xtdidregress groupvar: none of the groups is a control

[Jaan Cappelle]

Hello everybody

For my master's dissertation, I am doing a research about a certain type of loan SME's can get and wether or not this loan gives a difference in solvency rate (in this dataset defined as 'EQTotAss') in comparison to SME's who chose other types of financing. For this, I need to do a DID regression in Stata 17 with the xtdidregress command. My data, of which you find an extract below, can be considered as longitudinal panel data. My 'treated' group are all companies who got that particular type of loan in 2016, and their financial data from 2014-2019. The control group are other SME's who were matched to the treated group with the psmatch2 command (which explains the '_weight' in the dataset, made it easier to drop non-matched ID's afterwards too), with 3 nearest neighbors for every treated ID. From those, I also have the yearly financial data from 2014-2019.

* Example generated by -dataex-. For more info, type help dataex
clear
input long ID str1 Naceclass int(Year InstYear) double EQTotAss byte(Treat Post) double _weight byte NACEVALUE
 1 "G" 2014 1947  32.6406014702109 1 0 1  4
 1 "G" 2015 1947  33.8920049743329 1 0 1  4
 1 "G" 2016 1947  28.2327031474589 1 1 1  4
 1 "G" 2017 1947  14.8924649002703 1 1 1  4
 1 "G" 2018 1947 -9.87200666407716 1 1 1  4
 1 "G" 2019 1947 -43.9982995098913 1 1 1  4
 2 "G" 2014 1967  12.6942951466242 1 0 1  4
 2 "G" 2015 1967  19.0794423228928 1 0 1  4
 2 "G" 2016 1967  24.7809215661874 1 1 1  4
 2 "G" 2017 1967  26.9791549680904 1 1 1  4
 2 "G" 2018 1967  24.0291518840078 1 1 1  4
 2 "G" 2019 1967  26.1678007797327 1 1 1  4
 3 "N" 2014 1972  23.3552878594458 1 0 1 11
 3 "N" 2015 1972   15.920110911838 1 0 1 11
 3 "N" 2016 1972  9.40103436228268 1 1 1 11
 3 "N" 2017 1972  11.6781888375917 1 1 1 11
 3 "N" 2018 1972  15.9009691916184 1 1 1 11
 3 "N" 2019 1972  15.1472736871831 1 1 1 11
 4 "G" 2014 1974  20.6236539985521 1 0 1  4
 4 "G" 2015 1974  21.7696006916302 1 0 1  4
 4 "G" 2016 1974  17.4387218397215 1 1 1  4
 4 "G" 2017 1974  13.9735025667395 1 1 1  4
 4 "G" 2018 1974  12.5120257648545 1 1 1  4
 4 "G" 2019 1974  13.9218818223522 1 1 1  4
 5 "F" 2014 1976  45.9746975970946 1 0 1  3
 5 "F" 2015 1976  46.5163785796174 1 0 1  3
 5 "F" 2016 1976  48.6401965669444 1 1 1  3
 5 "F" 2017 1976  50.5212689718457 1 1 1  3
 5 "F" 2018 1976  39.8157249306825 1 1 1  3
 5 "F" 2019 1976  40.3284662251476 1 1 1  3
 6 "G" 2014 1981  55.7029418131835 1 0 1  4
 6 "G" 2015 1981  46.6763269195715 1 0 1  4
 6 "G" 2016 1981  50.8085941886075 1 1 1  4
 6 "G" 2017 1981  51.2017406491572 1 1 1  4
 6 "G" 2018 1981  54.0528553637935 1 1 1  4
 6 "G" 2019 1981  60.0368099213494 1 1 1  4
 7 "N" 2014 1982   12.412772256291 1 0 1 11
 7 "N" 2015 1982  25.8026389306743 1 0 1 11
 7 "N" 2016 1982  33.2674527669838 1 1 1 11
 7 "N" 2017 1982  16.3593316031508 1 1 1 11
 7 "N" 2018 1982 -69.9660602014132 1 1 1 11
 7 "N" 2019 1982 -17.7309842126474 1 1 1 11
 8 "G" 2014 1983   12.551288364862 1 0 1  4
 8 "G" 2015 1983  13.6327614424031 1 0 1  4
 8 "G" 2016 1983  14.4829319000867 1 1 1  4
 8 "G" 2017 1983  13.9116920401375 1 1 1  4
 8 "G" 2018 1983  11.5376267977135 1 1 1  4
 8 "G" 2019 1983  11.2173435098684 1 1 1  4
 9 "G" 2014 1983  41.2645473165449 1 0 1  4
 9 "G" 2015 1983  45.9550992397708 1 0 1  4
 9 "G" 2016 1983  41.8151126492942 1 1 1  4
 9 "G" 2017 1983  45.4197580531984 1 1 1  4
 9 "G" 2018 1983  45.5840512923262 1 1 1  4
 9 "G" 2019 1983   42.309229324435 1 1 1  4
10 "K" 2014 1983  90.3568817704526 1 0 1  8
10 "K" 2015 1983   43.352756611385 1 0 1  8
10 "K" 2016 1983  8.47506437186611 1 1 1  8
10 "K" 2017 1983   38.096260127703 1 1 1  8
10 "K" 2018 1983  21.3626810838561 1 1 1  8
10 "K" 2019 1983  32.6933890272311 1 1 1  8
11 "A" 2014 1984  9.76597818789505 1 0 1  1
11 "A" 2015 1984  10.4242072296313 1 0 1  1
11 "A" 2016 1984  9.21042939552771 1 1 1  1
11 "A" 2017 1984  11.2687789698305 1 1 1  1
11 "A" 2018 1984  14.9450847863529 1 1 1  1
11 "A" 2019 1984  17.5874947986905 1 1 1  1
12 "G" 2014 1984  23.3139955909947 1 0 1  4
12 "G" 2015 1984  18.4794890752551 1 0 1  4
12 "G" 2016 1984  15.9237637257454 1 1 1  4
12 "G" 2017 1984  12.0237062947727 1 1 1  4
12 "G" 2018 1984  10.3764739883332 1 1 1  4
12 "G" 2019 1984  5.52145020594037 1 1 1  4
13 "G" 2014 1984   77.315176580457 1 0 1  4
13 "G" 2015 1984  62.1478303981376 1 0 1  4
13 "G" 2016 1984  11.1298090872164 1 1 1  4
13 "G" 2017 1984  15.2496512780868 1 1 1  4
13 "G" 2018 1984  7.86969724855066 1 1 1  4
13 "G" 2019 1984  17.1587698103935 1 1 1  4
14 "G" 2014 1984  17.0968226814307 1 0 1  4
14 "G" 2015 1984 -15.5914219086402 1 0 1  4
14 "G" 2016 1984 -5.80762836817773 1 1 1  4
14 "G" 2017 1984  14.2140058002349 1 1 1  4
14 "G" 2018 1984  15.9713308982548 1 1 1  4
14 "G" 2019 1984  5.64779131465342 1 1 1  4
15 "F" 2014 1984  12.9673074012411 1 0 1  3
15 "F" 2015 1984  10.2540354591162 1 0 1  3
15 "F" 2016 1984  12.6596863163049 1 1 1  3
15 "F" 2017 1984  13.4509490263636 1 1 1  3
15 "F" 2018 1984  25.3390481064483 1 1 1  3
15 "F" 2019 1984  52.2303217730653 1 1 1  3
16 "G" 2014 1985  14.4541847445055 1 0 1  4
16 "G" 2015 1985  19.6117854618801 1 0 1  4
16 "G" 2016 1985  24.3806311396129 1 1 1  4
16 "G" 2017 1985  26.0357442075035 1 1 1  4
16 "G" 2018 1985  23.5179712829521 1 1 1  4
16 "G" 2019 1985  29.9285870826546 1 1 1  4
17 "G" 2014 1985 -1.19360704224253 1 0 1  4
17 "G" 2015 1985  12.4497488713605 1 0 1  4
17 "G" 2016 1985  17.3333444631797 1 1 1  4
17 "G" 2017 1985  14.4851065968165 1 1 1  4
end

