Problem with multiy-way clustering by dyad and year following Cameron et al. (2006)

[Matthias Walter]

Hey everyone,

I am working with a replication set that estimates the impact of trade on conflict with logit and uses an undirected dyad year data set. One specification clusters the standard errors by dyad and year according to the Cameron et al. (2006) method. However, Stata issues the warning "gatt not found" every time I execute the file.
I am using Stata 15 and the relevant code looks like this:

global opt ",robust cluster(dyad)"
global trade "ltrl ztl"
global multitrade "lopenl"
global interactions "muldis bildis"
global controlgrav "ldis comlng contig colony comcol fta_f gatt"
global controlbarb   "peace votcol pol totwar diswar larea allied"
global oldwar "con1 con2 con3 con4 con5 con6 con7 con8 con9 con10"
global oldwar2 "con11 con12 con13 con14 con15 con16 con17 con18 con19 con20"
global time "yr1-yr61"
*
//SE multi-way clustering by dyad and year following the Cameron et al. (2006) method:
egen dyear = group(dyad year)
matrix drop _all
logit mid $trade $multitrade $interactions $controlbarb $controlgrav $time $oldwar $oldwar2 , robust cluster(dyad)
matrix Vdyad = e(V)
logit mid $trade $multitrade $interactions $controlbarb $controlgrav $time $oldwar $oldwar2 , robust cluster(year)
matrix Vyear = e(V)
logit mid $trade $multitrade $interactions $controlbarb $controlgrav $time $oldwar $oldwar2 , robust cluster(dyear)
matrix Vdyear = e(V)
mat b=e(b)
matrix Vtot = Vdyad + Vyear - Vdyear
matrix Bsmall = b[1,"ltrl".."gatt"]
matrix Vtotsmall = Vtot["ltrl".."gatt","ltrl".."gatt"]
ereturn post Bsmall Vtotsmall
ereturn display

The warning appears after the line

matrix Bsmall = b[1,"ltrl".."gatt"]

. I am not sure how to get the code working and would appreciate any help. Thank you very much.

Best regards

Matthias

[Clyde Schechter]

Well, the code seems to include the variable gatt. So my best guess is that the variable gatt is being omitted from the logistic regression analyses either because it is colinear with some other variable or perhaps due to perfect prediction. Without seeing the -logit- output I can't be sure of this, but flying blind, that's my best guess.

If that's not right, please show complete output from this set of commands.

Side issue: it is unsafe programming practice to use global macros except where there is no alternative. In this case, local macros are an easily available alternative.

[Matthias Walter]

Dear Clyde,

thank you for your answer. However, I think omission of "gatt" is not the reason.

The logit output of

logit mid $trade $multitrade $interactions $controlbarb $controlgrav $time $oldwar $oldwar2 , robust cluster(dyear)

immediately before Stata issues the warning, looks like this:

 logit mid $trade $multitrade $interactions $controlba
> rb $controlgrav $time $oldwar $oldwar2 , robust clust
> er(dyear)

note: yr1 omitted because of collinearity
note: yr2 omitted because of collinearity
note: yr3 omitted because of collinearity
note: yr61 omitted because of collinearity
Iteration 0: log pseudolikelihood = -9160.0062
Iteration 1: log pseudolikelihood = -9121.5094
Iteration 2: log pseudolikelihood = -7139.7451
Iteration 3: log pseudolikelihood = -5781.7226
Iteration 4: log pseudolikelihood = -4999.4233
Iteration 5: log pseudolikelihood = -4646.5963
Iteration 6: log pseudolikelihood = -4174.7558
Iteration 7: log pseudolikelihood = -4132.5008
Iteration 8: log pseudolikelihood = -4132.1939
Iteration 9: log pseudolikelihood = -4132.1938

Logistic regression Number of obs = 299,542
Wald chi2(96) = 7043.42
Prob > chi2 = 0.0000
Log pseudolikelihood = -4132.1938 Pseudo R2 = 0.5489

