384000   197
"100140035" 2002 1     8893000   8079000   415
"100207201" 1998 1      510000         0   130
"100257110" 1999 3    18235000   1422000    68
"100257110" 2000 3     1800000         0    68
"100360652" 2001 2    28350000         0   180
"100360652" 2002 2    33180000         0   286
"100360652" 2003 2    42990000         0   270
"100360652" 2005 3    40400000         0   210
"10055211X" 2000 1     5300000         0    30
"100584066" 1998 4    25000000         0   250
"100584066" 1999 1    12000000         0   150
"100584066" 2000 1     5330000         0   140
"100725471" 2004 3    59283000         .   256
"100725471" 2005 3   190807000         0   256
"100725471" 2006 3   249957000         0   260
"100740735" 2001 1     9600000         0    40
"100740735" 2002 1    10500000         0    40
"1007722"   1999 2     6870000         0   150
"10091298X" 1998 1    50980000         0   391
"10100769"  2003 1        1000         0    45
"1010076X"  1998 1      499000         0    33
"1010076X"  1999 1     6812000         0    58
"1010076X"  2000 1     7286000         0    45
"1010076X"  2001 1     7434000         0    45
"1010076X"  2002 1     4846000         0    45
end

------------------ copy up to and including the previous line ------------------

Listed 100 out of 1716001 observations

[Clyde Schechter]

You did not say what to do if there is no observation for year 2001 to serve as the reference for the comparison for years 2002-2006 for controls. But from your example data, it appears that you do not want to apply criterion 2) if there is no such observation. So:

by id (year), sort:egen part1 = min(cond(year < 2002, inlist(regis_group, 1, 2, 3), .))

by id (year): egen regis2001 = max(cond(year == 2001, regis_group, .))

by id (year): egen part2 = min(cond(inrange(year, 2002, 2006), regis_group == 4, .))

by id (year): egen part3 = ///
    min(cond(inrange(year, 2002, 2006), ///
        !inlist(regis_group, 4, regis2001) & !missing(regis2001), .))

label define treat_control    0    "Control"    1    "Treated"
gen byte treat_control:treat_control = 1 if part1 == 1 & part2 == 1
replace treat_control = 0 if part1 == 1 & part3 == 1

[Andrew Yang]

Hi Clyde. Thanks for your help and code. Indeed I do not want to assign controls if no observation for year 2001.

Code works fine except for by id (year) : egen part3. Appreciate your help here. "///" may be typo?

Code:
by id (year): egen part3 = ///
unknown egen function ///()
r(133);

. min(cond(inrange(year, 2002, 2006), ///
command min is unrecognized
r(199);

. !inlist(regis_group, 4, regis2001) & !missing(regis2001), .))

zsh:1: parse error near `)'

[Nick Cox]

Clyde Schechter is writing code on the assumption that you will run it from the do-file editor. It’s not 3 commands to be typed in interactively one by one, but one command that Clyde typed on 3 physical lines. Each use of /// flags to Stata that the command has not yet finished but /// is not allowed interactively.

[Andrew Yang]

Thank you Clyde Schechter and Nick Cox , I tried it already and it works fine! Really appreciate it!