Conformability error r(503) when using RERI command

[Kate Dotsikas]

Hello,

I am trying to estimate interaction on an additive scale between two exposure variables, A and B. Exposure A is a 5-level categorigal variable measuring quintiles of neighbourhood deprivation, and exposure B is a 3-level categorical variable measuring individual-level socioeconomic status. I have therefore created dummy variables for each level of A and B (see code below). My outcome, Y, is a binary variable measuring smoking status. I have not included additional covariates in the model for now.

I am using the new reri in Stata 18 to estimate additive interaction. I have no problem running binomial regression using binreg for all combinations of A and B, but when I add the reri option, I get the 503 conformability error for all but one combination of A and B. Is anyone familiar with reri? Any advice would be very much appreciated! Thank you very much!

//additive interaction using reri

//create dummy variables for A (5-level variable with values 1-5)
generate A2=0 if A==1 
replace A2=1 if A==2

generate A3=0 if A==1 
replace A3=1 if A==3

generate A4=0 if A==1 
replace A4=1 if A==4

generateA5=0 if A==1 
replace A5=1 if A==5

//create dummy variables for B (3-level variable with values 0-2)
generate B2=0 if B==0 
replace B2=1 if B==1

generate B3=0 if B==0 
replace B3=1 if B==2

//multiplicative interaction
binreg Y A2#B2, rr 
binreg Y A2#B3, rr 

binreg Y A3#B2, rr 
​​​​​​​binreg Y A3#B3, rr 

​​​​​​​binreg Y A4#B2, rr 
​​​​​​​binreg Y A4#B3, rr 

​​​​​​​binreg Y A5#B2, rr 
​​​​​​​binreg Y A5#B3, rr 

//additive interaction
reri binreg Y A2#B2, rr //error
reri binreg Y A2#B3, rr //error

reri ​​​​​​​binreg Y A3#B2, rr 
reri ​​​​​​​binreg Y A3#B3, rr //error

reri ​​​​​​​binreg Y A4#B2, rr //error
reri ​​​​​​​binreg Y A4#B3, rr //error

reri ​​​​​​​binreg Y A5#B2, rr //error
reri ​​​​​​​binreg Y A5#B3, rr //error

[Jeff Pitblado (StataCorp)]

Thank you for showing us your code, but without an example dataset it is hard for us to help you.

Here is an example from the help file.

webuse nitrates
reri binreg tube drug nitrate, rr

Notice that drug and nitrate are not interacted. reri does the interaction of the first two binary predictor variables for you.

I expect you to get a syntax error with your specifications. For example, here is what I get when I interact drug and nitrate when using the reri prefix.

. reri binreg tube drug#nitrate, rr
too few variables specified
r(102);

When I try to reproduce your generated binary variables setup with a different dataset with categorical variables, I get the no observations error code (but not the message???, we try to will fix that soon) for the correct reri syntax and the above error message when I interact the generated binary indicator variables.

webuse nhanes2l
generate a2=0 if agegrp == 1
replace a2=1 if agegrp == 2
generate b2=0 if hsizgp == 1
replace b2=1 if hsizgp == 2
reri binreg heartatk a2 b2, rr
reri binreg heartatk a2#b2, rr

Here is the reri output from the above two commands.

. reri binreg heartatk a2 b2, rr
r(2000);

. reri binreg heartatk a2#b2, rr
too few variables specified
r(102);

If you can't share your data, via dataex, and cannot reproduce the problem with a simulated dataset, please share the Stata log that shows us what you typed and what Stata reported.

[Kate Dotsikas]

Hi Jeff,

Thank you so much for your reply. My mistake, when I copied my code into this forum I made a mistake that I did not make when actually running the code, hence the unnecessary interaction signs. Thanks for pointing that out.

I am unable to completely reproduce the error in an example data set, especially since I don't know why the error is occuring, but I have created a random sample of 100 observations of my data where I do get the error for two of the models. The remaining models get collinearity errors. Just for info, my real data set is over 170,000 observations with some missing data, this random sample includes only complete observations.

Thank you for your help!