I'll admit beforehand, my experience with Stata was nonexisting and the PSM only worked thanks to this forum. Now, with this remaining set of companies, I need to do an xtdidregress command to see what the 'EQTotAss' difference is between treated and control. Firstly, I converted the dataset to panel data by doing

. xtset ID Year

Panel variable: ID (strongly balanced)
 Time variable: Year, 2014 to 2019
         Delta: 1 unit

So far so good. Then, I wanted to do the actual Difference in Differences this way:

. xtdidregress (EQTotAss) (Treat), group (NACEVALUE) time(Year)
invalid group specification
    None of the groups defined by NACEVALUE is a control.
r(198);

Where EQTotAss is of course my variable of interest, Treat is a binary variable indicating treatment (=1) or control (=0) group, NACEVALUE groups the ID's in different sectors according to European NACE classification; this is the 'Naceclass' where every letter corresponds to a number. Year of course says what year it is, where 2014-2015 is pre-treatment and 2016-2019 is post-treatment. What am I doing wrong here, how can I group the ID's to make this work? I've tried dividing them in three new groups according to _weight. This because some controls have been matched to multiple treated ID's, which gave them different _weight and made it easy to divide into 3 groups. Problem was then Treat of course gets omitted because of collinearity with _weight==1.

I feel I'm overlooking something very obvious, but my days of searching didn't give any response so far. I've tried comparing with the 'hospitals' DiD example Stata includes, but the dataset isn't really comparable to mine. Any response in the right direction would be immensely helpful, thanks in advance!

[Maxence Morlet]

I'll admit I'm not too familiar with the xtdidregress package, so waiting on a better response would you consider doing it "manually"?

xtreg [depvar] [controls] i.year i.post#i.treat [aw=_weight], fe cluster([clustvar])

[Jared Greathouse]

Wait a minute, using your dataset, your real dataset, report the results of

tab Treat

If Stata is telling you nothing is treated, this is likely coding error.


Consider the following real example you can replicate:

u "http://econ.korea.ac.kr/~chirokhan/panelbook/data/basque-clean.dta", clear

loc int_time = 1975


g treated = cond(regionno==17 & year >= `int_time',1,0)

tempvar _XnormVar _xXnormVar

labvars year gdpcap "Year" "ln(GDP per 100,000)"

replace regionname = trim(regexr(regionname,"\(.+\) *",""))

egen id = group(regionname), label(regionname) // makes a unique ID

order id, b(year)

*keep if year >= 1960
drop if inlist(id,18) //12

drop regionno
xtset id year, y

cls

xtdidregress (gdpcap) (treat), group(id) time(year)

and when we tab the treatment, we get

tab treat

    treated |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |        708       96.85       96.85
          1 |         23        3.15      100.00
------------+-----------------------------------
      Total |        731      100.00

I don't do matching here, but this illustrates that for the command to work, you must have the treatment defined correctly.

[Enrique Pinzon (StataCorp)]

Hi Jaan,

Both Maxence and Jared give good suggestions as usual. Please take a look at the FAQ below, it might be of help.

https://www.stata.com/support/faqs/s...specification/

Also, note that the second parenthesis, where you have the variable -Treat-, should indicate the observations in your data that are treated. Put differently, they do not correspond to a dummy variable for belonging to a treatment group. The variable -Treat-, in a traditional 2x2 DID, corresponds to the interaction of belonging to the treatment group and the post-treatment period.