error for clogit with vce cluster option

[Marcel Issa]

I want to run fixed effects logit model and tell stata that the observations within each firm is not Independent. When i run:

 tabulate Industry, generate (g)

clogit Collateraldummy  Numberofemployees Corporationdummy Totalassets Grossprofit Profitability Leverage Loansize Maturity GDPGrowth g1 g3 Duration Housebank, vce(cluster Banks)

the error: group() required
r(198); appears. My data is:
input float Collateraldummy byte Numberofemployees float Corporationdummy long(Totalassets Grossprofit) double(Profitability Leverage) long Loansize byte Maturity str14 Industry double Duration byte Housebank str6 Firm float Banks int Time
1 28 1 1500 1600000 .0625 .95 475000 10 "Other Industry" 0 0 "Firm A" 7 2018
0 28 1 1500 1600000 .0625 .95 475000 10 "Other Industry" 0 0 "Firm A" 1 2018
1 15 1 500 800000 .0875 .5 150000 10 "Other Industry" 5.75 1 "Firm B" 8 2018
1 15 1 500 800000 .0875 .5 30000 1 "Other Industry" 5.75 1 "Firm B" 8 2018
1 15 1 500 800000 .0875 .5 20000 1 "Other Industry" 6 1 "Firm B" 8 2018
1 10 0 387 815000 .0343558282208589 .72 80000 1 "Handcraft" 10 1 "Firm C" 8 2016
1 10 0 415 830000 .05060240963855422 .77 80000 1 "Handcraft" 11 1 "Firm C" 8 2017
1 10 0 400 850000 .03529411764705882 .9 120000 1 "Handcraft" 12 1 "Firm C" 8 2018
0 10 0 415 830000 .05060240963855422 .77 60000 6 "Handcraft" 1 0 "Firm C" 7 2017
1 25 1 800 3500000 .03428571428571429 .2 100000 1 "Other Industry" 4.666666666666667 0 "Firm D" 3 2018
1 25 1 800 3500000 .03428571428571429 .2 620000 20 "Other Industry" 0 0 "Firm D" 6 2018
1 25 1 800 3500000 .03428571428571429 .2 230000 3 "Other Industry" 5 0 "Firm D" 5 2018
0 8 0 130 300000 .23333333333333334 .4 50000 10 "Gastronomic" 4.75 1 "Firm E" 1 2018
0 3 0 60 190000 0 0 20000 10 "Gastronomic" 0 1 "Firm E" 1 2012
0 8 0 130 300000 .23333333333333334 .4 15000 3 "Gastronomic" 3 0 "Firm E" 3 2016
1 12 1 450 800000 .08125 .26 50000 10 "Handcraft" 10.083333333333334 0 "Firm F" 8 2018
1 12 1 462 830000 .0819277108433735 .32 125000 5 "Handcraft" 8 0 "Firm F" 8 2016
1 12 1 438 755000 .07549668874172186 .3 100000 5 "Handcraft" 0 0 "Firm F" 4 2017
1 12 1 450 800000 .08125 .26 15000 1 "Handcraft" 10 0 "Firm F" 8 2018
1 12 1 438 755000 .07549668874172186 .3 15000 1 "Handcraft" 9 0 "Firm F" 8 2017
1 12 1 462 830000 .0819277108433735 .32 15000 1 "Handcraft" 8 0 "Firm F" 8 2016
1 12 1 438 755000 .07549668874172186 .3 120000 1 "Handcraft" 10 0 "Firm F" 5 2017