//sample data of 100 observations

clear
input float(ID A B Y)
  2251 3 2 0
  2616 2 0 0
  2739 2 1 0
  8656 3 1 1
 11635 2 2 0
 12063 1 0 1
 14556 3 2 1
 17692 1 0 0
 22288 3 2 0
 23194 4 2 1
 26862 1 2 0
 27146 1 0 0
 29556 2 0 0
 32998 1 0 0
 35402 1 1 0
 35724 4 2 1
 37155 3 1 1
 37353 1 0 0
 40091 1 1 0
 40118 1 0 0
 44620 1 0 1
 46530 2 2 1
 48366 1 0 0
 48441 4 0 0
 50893 5 2 0
 50968 4 1 0
 50986 1 2 1
 53426 4 2 0
 53797 3 2 0
 55485 4 2 0
 56018 2 1 0
 56735 2 0 0
 59852 1 1 0
 60314 1 1 1
 63463 1 0 0
 63992 1 0 0
 66555 1 0 0
 67714 4 2 1
 70785 3 1 1
 71674 1 0 0
 72302 3 2 1
 72856 3 2 0
 73646 4 2 1
 73738 2 0 0
 75919 2 2 0
 76623 1 1 0
 77686 2 2 0
 78702 2 0 0
 79130 2 1 0
 79761 5 2 0
 82065 2 1 0
 83848 2 2 0
 84052 1 0 0
 86190 2 0 0
 89568 3 0 0
 93216 5 1 0
 96708 5 1 1
 96927 4 2 1
 99393 2 1 1
104569 3 2 0
104707 1 0 0
109109 2 2 0
110010 1 0 0
111407 2 0 0
113956 4 1 1
114172 3 0 0
116870 2 2 0
118758 1 1 0
126201 3 2 0
128127 1 0 0
128382 3 0 0
129519 4 2 0
130159 4 0 0
130356 5 2 0
132228 1 0 0
138178 1 0 0
140007 3 1 1
140070 3 0 0
140767 4 0 0
142048 2 1 1
143234 3 2 1
143501 1 1 0
145192 1 2 0
150448 1 0 0
153422 4 1 1
156159 1 1 0
158447 1 1 0
160731 3 0 0
161313 1 1 0
161759 1 0 0
163748 2 2 1
163834 3 2 1
167091 3 1 0
167868 1 1 1
168289 5 1 0
168775 2 0 1
170219 1 1 0
171844 3 0 0
171879 2 0 0
178080 1 0 0
end

//create dummy variables for A (5-level variable with values 1-5)
generate A2=0 if A==1 
replace A2=1 if A==2

generate A3=0 if A==1 
replace A3=1 if A==3

generate A4=0 if A==1 
replace A4=1 if A==4

generate A5=0 if A==1 
replace A5=1 if A==5

//create dummy variables for B (3-level variable with values 0-2)
generate B2=0 if B==0 
replace B2=1 if B==1

generate B3=0 if B==0 
replace B3=1 if B==2

//additive interaction
reri binreg Y A2 B2, rr //error 503 - conformability
reri binreg Y A2 B3, rr //error 503 - conformability

reri binreg Y A3 B2, rr //error 459 - collinearity or empty cell
reri binreg Y A3 B3, rr //error 459 - collinearity or empty cell

reri binreg Y A4 B2, rr //error 459 - collinearity or empty cell
reri binreg Y A4 B3, rr //error 459 - collinearity or empty cell

reri binreg Y A5 B2, rr //error 459 - collinearity or empty cell
reri binreg Y A5 B3, rr //error 459 - A5 has only one value in estimation sample of 24 observations

[Jeff Pitblado (StataCorp)]

Thank you for the data, that helped me find the source of the problem.

The conformability error you are getting is due to a bug in reri.
This bug is triggered when there is a missing value in the first
observation of either of the binary predictor variables. We hope to
have this fixed in a future update to Stata 18.

The other error messages are legitimate. To get a better sense of the
problem reri has with your model and data, try running the
underlying model without the reri prefix. For example, the
following shows that reri exits with an error complaining about a
problem with an element of the interaction of the binary variables. The
output from fitting the underlying model tells us that this element
prefectly predicts an outcome, meaning its effect is absolute (i.e.
infinte) and thus those observations must be removed in order to fit the
model for the remaining data (this also causes an empty cell in the
interaction). reri will only produce a table of results if the
interaction of the binary variables is fully estimable -- in this case
it is not.

. reri binreg Y A3 B2, rr //error 459 - collinearity or empty cell

Fitting binreg ...
interaction 1.A3#0b.B2 omitted due to collinearity or empty cell
r(459);

. binreg Y A3#B2, rr
note: 1.A3#0.B2 != 0 predicts failure perfectly;
      1.A3#0.B2 omitted and 6 obs not used.


