[Bug]: Stratified Wilson and Newcombe CI's when zero successes are observed

[srigby-argx]

What happened?

In case there are no successes in one of the groups, I get the error: Error in [.default(tbl, "TRUE", ) : subscript out of bounds. Apparently this is because tbl has no row called "TRUE".

I don't know of any reason why we can't calculate a CI here; in theory it should be possible. The unstratified version apparently works fine.

Note that if I swap successes <-> failures then I get different errors.

sessionInfo()

> sessionInfo()
R version 4.2.2 (2022-10-31)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04.6 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] tern_0.9.0        rtables_0.6.12    magrittr_2.0.3    formatters_0.5.11

loaded via a namespace (and not attached):
 [1] rstudioapi_0.17.1 splines_4.2.2     tidyselect_1.2.0  munsell_0.5.0    
 [5] lattice_0.20-45   colorspace_2.1-0  R6_2.5.1          rlang_1.1.1      
 [9] fastmap_1.1.1     dplyr_1.1.4       rbibutils_2.2.16  tools_4.2.2      
[13] grid_4.2.2        broom_1.0.5       checkmate_2.3.0   gtable_0.3.3     
[17] cli_3.6.3         htmltools_0.5.8.1 survival_3.7-0    digest_0.6.32    
[21] tibble_3.2.1      lifecycle_1.0.4   Matrix_1.6-0      tidyr_1.3.1      
[25] purrr_1.0.4       ggplot2_3.5.1     Rdpack_2.6        vctrs_0.6.4      
[29] glue_1.6.2        stringi_1.7.12    compiler_4.2.2    pillar_1.10.1    
[33] generics_0.1.3    scales_1.3.0      backports_1.4.1   pkgconfig_2.0.3  
>

Relevant log output

set.seed(5)

library(tern)
#> Loading required package: rtables
#> Loading required package: formatters
#> Loading required package: magrittr
#> 
#> Attaching package: 'rtables'
#> The following object is masked from 'package:utils':
#> 
#>     str
#> Registered S3 method overwritten by 'tern':
#>   method   from 
#>   tidy.glm broom

grp <- factor(c(rep("T", 50), rep("C", 50)))
rsp <- as.logical(rbinom(n = 100, size = 1, prob = 0.2))
strata <- factor(sample(1:4, size = 100, replace = TRUE))

prop_wilson(rsp[grp == "C"], conf_level = 0.95)
#> [1] 0.2243695 0.4784617

prop_diff_strat_nc(rsp, grp, strata,
                   weights_method = "cmh")
#> $diff
#> [1] -0.128047
#> 
#> $diff_ci
#>       lower       upper 
#> -0.29644748  0.04999571

prop_strat_wilson(rsp[grp == "C"], strata[grp == "C"])
#> $conf_int
#>     lower     upper 
#> 0.2127322 0.4590773 
#> 
#> $weights
#>         1         2         3         4 
#> 0.2993060 0.2090818 0.3423156 0.1492966

# Change all controls to failures

rsp[grp == "C"] <- FALSE

prop_wilson(rsp[grp == "C"], conf_level = 0.95)
#> [1] 0.0000000 0.0713476

prop_diff_strat_nc(rsp, grp, strata,
                   weights_method = "cmh")
#> Error in `[.default`(tbl, "TRUE", ): subscript out of bounds

prop_strat_wilson(rsp[grp == "C"], strata[grp == "C"])
#> Error in `[.default`(tbl, "TRUE", ): subscript out of bounds

# Swap TRUEs and FALSEs
rsp <- !rsp

prop_wilson(rsp[grp == "C"], conf_level = 0.95)
#> [1] 0.9286524 1.0000000

prop_diff_strat_nc(rsp, grp, strata,
                   weights_method = "cmh")
#> Error in stats::prop.test(x, n, correct = correct, conf.level = strata_conf_level): 'conf.level' must be a single number between 0 and 1

prop_strat_wilson(rsp[grp == "C"], strata[grp == "C"])
#> Error in if (diff_v[length(diff_v)] < tol) break: missing value where TRUE/FALSE needed

^{Created on 2025-05-08 with reprex v2.1.1}

Code of Conduct

• I agree to follow this project's Code of Conduct.

Contribution Guidelines

• I agree to follow this project's Contribution Guidelines.

Security Policy

• I agree to follow this project's Security Policy.

[Melkiades]

Could you please render the code with the {reprex} package? so we can run it straightforwardly. Thanks

[srigby-argx]

Update: I checked the details and in fact the stratified Newcombe CI is not defined when one of the groups has only successes or only failures in each stratum. I.e., for each group there needs to be at least 1 stratum with a mixture of successes and failures. Otherwise Equation (6) in Yan and Su (2010) is unevaluable because the denominator is zero. Would be nice to have an informative error message when this happens.

The case when there is one stratum with 0 successes, and other strata have a mixture of successes and failures, still leads to an error using prop_diff_strat_nc. In this case it should be possible to calculate the CI - see the last example in Table 2 of Yan and Su (2010).

Thanks very much.

[Melkiades]

Update: I checked the details and in fact the stratified Newcombe CI is not defined when one of the groups has only successes or only failures in each stratum. I.e., for each group there needs to be at least 1 stratum with a mixture of successes and failures. Otherwise Equation (6) in Yan and Su (2010) is unevaluable because the denominator is zero. Would be nice to have an informative error message when this happens.

The case when there is one stratum with 0 successes, and other strata have a mixture of successes and failures, still leads to an error using prop_diff_strat_nc. In this case it should be possible to calculate the CI - see the last example in Table 2 of Yan and Su (2010).

Thanks very much.

Thanks for finding this error!! Could you try a PR for adding the informative error message? I will review it swiftly so to expedite the addition