1 12 1 462 830000 .0819277108433735 .32 120000 1 "Handcraft" 9 0 "Firm F" 5 2016
0 12 1 450 800000 .08125 .26 10000 1 "Handcraft" 10.583333333333334 0 "Firm F" 5 2018
1 10 1 320 1000000 .08 .55 70000 6 "Gastronomic" 7 0 "Firm G" 5 2018
1 10 1 320 1000000 .08 .55 100000 5 "Gastronomic" 5.166666666666667 0 "Firm G" 4 2018
1 12 1 277 800000 .09375 .6 150000 4 "Gastronomic" 5.083333333333333 1 "Firm H" 5 2018
1 25 1 720 1800000 .11388888888888889 .45 350000 3 "Gastronomic" 12 1 "Firm I" 5 2016
0 25 1 695 2000000 .105 .45 300000 6 "Gastronomic" 14 1 "Firm I" 5 2018
1 3 1 248 500000 .11 .44 30000 4 "Handcraft" 0 0 "Firm J" 3 2015
1 3 1 250 600000 .08333333333333333 .5 50000 5 "Handcraft" 1.33 0 "Firm J" 4 2016
0 3 1 248 500000 .11 .44 8000 1 "Handcraft" 0 0 "Firm J" 7 2015
0 3 1 250 600000 .08333333333333333 .5 8000 1 "Handcraft" 1 0 "Firm J" 7 2016
0 3 1 250 600000 .08333333333333333 .5 10000 3 "Handcraft" 1.083 0 "Firm J" 9 2016
1 25 1 462 1750000 .022857142857142857 .45 100000 1 "Handcraft" 0 0 "Firm K" 9 2016
1 29 1 450 1900000 .027105263157894736 .5 200000 3 "Handcraft" .5833333333333334 0 "Firm K" 9 2017
1 29 1 450 1900000 .027105263157894736 .5 100000 1 "Handcraft" 1 0 "Firm K" 9 2017
1 25 1 462 1750000 .022857142857142857 .45 250000 5 "Handcraft" 0 0 "Firm K" 4 2016
1 29 1 440 2000000 .025 .5 200000 5 "Handcraft" 1.4166666666666667 0 "Firm K" 4 2018
1 9 1 360 415000 .18795180722891566 .25 15000 1 "Handcraft" 5 1 "Firm L" 7 2015
1 9 1 350 435000 .18620689655172415 .25 25000 1 "Handcraft" 6 1 "Firm L" 7 2016
1 9 1 345 430000 .18604651162790697 .3 15000 1 "Handcraft" 7 1 "Firm L" 7 2017
1 14 0 1000 1450000 .07931034482758621 .6 350000 7 "Gastronomic" 15 1 "Firm M" 7 2013
0 15 0 1050 1500000 .06666666666666667 .7 300000 10 "Gastronomic" 20 1 "Firm M" 7 2018
1 14 0 1000 1450000 .07931034482758621 .6 150000 1 "Gastronomic" 15 1 "Firm M" 7 2013
1 15 0 970 1400000 .06785714285714285 .7 150000 1 "Gastronomic" 16.5 1 "Firm M" 7 2014
1 15 0 960 1475000 .06779661016949153 .7 150000 1 "Gastronomic" 17.75 1 "Firm M" 7 2015
1 3 0 350 400000 .125 .5 20000 1 "Handcraft" 7 1 "Firm N" 6 2018
1 3 0 350 400000 .125 .5 15000 5 "Handcraft" 7 1 "Firm N" 6 2018
0 25 1 500 1100000 .18181818181818182 .8 150000 10 "Handcraft" 15 1 "Firm O" 5 2018
0 25 1 500 1100000 .18181818181818182 .8 400000 15 "Handcraft" 15 1 "Firm O" 5 2018
0 25 1 500 1100000 .18181818181818182 .8 50000 1 "Handcraft" 15 1 "Firm O" 5 2018
0 25 0 620 2000000 .15 .2 150000 10 "Handcraft" 20 1 "Firm P" 5 2018
0 25 0 620 2000000 .15 .2 50000 1 "Handcraft" 20 1 "Firm P" 5 2018