Iteration 1:  Deviance =  262.2823
Iteration 2:  Deviance =  251.0156
Iteration 3:  Deviance =  250.1016
Iteration 4:  Deviance =  250.0744
Iteration 5:  Deviance =  250.0743
Iteration 6:  Deviance =  250.0743

Generalized linear models                         Number of obs   =         44
Optimization     : MQL Fisher scoring             Residual df     =         41
                   (IRLS EIM)                     Scale parameter =          1
Deviance         =  250.0743346                   (1/df) Deviance =   6.099374
Pearson          =    600000029                   (1/df) Pearson  =   1.46e+07

Variance function: V(u) = u*(1-u)                 [Bernoulli]
Link function    : g(u) = ln(u)                   [Log]

                                                  BIC             =   94.92256

------------------------------------------------------------------------------
             |                 EIM
           Y | Risk ratio   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       A3#B2 |
        0 1  |       1.75   1.631393     0.60   0.548     .2815315      10.878
        1 0  |          1  (empty)
        1 1  |        8.4   5.953796     3.00   0.003     2.093921    33.69755
             |
       _cons |   .0952381   .0640561    -3.50   0.000      .025486    .3558933
------------------------------------------------------------------------------
Note: _cons estimates baseline risk.

[Kate Dotsikas]

Hi Jeff,
Thank you very much for your help. Your explanation is very clear. I look forward to the bug being fixed in a future update, in the mean time I'll calculate the additive interaction by hand.
Best,
Kate

[Jeff Pitblado (StataCorp)]

The fix to reri is now live, as of Stata 18 update on 20dec2023.

[Mark Russell]

Dear Jeff/All,

I am also struggling with a conformability error with the RERI command, despite updating to the latest version of Stata 18.

* Example generated by -dataex-. For more info, type help dataex
clear
input double(n Age_band A2 A3) byte Ethnicity double(E2 E3 E4 outcome)
  1 3 . 1 1 0 0 0 0
  2 2 1 . 1 0 0 0 0
  3 3 . 1 1 0 0 0 0
  4 2 1 . 1 0 0 0 0
  5 2 1 . 1 0 0 0 0
  6 3 . 1 1 0 0 0 0
  7 3 . 1 1 0 0 0 0
  8 2 1 . 4 . . 1 0
  9 2 1 . 1 0 0 0 1
 10 2 1 . 1 0 0 0 0
 11 2 1 . 1 0 0 0 0
 12 3 . 1 1 0 0 0 0
 13 1 0 0 3 . 1 . 0
 14 3 . 1 1 0 0 0 0
 15 2 1 . 1 0 0 0 0
 16 2 1 . 1 0 0 0 0
 17 1 0 0 1 0 0 0 0
 18 2 1 . 1 0 0 0 0
 19 3 . 1 1 0 0 0 0
 20 2 1 . 1 0 0 0 0
 21 2 1 . 1 0 0 0 0
 22 2 1 . 1 0 0 0 0
 23 2 1 . 1 0 0 0 0
 24 2 1 . 1 0 0 0 0
 25 3 . 1 1 0 0 0 0
 26 3 . 1 1 0 0 0 0
 27 3 . 1 1 0 0 0 0
 28 1 0 0 1 0 0 0 0
 29 3 . 1 1 0 0 0 0
 30 2 1 . 1 0 0 0 0
 31 3 . 1 1 0 0 0 0
 32 2 1 . 1 0 0 0 0
 33 1 0 0 1 0 0 0 0
 34 3 . 1 1 0 0 0 0
 35 3 . 1 2 1 . . 1
 36 1 0 0 1 0 0 0 0
 37 3 . 1 1 0 0 0 0
 38 3 . 1 1 0 0 0 0
 39 1 0 0 1 0 0 0 0
 40 3 . 1 1 0 0 0 0
 41 2 1 . 1 0 0 0 0
 42 1 0 0 3 . 1 . 0
 43 2 1 . 1 0 0 0 0
 44 3 . 1 1 0 0 0 0
 45 2 1 . 4 . . 1 0
 46 2 1 . 1 0 0 0 0
 47 2 1 . 1 0 0 0 0
 48 2 1 . 1 0 0 0 0
 49 3 . 1 1 0 0 0 1
 50 3 . 1 1 0 0 0 0
 51 2 1 . 4 . . 1 1
 52 2 1 . 2 1 . . 1
 53 1 0 0 4 . . 1 0
 54 2 1 . 1 0 0 0 0
 55 3 . 1 3 . 1 . 0
 56 3 . 1 1 0 0 0 1
 57 3 . 1 1 0 0 0 0
 58 2 1 . 1 0 0 0 0
 59 2 1 . 1 0 0 0 0
 60 1 0 0 2 1 . . 0
 61 2 1 . 1 0 0 0 0
 62 3 . 1 1 0 0 0 0
 63 1 0 0 1 0 0 0 0
 64 2 1 . 2 1 . . 0
 65 2 1 . 1 0 0 0 0
 66 1 0 0 1 0 0 0 0
 67 3 . 1 1 0 0 0 0
 68 3 . 1 1 0 0 0 0
 69 2 1 . 2 1 . . 0
 70 3 . 1 1 0 0 0 0
 71 1 0 0 1 0 0 0 0
 72 2 1 . 1 0 0 0 1
 73 2 1 . 2 1 . . 0
 74 3 . 1 1 0 0 0 0
 75 2 1 . 1 0 0 0 0
 76 3 . 1 1 0 0 0 0
 77 3 . 1 1 0 0 0 0
 78 3 . 1 1 0 0 0 0
 79 1 0 0 2 1 . . 0
 80 3 . 1 1 0 0 0 1
 81 3 . 1 1 0 0 0 0
 82 2 1 . 1 0 0 0 0
 83 3 . 1 2 1 . . 0
 84 3 . 1 1 0 0 0 0
 85 3 . 1 1 0 0 0 0
 86 2 1 . 1 0 0 0 0
 87 1 0 0 1 0 0 0 0
 88 2 1 . 1 0 0 0 0
 89 2 1 . 1 0 0 0 0
 90 3 . 1 1 0 0 0 1
 91 3 . 1 1 0 0 0 0
 92 2 1 . 4 . . 1 0
 93 3 . 1 1 0 0 0 0
 94 2 1 . 1 0 0 0 0
 95 2 1 . 1 0 0 0 0
 96 2 1 . 1 0 0 0 0
 97 3 . 1 1 0 0 0 0
 98 3 . 1 1 0 0 0 0
 99 2 1 . 1 0 0 0 0
100 2 1 . 1 0 0 0 0
end
label def Age_band 1 "<40", modify
label def Age_band 2 "40-65", modify
label def Age_band 3 ">65", modify
label def Ethnicity 1 "White", modify
label def Ethnicity 2 "Black", modify
label def Ethnicity 3 "Asian", modify
label def Ethnicity 4 "Mixed/Other", modify
label def outcome 0 "No drug initiated", modify
label def outcome 1 "Drug initiated", modify

