Error code "r(123); invalid numlist has too many elements"

[Sigrid Vikjord]

Dear Statalist,

I'm running this code, generating cubic splines for the variable FFMI against HRs. This gives me the error code "r(123); invalid numlist has too many elements". Why does this happen, and how can I improve my code to avoid this?

foreach num of numlist 3/7{
        preserve
        keep if cohort==1 & PartAg_NT4BLQ1>40.0 & PartAg_NT4BLQ1!=.    
        stset enddate_ffmi, id(PID_109925) failure(RegisStat==5) origin(time BirthYear) enter(time PartDat_NT4BLQ1) scale(365.25)
        mkspline spl_ffmi = FFMI, cubic nknots(`num') displayknots            
        stcox c.spl_ffmi* i.Sex, vce(robust)
        estat ic
        levelsof FFMI, local(levels)
        xblc spl_ffmi*, covname(FFMI) at(`levels') reference(18.5) eform gen(wffmi hr lb ub)
        keep if wffmi>12.0 & wffmi<29.0
        twoway rline lb ub wffmi, lp(solid) lc(red) lw(medium) || line hr wffmi, lc(blue) lp(solid) lw(medium) ///
        legend(off) ylabel(, nogrid format(%3.2f)) yscale(log) yticks() xlabel(12(1)29, format(%3.0f)) xticks() xtitle("{bf:FFM     (kg/m{sup:2})}") ytitle("{bf:HR (95% CI) for all-cause mortality}") ///
        yline(1, lc(black) lp(dash) lw(medthin)) name(no_copd_death_spl_ffmi_knot`num', replace) saving("M:\FFMIsplines.gph",             replace)
        restore
}

* Example generated by -dataex-. For more info, type help dataex
clear
input str13 PID_109925 float FFMI
"1099250000068" 20.257673
"1099250000075"  16.41549
"1099250000303" 16.261368
"1099250000433" 18.844908
"1099250000440" 19.078123
"1099250000556" 16.711199
"1099250000594" 16.096815
"1099250000747" 17.166037
"1099250000808"  23.12851
"1099250000815" 16.872831
"1099250000839"  21.20099
"1099250000976" 18.357824
"1099250001072" 15.002635
"1099250001102" 16.622509
"1099250001218"  15.96405
"1099250001379" 17.252134
"1099250001645" 18.876345
"1099250001690" 16.994984
"1099250001874"  20.27285
"1099250001911"  24.86499
"1099250001980" 15.426138
"1099250001997"  20.51911
"1099250002000" 21.389755
"1099250002017" 18.948357
"1099250002031" 20.587954
"1099250002079" 16.991798
"1099250002086"  22.66436
"1099250002154" 17.340462
"1099250002185" 21.074715
"1099250002222"  14.76981
"1099250002307"  20.29168
"1099250002499"  20.79452
"1099250002604" 15.637706
"1099250002680" 13.729548
"1099250002727" 24.500887
"1099250002840" 17.770155
"1099250002956" 17.600523
"1099250002970" 16.451637
"1099250003007"  22.63712
"1099250003052" 19.406385
"1099250003144" 16.657373
"1099250003243"  20.07525
"1099250003366" 17.922968
"1099250003373" 18.827162
"1099250003618"  16.72992
"1099250003700" 15.712954
"1099250003731" 18.618073
"1099250003984" 16.122509
"1099250004004" 22.994595
"1099250004035" 19.515856
"1099250004042" 14.438632
"1099250004073" 17.067698
"1099250004103" 17.162767
"1099250004134" 20.198105
"1099250004141"  21.98878
"1099250004196" 20.644033
"1099250004219"  19.55816
"1099250004233" 19.101213
"1099250004295"  19.83154
"1099250004486" 14.690882
"1099250004516" 16.955017
"1099250004615" 19.148285
"1099250004752"  19.73008
"1099250004783" 17.988977
"1099250004868" 16.573256
"1099250004899" 17.834265
"1099250004998" 17.433004
"1099250005025"  20.97489
"1099250005049" 18.173464
"1099250005216"  21.99749
"1099250005278"  19.14934
"1099250005285" 16.772224
"1099250005360"  15.84676
"1099250005421"  22.20912
"1099250005445" 15.887088
"1099250005544" 17.749592
"1099250005681" 17.077469
"1099250005728"   21.3407
"1099250005797" 16.324526
"1099250005803"  19.90521
"1099250005834" 20.154686
"1099250005865" 16.670069
"1099250005988"  20.57978
"1099250006190" 17.966629
"1099250006213" 24.858364
"1099250006244" 17.207388
"1099250006275" 18.172617
"1099250006305" 19.501144
"1099250006374" 19.584696
"1099250006398" 16.620935
"1099250006435" 19.601484
"1099250006558" 18.326319
"1099250006732" 17.016905
"1099250006749"  21.97266
"1099250006787" 17.824898
"1099250006794" 17.189758
"1099250006909" 16.591694
"1099250006916" 18.457619
"1099250006923"  19.51468
"1099250006985" 17.643011
end