0 12 1 380 1500000 .06666666666666667 .3 25000 5 "Handcraft" 15 1 "Firm Q" 5 2018
1 7 0 400 950000 .1368421052631579 .25 300000 5 "Handcraft" 3 1 "Firm R" 7 2015
0 9 0 425 1000000 .123 .2 250000 7 "Handcraft" 6 1 "Firm R" 7 2018
1 7 0 400 950000 .1368421052631579 .25 50000 1 "Handcraft" 3 1 "Firm R" 7 2015
1 8 0 415 975000 .14358974358974358 .2 80000 1 "Handcraft" 4.333333333333333 1 "Firm R" 7 2016
1 9 0 410 935000 .13368983957219252 .2 80000 1 "Handcraft" 5.333333333333333 1 "Firm R" 7 2017
1 9 0 425 1000000 .123 .2 80000 1 "Handcraft" 6 1 "Firm R" 7 2018
1 6 0 370 427000 .14285714285714285 .42 80000 5 "Handcraft" 23 1 "Firm S" 5 2016
1 6 0 370 427000 .14285714285714285 .42 30000 1 "Handcraft" 8 0 "Firm S" 6 2016
1 6 0 375 430000 .13953488372093023 .45 45000 1 "Handcraft" 8.75 0 "Firm S" 6 2017
0 6 0 370 427000 .14285714285714285 .42 80000 5 "Handcraft" 0 0 "Firm S" 2 2016
0 28 1 3500 2875000 .05495652173913043 .38 500000 10 "Other Industry" 14 1 "Firm T" 5 2012
0 30 1 3625 3000000 .05 .4 400000 7 "Other Industry" 4 0 "Firm T" 3 2017
1 30 1 3625 3000000 .05 .4 60000 2 "Other Industry" 5 0 "Firm T" 4 2017
1 30 1 3625 3000000 .05 .4 50000 2 "Other Industry" .16666666666666666 0 "Firm T" 4 2017
0 15 1 3100 2600000 .06538461538461539 .5 150000 3 "Other Industry" 5 0 "Firm U" 3 2018
0 15 1 3100 2600000 .06538461538461539 .5 130000 4 "Other Industry" 4 0 "Firm U" 4 2018
0 15 1 3100 2600000 .06538461538461539 .5 50000 2 "Other Industry" 4 0 "Firm U" 4 2018
1 35 1 2650 2300000 .09 .21 300000 5 "Other Industry" 22 1 "Firm V" 7 2014
1 35 1 2710 2425000 .09278350515463918 .28 250000 7 "Other Industry" 23 1 "Firm V" 7 2015
0 33 1 2665 2400000 .0875 .25 50000 9 "Other Industry" 25.25 1 "Firm V" 7 2017
0 33 1 2700 2350000 .08297872340425531 .25 80000 10 "Other Industry" 26.333333333333332 1 "Firm V" 7 2018
1 34 1 2710 2425000 .09278350515463918 .28 80000 1 "Other Industry" 23.166666666666668 1 "Firm V" 7 2015
1 26 1 1980 1650000 .0893939393939394 .26 325000 10 "Handcraft" 16 1 "Firm W" 7 2015
0 26 1 2050 1700000 .08941176470588236 .31 150000 8 "Handcraft" 18.333333333333332 1 "Firm W" 7 2017
0 26 1 1930 1750000 .08857142857142856 .33 220000 5 "Handcraft" 19.166666666666668 1 "Firm W" 7 2018
0 26 1 2050 1700000 .08941176470588236 .31 80000 1 "Handcraft" 18.166666666666668 1 "Firm W" 7 2017
0 . 0 . . . . . . "" . . "" . .
end
[/CODE]
------------------ copy up to and including the previous line ------------------
I hope someone can assist.

