generating a Diff-inDiff outcome

[Ali Taleb]

Hi Everyone,

I have this data example below where I need to generate a Diff-outcome. The outcome is KW, treat is a dummy indicating that county is treated or not. year is the date of the outcome KW.
So I need a variable that indicates the KW for:

- a treated county before and after treatment
- a control (untreated) county before and after the treatment

But the treatment, as you would see, has been introduced in different periods.

Any help would be appreciated.

Thank you,
Ali

* Example generated by -dataex-. To install: ssc install dataex
clear
input str17 county float(year treat) double KW
                 "sierra"    2012 0 1.7300000190734863
"mono"      2010 0  2.115000009536743
"mono"      2011 0  1047.594475159645
"mono"      2016 1  615.0648956298828
"mono"      2012 0  53.47581745147705
"mono"      2014 0 17.968571462631225
"mono"      2013 0  19.12836738586426
"inyo"      2009 0   332.690810918808
"mariposa"  2016 1  9.930999755859375
"mariposa"  2017 1 30.239999771118164
"mariposa"  2015 0  7.427000045776367
"mariposa"  2012 0 141.42100143432617
"mariposa"  2011 0  596.9810292720795
"mariposa"  2010 0   85.7300033569336
"inyo"      2016 0   538.287145614624
"inyo"      2015 0  7.857142925262451
"inyo"      2010 0                  6
"inyo"      2014 0             110.05
"inyo"      2011 0   172.864492893219
"inyo"      2012 0 130.51142597198486
"plumas"    2017 0 2.2899999618530273
"plumas"    2016 0 20.330999851226807
"plumas"    2014 0 14.689000129699707
"plumas"    2013 0 36.129999947547915
"plumas"    2012 0  5.629000186920166
"plumas"    2011 0   87.7210021018982
"plumas"    2010 0               3.36
"plumas"    2009 0               4.73
"colusa"    2009 0 2166.0669903564453
"colusa"    2010 0  2214.139026412964
"colusa"    2011 0 1735.3049488449096
"colusa"    2012 0  2447.334988412857
"colusa"    2016 1  5633.800985813141
"colusa"    2013 0  491.6379928588867
"colusa"    2015 1 2466.3309755325317
"colusa"    2014 0  4003.920999794006
"colusa"    2017 1 3710.9460051059723
"glenn"     2009 0  93.39200065612792
"glenn"     2010 0  65.02400117397309
"glenn"     2017 1 3352.4740171432495
"glenn"     2016 1 2882.8930139541626
"glenn"     2014 0 1505.9709854125977
"glenn"     2011 0  526.4140105819702
"glenn"     2015 1 2276.3009951114655
"glenn"     2013 0 1979.0520255088807
"glenn"     2012 0  981.4779928994179
"lassen"    2017 0 188.26600074768066
"lassen"    2016 0 1137.7200088500977
"lassen"    2015 0  138.7519989013672
"lassen"    2014 0  36.76900100708008
"lassen"    2011 0  77.89299774169922
"lassen"    2010 0             32.175
"amador"    2016 0  210.7160016298294
"amador"    2015 0 126.29799842834473
"amador"    2014 0 18.808000087738037
"amador"    2017 0 2059.2509994506836
"amador"    2013 0  186.4290008544922
"amador"    2012 0  252.5989990234375
"amador"    2009 0                  7
"amador"    2011 0 155.96100211143494
"amador"    2010 0  39.49600019454956
"calaveras" 2015 0  40.20900011062622
"calaveras" 2016 0 154.52699947357178
"calaveras" 2014 0  65.48999977111816
"calaveras" 2017 1  86.69000101089478
"calaveras" 2013 0   65.9020004272461
"calaveras" 2012 0 38.855000534057616
"calaveras" 2011 0 245.75099277496338
"calaveras" 2010 0 26.644999694824218
"calaveras" 2009 0 1.9600000381469727
"tuolumne"  2016 0 179.29499912261963
"tuolumne"  2017 0 148.83900141716003
"tuolumne"  2015 0  17.88599967956543
"tuolumne"  2014 0  64.86000061035156
"sanbenito" 2010 0 318.01400032043455
"tuolumne"  2013 0               70.5
"sanbenito" 2011 0  2410.385025024414
"tuolumne"  2012 0 198.72000122070312
"sanbenito" 2012 0 25.486000537872314
"tuolumne"  2011 0 11.892999649047852
"tuolumne"  2009 0  7.289999771118164
"tuolumne"  2010 0  7.474000205993653
"sanbenito" 2013 0 352.14100074768066
"sanbenito" 2014 0  59.91600036621094
"sanbenito" 2015 0  58.88300132751465
"sanbenito" 2016 0  771.6840028762817
"sanbenito" 2017 1  637.1600160598755
"tehama"    2009 0 30.239999771118164
"tehama"    2010 0 1408.9979998207093
"tehama"    2011 0  1119.916003227234
"tehama"    2016 0  2279.308996319771
"tehama"    2015 0 1661.4869998693466
"tehama"    2012 0 1752.4280173778534
"tehama"    2013 0 3259.2540171051023
"tehama"    2017 0 1602.0799750089645
"tehama"    2014 0  836.8469922542572
"lake"      2016 0 431.91299533843994
"lake"      2017 0 232.24800395965576
"lake"      2015 0 343.82999897003174
"lake"      2014 0 495.46099758148193
end

[Clyde Schechter]

Because the treatment begins at different times in different counties, you need to use a generalized DID estimator. See https://www.annualreviews.org/doi/pd...-040617-013507 for a full explanation.

In your example data it appears that your variable treat remains 0 at all times in the untreated group, and is 0 before treatment and 1 after treatment in the treated group. If that is correct in your whole data (and the code shown below verifies this before proceeding), then it is already the right setup for your analysis.

encode county, gen(n_county)
xtset n_county year

//    VERIFY THAT TREAT ONLY SWITCHES FROM 0 TO 1, AND DOES SO
by n_county (year): gen byte OK =  treat[_n+1] == treat ///
    | (treat[_n+1] == 1 & treat == 0) | _n == _N
    
xtreg KW i.treat i.year, fe

The coefficient of treat will be the DID estimator of the treatment effect.
You may want to use clustered standard errors with the regression.

[Ali Taleb]

Sorry for the slow reply. Thank you so much!