[Richard Williams]

I'm guessing it is because FFMI has too many possible values. The help for levelsof says "levelsof may hit the limits imposed by your Stata. However, it is typically used when the number of distinct values of varname is not extremely large."

[daniel klein]

I also guess that there are too many levels; somewhere above 250. But it is probably not levelsof that chokes here (it should be able to handle many more levels). My guess is that the at() option of xblc (probably from SJ?), to which the levels are fed, cannot handle the number of levels.

[Richard Williams]

The problems isn't replicable because not all the variables are included in the dataex extract.

I would probably have the program set to stop immediately after the levelsof command, so you could see if it was already bombing at that point.

I suspect that, whether it runs or not, the command

levelsof FFMI, local(levels)

is mistaken anyway. Just with the data presented the value of FFMI is uniquie, i.e. you already have 100 levels with only 100 cases. I bet you meant to use some other variable here; or, you are just misunderstanding how to do what you want to do.

In the help examples for the xblc command, in the -at- option they only list a few of the possible values for covname() variable, not every possible value. I suspect Sigrid wants to do the same.

[Sigrid Vikjord]

Dear Mr. William and mr. Klein,

Thank you for valuable input. The syntax above works fine with the variable Ffm_NT4BLM, but not FFMI which is derived from it. How could the number of levels be different between these two?

gen Height_m = Hei_NT4BLM/100
gen FFMI = ((Ffm_NT4BLM)/(Height_m^2))

* Example generated by -dataex-. For more info, type help dataex
clear
input double(Hei_NT4BLM Ffm_NT4BLM) float FFMI
180.5   66 20.257673
  159 41.5  16.41549
172.7 48.5 16.261368
163.7 50.5 18.844908
187.4   67 19.078123
149.8 37.5 16.711199
167.2   45 16.096815
160.1   44 17.166037
185.4 79.5  23.12851
162.4 44.5 16.872831
  183   71  21.20099
161.7   48 18.357824
158.1 37.5 15.002635
174.3 50.5 16.622509
173.4   48  15.96405
159.7   44 17.252134
154.4   45 18.876345
169.8   49 16.994984
187.8 71.5  20.27285
183.8   84  24.86499
151.7 35.5 15.426138
180.7   67  20.51911
  186   74 21.389755
174.2 57.5 18.948357
  177 64.5 20.587954
151.5   39 16.991798
  170 65.5  22.66436
168.1   49 17.340462
182.9 70.5 21.074715
160.4   38  14.76981
176.9 63.5  20.29168
176.8   65  20.79452
175.2   48 15.637706
160.8 35.5 13.729548
183.5 82.5 24.500887
172.7   53 17.770155
  166 48.5 17.600523
156.9 40.5 16.451637
177.1   71  22.63712
175.1 59.5 19.406385
161.6 43.5 16.657373
  199 79.5  20.07525
164.5 48.5 17.922968
  180   61 18.827162
168.5 47.5  16.72992
171.1   46 15.712954
161.4 48.5 18.618073
165.2   44 16.122509
180.6   75 22.994595
182.5   65 19.515856
165.4 39.5 14.438632
157.8 42.5 17.067698
155.5 41.5 17.162767
178.7 64.5 20.198105
176.5 68.5  21.98878
173.3   62 20.644033
181.6 64.5  19.55816
164.2 51.5 19.101213
176.1 61.5  19.83154
158.7   37 14.690882
  170   49 16.955017
163.2   51 19.148285
179.4 63.5  19.73008
169.2 51.5 17.988977
166.6   46 16.573256
166.6 49.5 17.834265
170.2 50.5 17.433004
191.6   77  20.97489
177.1   57 18.173464
187.7 77.5  21.99749
182.1 63.5  19.14934
157.3 41.5 16.772224
162.8   42  15.84676
186.2   77  22.20912
168.3   45 15.887088
172.8   53 17.749592
159.6 43.5 17.077469
182.4   71   21.3407
160.4   42 16.324526
181.4 65.5  19.90521
176.8   63 20.154686
164.3   45 16.670069
168.6 58.5  20.57978
164.3 48.5 17.966629
  186   86 24.858364
180.4   56 17.207388
159.1   46 18.172617
173.2 58.5 19.501144
170.6   57 19.584696
171.7   49 16.620935
182.8 65.5 19.601484
  166 50.5 18.326319
160.8   44 17.016905
  192   81  21.97266
165.8   49 17.824898
171.4 50.5 17.189758
165.6 45.5 16.591694
162.1 48.5 18.457619
164.8   53  19.51468
165.8 48.5 17.643011
end