[Andrew Musau]

Since you have a cluster variable, the group variable can be that variable or some other variable nested within the cluster variable. What identifies groups in your data?

[Marcel Issa]

My group variable is banks. Does that mean I cannot set a different group variable than the variable i set as cluster variable?

[Andrew Musau]

You can do that as long as the variable is nested within that cluster variable. In your case, what is needed is that you explicitly specify the group variable in your syntax.

 clogit Collateraldummy  ..., group(Banks) vce(cluster Banks)

[Marcel Issa]

The Problem is that i have a cross section of Banks and firms so I thought to Group Banks and to use firm clustered Standard Errors. My sample size doesnt allow me to Group firms btw.

[Andrew Musau]

clogit is fixed effects logit, so you need a minimum of two observations per group because you are looking at variation within groups. I cannot advise on the best course of action as I do not know what your research goals are, but it is normal to lose cases if you choose to run a fixed effects model. For the data in #1, you still have 50 cases if you both group and cluster by firm, which is not entirely not useful.

[Marcel Issa]

Ok thx Andrew. I have another Problem related to my Regression of the loanspread. It is based on the same data + :
input double Loanspread float Lnduration
.0545 0
.0125 0
.0252 1.9095426
.09140000000000001 1.9095426
.0907 1.94591
.0872 2.397895
.0867 2.484907
.0855 2.564949
.0502 .6931472
.09419999999999999 1.734601
.0407 0
.0931 1.7917595
.017599999999999998 1.7492
.007099999999999999 0
.09159999999999999 1.3862944
.0326 2.4054425
.0433 2.1972246
.0517 0
.0815 2.397895
.08259999999999999 2.3025851
.07999999999999999 2.1972246
.0884 2.397895
.0897 2.3025851
.0867 2.449567
.039 2.0794415
.0409 1.8191584
.033299999999999996 1.805553
.0279 2.564949
.0245 2.70805
.0455 0
.0491 .8458683
.1005 0
.1002 .6931472
.1003 .7338092
.0984 0
.0943 .4595323
.0962 .6931472
.0526 0
.051199999999999996 .8823892
.0816 1.7917595
.0811 1.94591
.0804 2.0794415
.025800000000000003 2.772589
.0227 3.0445225
.0823 2.772589
.08070000000000001 2.862201
.0805 2.931194
.08779999999999999 2.0794415
.0339 2.0794415
.029 2.772589
.026000000000000002 2.772589
.0852 2.772589
.020999999999999998 3.0445225
.0812 3.0445225
.0257 2.772589
.0312 1.3862944
.0254 1.94591
.08760000000000001 1.3862944
.0882 1.6739764
.0887 1.8458267
.0866 1.94591
.0285 3.178054
.09240000000000001 2.1972246
.0917 2.2772672
.016 0
.0191 2.70805
.0376 1.609438
.0936 1.7917595
.0968 .1541507
.0461 1.7917595
.0465 1.609438
.094 1.609438
.0209 3.135494
.0215 3.178054
.019799999999999998 3.267666
.0175 3.308107
.078 3.184974
.0274 2.833213
.027 2.9618306
.0255 3.004031
.0815 2.953173
end
I ran:

 xtreg Loanspread Age Totalassets Numberofemployees Corporationdummy Grossprofit Profitability Leverage Loansize Maturity g1 g3 GDPGrowth Lnduration Housebank if Loantype=="Credit", fe vce (cluster Banks)

and since Lnduration is significant and negative, I want to estimate the effect of housebank on loanspread, when the Duration equals 10 compared to Duration of 1. Could you please help me with the command?

[Andrew Musau]

I want to estimate the effect of housebank on loanspread, when the Duration equals 10 compared to Duration of 1. Could you please help me with the command?

You can generate a dummy for 3 levels of duration. Below, the dummy equals one if duration=1, equals two if duration=10 and equals zero otherwise. Interact this with the housebank variable. Below, I set the base of the duration dummy to duration=1.

gen duration= exp(Lnduration)
gen dduration= duration==1
replace dduration= 2 if duration==10 & dduration==0
xtreg Loanspread Age ...c.Housebank##ib1.dduration, fe cluster(Banks)

Therefore, the coefficient of the interaction term between housebank and duration=10 is the difference between housebank at duration=1 and duration equals 10 (c.Housebank#2.dduration). To test the difference between housebank at duration=10 and all other levels of duration except duration=1, use the test command

test c.Housebank#2.dduration = c.Housebank#0.dduration

[Marcel Issa]

Hi Andrew thx for the Suggestion. However, this is not exactly what i was Looking for. I dont want to know the effect at specified values for Duration. Thus I constructed one Dummy:

generate DDuration = (Duration>=10)

and used it in interaction with the housedummy:

xtreg Loanspread Age Totalassets Numberofemployees Corporationdummy Grossprofit Profitability Leverage Loansize Maturity g1 g3 GDPGrowth Duration Housebank i.DDuration##ib1.Housebank if Loantype=="Credit", fe