(Std. Err. adjusted for 299,542 clusters in dyear)
------------------------------------------------------------------------------
| Robust
mid | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ltrl | -.2816831 .1107804 -2.54 0.011 -.4988087 -.0645574
ztl | -.1696852 .1207617 -1.41 0.160 -.4063737 .0670033
lopenl | 1.521421 .3482049 4.37 0.000 .838952 2.20389
muldis | -.1927734 .0439256 -4.39 0.000 -.278866 -.1066809
bildis | .0451235 .0140407 3.21 0.001 .0176042 .0726428
peace | -.0196064 .0019994 -9.81 0.000 -.0235251 -.0156876
votcol | -.8354898 .1111199 -7.52 0.000 -1.053281 -.6176988
pol | -.1734717 .0576889 -3.01 0.003 -.2865399 -.0604034
totwar | .2356999 .0091461 25.77 0.000 .217774 .2536258
diswar | .0162581 .0579463 0.28 0.779 -.0973145 .1298307
larea | .16727 .0211027 7.93 0.000 .1259095 .2086305
allied | .0634088 .0935155 0.68 0.498 -.1198782 .2466959
ldis | -.7705435 .1304422 -5.91 0.000 -1.026206 -.5148816
comlng | .2389604 .090768 2.63 0.008 .0610584 .4168625
contig | 1.180967 .125122 9.44 0.000 .9357318 1.426201
colony | .3257941 .1476245 2.21 0.027 .0364554 .6151327
comcol | .1246893 .1190458 1.05 0.295 -.1086362 .3580148
fta_f | -.2837154 .1528753 -1.86 0.063 -.5833454 .0159146
gatt | .0845778 .0611329 1.38 0.167 -.0352405 .204396
yr1 | 0 (omitted)
yr2 | 0 (omitted)
yr3 | 0 (omitted)
yr4 | 1.099007 .6407719 1.72 0.086 -.156883 2.354897
yr5 | .3256495 .6748004 0.48 0.629 -.996935 1.648234
yr6 | 1.386339 .5403148 2.57 0.010 .3273415 2.445337
yr7 | 1.715505 .517751 3.31 0.001 .7007319 2.730279
yr8 | 1.442397 .5696231 2.53 0.011 .325956 2.558837
yr9 | .8098438 .6118651 1.32 0.186 -.3893898 2.009077
yr10 | 1.132092 .5661676 2.00 0.046 .022424 2.24176
yr11 | 1.398984 .5024139 2.78 0.005 .4142709 2.383697
yr12 | 1.135468 .4931756 2.30 0.021 .1688613 2.102074
yr13 | .7713696 .51707 1.49 0.136 -.242069 1.784808
yr14 | .5676775 .5732175 0.99 0.322 -.5558083 1.691163
yr15 | .5117561 .4927286 1.04 0.299 -.4539742 1.477486
yr16 | .5446511 .4682302 1.16 0.245 -.3730632 1.462365
yr17 | .1054876 .5270127 0.20 0.841 -.9274384 1.138414
yr18 | .5650935 .4881192 1.16 0.247 -.3916026 1.52179
yr19 | -.615368 .60816 -1.01 0.312 -1.80734 .5766038
yr20 | .6322047 .4911201 1.29 0.198 -.3303731 1.594782
yr21 | .5138907 .4967369 1.03 0.301 -.4596959 1.487477
yr22 | .5404488 .5059239 1.07 0.285 -.4511438 1.532041
yr23 | .49463 .5027108 0.98 0.325 -.4906651 1.479925
yr24 | .8810904 .4640019 1.90 0.058 -.0283367 1.790517
yr25 | .2714271 .5313844 0.51 0.609 -.7700671 1.312921
yr26 | 1.002598 .4620235 2.17 0.030 .097049 1.908148
yr27 | .7542277 .5052595 1.49 0.136 -.2360627 1.744518
yr28 | 1.07989 .4504998 2.40 0.017 .1969265 1.962853
yr29 | .3938823 .4845327 0.81 0.416 -.5557844 1.343549
yr30 | .8211113 .4554662 1.80 0.071 -.0715859 1.713809
yr31 | .9638972 .4594783 2.10 0.036 .0633364 1.864458
yr32 | 1.05334 .4630383 2.27 0.023 .1458019 1.960879
yr33 | .9768761 .4513419 2.16 0.030 .0922622 1.86149
yr34 | .752318 .4460672 1.69 0.092 -.1219575 1.626594
yr35 | 1.282131 .4302244 2.98 0.003 .4389071 2.125356
yr36 | .6622246 .4556938 1.45 0.146 -.2309188 1.555368
yr37 | .7030175 .4484018 1.57 0.117 -.1758338 1.581869
yr38 | .4654001 .4634996 1.00 0.315 -.4430425 1.373843
yr39 | -.1640528 .5072977 -0.32 0.746 -1.158338 .8302324
yr40 | .2973025 .5294433 0.56 0.574 -.7403874 1.334992
yr41 | .3105018 .4574789 0.68 0.497 -.5861403 1.207144
yr42 | -.7175371 .4917758 -1.46 0.145 -1.6814 .2463257
yr43 | -.2175545 .5278257 -0.41 0.680 -1.252074 .8169648
yr44 | .9413835 .4570702 2.06 0.039 .0455424 1.837225
yr45 | .9033584 .4466471 2.02 0.043 .0279462 1.778771
yr46 | .916239 .4505581 2.03 0.042 .0331612 1.799317
yr47 | .7530451 .4425851 1.70 0.089 -.1144058 1.620496
yr48 | 1.014713 .4280599 2.37 0.018 .1757308 1.853695
yr49 | .5301066 .4468959 1.19 0.236 -.3457933 1.406006
yr50 | .4008088 .4492178 0.89 0.372 -.479642 1.28126
yr51 | -.2211528 .4691557 -0.47 0.637 -1.140681 .6983755
yr52 | .8847973 .4489987 1.97 0.049 .004776 1.764819
yr53 | .6264586 .4886909 1.28 0.200 -.331358 1.584275
yr54 | .8290393 .4737594 1.75 0.080 -.0995121 1.757591
yr55 | 1.151282 .4483185 2.57 0.010 .2725941 2.02997
yr56 | .777546 .4768046 1.63 0.103 -.1569739 1.712066
yr57 | .4585431 .4853814 0.94 0.345 -.492787 1.409873
yr58 | .2079696 .4973917 0.42 0.676 -.7669002 1.182839
yr59 | .6279458 .4774949 1.32 0.188 -.3079271 1.563819
yr60 | .560491 .4657376 1.20 0.229 -.3523379 1.47332
yr61 | 0 (omitted)
con1 | 2.515713 .1070426 23.50 0.000 2.305913 2.725513
con2 | .693324 .1230568 5.63 0.000 .4521371 .9345109
con3 | .7286335 .1364371 5.34 0.000 .4612216 .9960453
con4 | .5031502 .1413471 3.56 0.000 .2261149 .7801855
con5 | .105 .1566648 0.67 0.503 -.2020573 .4120574
con6 | .3270146 .1635143 2.00 0.046 .0065324 .6474968
con7 | .0089965 .1587087 0.06 0.955 -.3020669 .3200598
con8 | .4167175 .1680716 2.48 0.013 .0873032 .7461318
con9 | -.1599342 .183897 -0.87 0.384 -.5203658 .2004974
con10 | .2357253 .1750691 1.35 0.178 -.1074038 .5788543
con11 | -.1887196 .2035503 -0.93 0.354 -.5876708 .2102317
con12 | .3044805 .2001996 1.52 0.128 -.0879035 .6968646
con13 | .2346122 .1900529 1.23 0.217 -.1378846 .6071089
con14 | -.3056957 .215135 -1.42 0.155 -.7273526 .1159612
con15 | .016666 .2123077 0.08 0.937 -.3994493 .4327814
con16 | .0742227 .2035501 0.36 0.715 -.3247281 .4731735
con17 | .0747167 .2303936 0.32 0.746 -.3768464 .5262799
con18 | .1191099 .2158417 0.55 0.581 -.3039321 .5421519
con19 | .4443655 .2356611 1.89 0.059 -.0175218 .9062527
con20 | .0814065 .2157455 0.38 0.706 -.3414469 .5042598
_cons | -3.745683 1.116557 -3.35 0.001 -5.934094 -1.557272
------------------------------------------------------------------------------