[Richard Williams]

The syntax above works fine with the variable Ffm_NT4BLM, but not FFMI which is derived from it. How could the number of levels be different between these two?

It isn't like you are converting from feet to inches. There isn't a one to one correspondence between the two variables. Because you divide the original variable by Height_m^2, which has a huge number of unique values, FFMI could easily wind up with a unique value for every single case. See the following:

. levelsof Ffm_NT4BLM, local(levels1)
35.5 37 37.5 38 39 39.5 40.5 41.5 42 42.5 43.5 44 44.5 45 45.5 46 47.5 48 48.5 49 49.5 50.5 51 51.5 53 56 57 57.5 58.5 59.5 61 61.5 62 63 63
> .5 64.5 65 65.5 66 67 68.5 70.5 71 71.5 74 75 77 77.5 79.5 81 82.5 84 86

. di r(r)
53

. levelsof FFMI, local(levels2)
13.72954845428467 14.43863201141357 14.69088172912598 14.76980972290039 15.00263500213623 15.42613792419434 15.63770580291748 15.71295356750
> 488 15.84675979614258 15.88708782196045 15.96405029296875 16.09681510925293 16.12250900268555 16.26136779785156 16.32452583312988 16.41548
> 919677734 16.45163726806641 16.57325553894043 16.59169387817383 16.62093544006348 16.62250900268555 16.65737342834473 16.67006874084473 16
> .7111988067627 16.72991943359375 16.77222442626953 16.87283134460449 16.95501708984375 16.99179840087891 16.9949836730957 17.0169048309326
> 2 17.06769752502441 17.07746887207031 17.16276741027832 17.16603660583496 17.18975830078125 17.20738792419434 17.25213432312012 17.3404617
> 3095703 17.43300437927246 17.60052299499512 17.64301109313965 17.74959182739258 17.77015495300293 17.82489776611328 17.83426475524902 17.9
> 229679107666 17.96662902832031 17.98897743225098 18.17261695861816 18.17346382141113 18.32631874084473 18.35782432556152 18.45761871337891
>  18.61807250976563 18.82716178894043 18.84490776062012 18.87634468078613 18.94835662841797 19.07812309265137 19.1012134552002 19.148284912
> 10938 19.14933967590332 19.40638542175293 19.50114440917969 19.51468086242676 19.51585578918457 19.55816078186035 19.58469581604004 19.601
> 48429870605 19.73007965087891 19.83153915405273 19.90521049499512 20.07525062561035 20.15468597412109 20.19810485839844 20.2576732635498 2
> 0.27285003662109 20.29167938232422 20.51910972595215 20.57978057861328 20.58795356750488 20.64403343200684 20.79451942443848 20.9748897552
> 4902 21.07471466064453 21.20099067687988 21.34070014953613 21.38975524902344 21.97266006469727 21.98877906799316 21.99748992919922 22.2091
> 1979675293 22.63711929321289 22.66436004638672 22.99459457397461 23.12850952148438 24.50088691711426 24.85836410522461 24.864990234375

. di r(r)
100

If you have hundreds or thousands of cases the results will be even more striking.

I don't know what you are trying to do, but in the examples for xblc, they plugged in a set of 10 possible values, not hundreds or thousands.