and got: -------------------------------------------------------------------------------------
Loanspread | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------------+----------------------------------------------------------------
Age | .0002211 .0000991 2.23 0.039 .0000119 .0004303
Totalassets | 2.03e-06 1.76e-06 1.15 0.265 -1.68e-06 5.73e-06
Numberofemployees | .0003652 .0001991 1.83 0.084 -.000055 .0007853
Corporationdummy | .002254 .0034194 0.66 0.519 -.0049603 .0094682
Grossprofit | -8.90e-07 2.85e-06 -0.31 0.759 -6.91e-06 5.13e-06
Profitability | .0519911 .0248535 2.09 0.052 -.0004452 .1044274
Leverage | .0006589 .0049496 0.13 0.896 -.0097838 .0111017
Loansize | -.0000237 9.33e-06 -2.54 0.021 -.0000434 -3.99e-06
Maturity | -.0001406 .000308 -0.46 0.654 -.0007905 .0005093
g1 | .0013849 .0022332 0.62 0.543 -.0033267 .0060965
g3 | -.0039894 .0036296 -1.10 0.287 -.0116472 .0036683
GDPGrowth | -.0106406 .226572 -0.05 0.963 -.4886657 .4673845
Duration | -.0016098 .0004347 -3.70 0.002 -.0025269 -.0006928
Housebank | -.0103119 .0038512 -2.68 0.016 -.0184373 -.0021865
1.DDuration | .0047377 .0055844 0.85 0.408 -.0070445 .0165198
0.Housebank | 0 (omitted)
|
DDuration#Housebank |
1 0 | -.0110233 .0075315 -1.46 0.162 -.0269134 .0048669
|
_cons | .0388755 .0067726 5.74 0.000 .0245867 .0531644
--------------------+----------------------------------------------------------------
sigma_u | .02095341
sigma_e | .00401662
rho | .96455632 (fraction of variance due to u_i)
-------------------------------------------------------------------------------------
F test that all u_i=0: F(7, 17) = 10.05 Prob > F = 0.0001
Since, I have two Dummies interacting with each other, shouldnt Stata refer in the ouput to the base form where the interactiondummy equals 1 and therefore give 3 coefficients of the Dummies being 0:0 0:1 and 1:0? Why do I get only one coefficient in the Output and how do I interprete this?

[Andrew Musau]

Since, I have two Dummies interacting with each other, shouldnt Stata refer in the ouput to the base form where the interactiondummy equals 1 and therefore give 3 coefficients of the Dummies being 0:0 0:1 and 1:0? Why do I get only one coefficient in the Output and how do I interprete this?

You already include 1.DDuration and Housebank in the regression, so two of the dummies are gone. See the example below:

webuse lbw, clear
reg low age i.ht##i.ui
note: 1.ht#1.ui identifies no observations in the sample

      Source |       SS           df       MS      Number of obs   =       189
-------------+----------------------------------   F(3, 185)       =      4.54
       Model |  2.78222218         3  .927407393   Prob > F        =    0.0043
    Residual |  37.7997884       185  .204323181   R-squared       =    0.0686
-------------+----------------------------------   Adj R-squared   =    0.0535
       Total |  40.5820106       188  .215861758   Root MSE        =    .45202

------------------------------------------------------------------------------
         low |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |  -.0090077   .0062412    -1.44   0.151    -.0213208    .0033054
        1.ht |   .3235592   .1356798     2.38   0.018     .0558805    .5912378
        1.ui |   .2345424   .0933848     2.51   0.013     .0503064    .4187784
             |
       ht#ui |
        1 1  |          0  (empty)
             |
       _cons |   .4662003   .1509261     3.09   0.002     .1684427     .763958
------------------------------------------------------------------------------

. reg low age 1.ht#1.ui 1.ht#0.ui 0.ht#1.ui 0.ht#0.ui
note: 1.ht#1.ui identifies no observations in the sample

      Source |       SS           df       MS      Number of obs   =       189
-------------+----------------------------------   F(3, 185)       =      4.54
       Model |  2.78222218         3  .927407393   Prob > F        =    0.0043
    Residual |  37.7997884       185  .204323181   R-squared       =    0.0686
-------------+----------------------------------   Adj R-squared   =    0.0535
       Total |  40.5820106       188  .215861758   Root MSE        =    .45202

------------------------------------------------------------------------------
         low |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |  -.0090077   .0062412    -1.44   0.151    -.0213208    .0033054
             |
       ht#ui |
        0 1  |   .2345424   .0933848     2.51   0.013     .0503064    .4187784
        1 0  |   .3235592   .1356798     2.38   0.018     .0558805    .5912378
        1 1  |          0  (empty)
             |
       _cons |   .4662003   .1509261     3.09   0.002     .1684427     .763958
------------------------------------------------------------------------------

The coefficient of the 1/0 combination is equal to that of 1.ht and 0/1 is equal to 1.ui. So this answers your question on where did the coefficients go to. Hope that you have figured out what you wanted in case my advice in #8 was not it.