gen A2=0 if Age_band==1
replace A2=1 if Age_band==2

gen A3=0 if Age_band==1
replace A3=1 if Age_band==3

gen E2=0 if Ethnicity==1
replace E2=1 if Ethnicity==2

gen E3=0 if Ethnicity==1
replace E3=1 if Ethnicity==3

gen E4=0 if Ethnicity==1
replace E4=1 if Ethnicity==4

reri poisson outcome A2 E3

Fitting poisson ...

Interaction of A2 and E3 on an additive scale

Model: Poisson                                  Number of observations = 3,377

       A2#E3
        - +    0 1
        + -    1 0
        + +    1 1

------------------------------------------------------------------------------
             |               Robust
             |        ERR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       A2#E3 |
        - +  |          .          .        .       .            .           .
conformability error
r(503);

Whereas it seems to work ok for other levels of the variables:

. reri poisson outcome A3 E2

Fitting poisson ...

Interaction of A3 and E2 on an additive scale

Model: Poisson                                  Number of observations = 2,814

       A3#E2
        - +    0 1
        + -    1 0
        + +    1 1

------------------------------------------------------------------------------
             |               Robust
             |        ERR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       A3#E2 |
        - +  |   .7635601   .5371481     1.86   0.063    -.0291998    2.203691
        + -  |  -.6391482    .049558    -7.42   0.000    -.7243055   -.5276873
        + +  |  -.1924439   .3879773    -0.44   0.656    -.6850608    1.070707
-------------+----------------------------------------------------------------
        RERI |  -.3168558   .6462803    -0.49   0.624    -1.583542    .9498304
 Attr. prop. |  -.3923638   .9134194    -0.43   0.668    -2.182633    1.397905
Synergy ind. |   -1.54683          .        .       .            .           .
------------------------------------------------------------------------------
Notes: P>|z| for synergy index (SI) is for test H0: SI = 1.
       Some estimates of excess relative risk are not positive.

Any help would be much appreciated!

Thanks,
Mark

[Jeff Pitblado (StataCorp)]

Your data example does not reproduce the output you show.

Given the output from your first call to reri, I suspect your Stata does not have the update I reference in #6. Run update query to check for the latest Stata update.

Here is the output I get when running the above (ignoring the calls to generate since the A and E variables already exist from the dataex listing).

. reri poisson outcome A2 E3

Fitting poisson ...
interaction 1.A2#1.E3 omitted due to collinearity or empty cell
r(459);

The error message explains the exact problem with the dataset. Here is a tabulation that shows where the empty cell is:

. tabulate A2 E3

           |          E3
        A2 |         0          1 |     Total
-----------+----------------------+----------
         0 |         9          2 |        11 
         1 |        37          0 |        37 
-----------+----------------------+----------
     Total |        46          2 |        48

If after updating your Stata you still get the conformability error, please provide us with a dataset that reproduces the error.