. matrix Vdyear = e(V)

. //the command e(b) produces the coefficient vector
. mat b=e(b)

. //Connect all three variance covariance matrices
. matrix Vtot = Vdyad + Vyear - Vdyear

. matrix Bsmall = b[1,"ltrl".."gatt"]
gatt not found
r(111);

Concerning the macros:they are used in the code of the replication set, which I simply adopted. I´ll try with locals.

Best regards,

Matthias

[Nick Cox]

Although that model looks way over-complicated for many tastes, this conversation suggests that the error is just in referencing matrix columns:

. sysuse auto, clear
(1978 Automobile Data)

. logit foreign mpg weight price

Iteration 0:   log likelihood =  -45.03321  
Iteration 1:   log likelihood = -22.244792  
Iteration 2:   log likelihood = -18.069284  
Iteration 3:   log likelihood = -17.184699  
Iteration 4:   log likelihood = -17.161975  
Iteration 5:   log likelihood = -17.161893  
Iteration 6:   log likelihood = -17.161893  

Logistic regression                             Number of obs     =         74
                                                LR chi2(3)        =      55.74
                                                Prob > chi2       =     0.0000
Log likelihood = -17.161893                     Pseudo R2         =     0.6189

------------------------------------------------------------------------------
     foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -.1210918   .0956855    -1.27   0.206     -.308632    .0664483
      weight |  -.0068497   .0019996    -3.43   0.001    -.0107688   -.0029306
       price |   .0009264   .0003074     3.01   0.003      .000324    .0015288
       _cons |   14.42237   5.414367     2.66   0.008      3.81041    25.03434
------------------------------------------------------------------------------

. mat b = e(b)

. mat li b

b[1,4]
       foreign:    foreign:    foreign:    foreign:
           mpg      weight       price       _cons
y1  -.12109183  -.00684974   .00092639   14.422375

. mat B = b[1, "weight".."price"]
price not found
r(111);

. mat B = b[1, "foreign:weight".."foreign:price"]

. mat li B

B[1,2]
       foreign:    foreign:
        weight       price
y1  -.00684974   .00092639

[Clyde Schechter]

Try this:

matrix Bsmall = b[1,"mid:ltrl".."mid:gatt"]

Added: Crossed with #4 which gives the same fix.

[Matthias Walter]

Thanks to both of you. It worked.