[Marcel Issa]

So in my case, when I interact a dummy that quals 1 if Duration exceeds 10 years with another dummy that equals 1 if it is the housebank then the ouput shows:

generate DDuration = (Duration>=10)

xtreg Loanspread_pct Age Totalassets Numberofemployees Corporationdummy Grossprofit Profitability_pct Leverage_pct Loansize Maturity g1 g3 GDPGrowth_pct Duration Housebank i.Housebank#i.DDuration if Loantype=="Credit", fe

note: 1.DDuration#1.Housebank omitted because of collinearity

Fixed-effects (within) regression Number of obs = 41
Group variable: Banks Number of groups = 8

R-sq: Obs per group:
within = 0.8755 min = 1
between = 0.0282 avg = 5.1
overall = 0.3149 max = 13

F(16,17) = 7.47
corr(u_i, Xb) = -0.4907 Prob > F = 0.0001


Loanspread_pct Coef. Std. Err. t P>t [95% Conf. Interval]

Age .0221114 .0099144 2.23 0.039 .0011939 .0430289
Totalassets .0002025 .0001756 1.15 0.265 -.0001679 .0005729
Numberofemployees .0365158 .0199131 1.83 0.084 -.0054972 .0785288
Corporationdummy .225397 .3419371 0.66 0.519 -.4960272 .9468213
Grossprofit -.000089 .0002855 -0.31 0.759 -.0006913 .0005133
Profitability_pct .0519911 .0248535 2.09 0.052 -.0004452 .1044274
Leverage_pct .0006589 .0049496 0.13 0.896 -.0097838 .0111017
Loansize -.0023671 .0009328 -2.54 0.021 -.0043351 -.0003992
Maturity -.014058 .0308022 -0.46 0.654 -.079045 .050929
g1 .1384901 .2233168 0.62 0.543 -.3326672 .6096474
g3 -.3989448 .3629563 -1.10 0.287 -1.164716 .366826
GDPGrowth_pct -.0106406 .226572 -0.05 0.963 -.4886657 .4673844
Duration -.160985 .0434656 -3.70 0.002 -.2526893 -.0692807
Housebank -.557419 .5235734 -1.06 0.302 -1.662062 .5472242

DDuration#Housebank
0 1 -.473769 .5584445 -0.85 0.408 -1.651984 .7044458
1 0 -.6285598 .5743228 -1.09 0.289 -1.840275 .5831554
1 1 0 (omitted)

_cons 3.887554 .6772577 5.74 0.000 2.458665 5.316443

sigma_u 2.0953408
sigma_e .40166154
rho .96455633 (fraction of variance due to u_i)

F test that all u_i=0: F(7, 17) = 10.05 Prob > F = 0.0001


So the note"

note: 1.DDuration#1.Housebank omitted because of collinearity" is normal and Always ommitted? Also by interpreting the coefficients does the 0 1 coefficient mean that Housebank has 47.3% less effect on loanspread when the Duration is less than 10 years compared to the effect of housebank when Duration is equal or higher than 10 years? So do I compare the coefficients with the case where both Dummies are qual to 1 or where both Dummies are equal to 0?

[Marcel Issa]

I am still unsure about the interpretation of the interaction term. So usually the coefficient of a variable in a model that includes the interaction term of this variable with another, shows the effect of this variable when the variable that it interacts with is equal to zero. In my case I transformed this variable to a dummy and thus Stata returns me the coefficient of the continous duration variable and the interaction when the dummy of duration is equal to zero, which Stata wouldnt have showed me if I hadnt transformed duration if I am right? Still, I wonder why I only get two coefficients for the interaction. I assume they are both compared to the case where both dummies are equal to 1. But what aboout both dummies are equal to zero? I would also like to know if the effect of the interaction when both dummies are equal to 1 significantly differs from both dummies being equal to zero? I hope u can identify my uncertainty and can assist me on the interpretation.
Many thanks in advance.