Bug: Cannot reorder residual SD for mixed effects models

[mccarthy-m-g]

The coef_map argument is unable to reorder the residual SD for mixed effects models; the residual SD is always forced to the bottom. Here's a reprex:

library(modelsummary)
library(lme4)

sleepstudy_fit <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)

# Bug: The residual SD is always forced to the bottom of the estimates section.
modelsummary(
  sleepstudy_fit,
  coef_map = c("SD (Observations)", "(Intercept)")
)

	(1)
(Intercept)	251.405
	(6.825)
SD (Observations)	25.592

# As a consequence, it's impossible to group all the residuals together in the
# table.
modelsummary(
  sleepstudy_fit,
  coef_map = c(
    "SD (Intercept Subject)",
    "SD (Days Subject)",
    "SD (Observations)",
    "Cor (Intercept~Days Subject)"
  )
)

	(1)
SD (Intercept Subject)	24.741
SD (Days Subject)	5.922
Cor (Intercept~Days Subject)	0.066
SD (Observations)	25.592

^{Created on 2024-04-23 with reprex v2.0.2}

Session info:

R version 4.2.2 (2022-10-31)
Platform: x86_64-apple-darwin17.0 (64-bit)
Running under: macOS Mojave 10.14.6

Matrix products: default
BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRlapack.dylib

locale:
[1] en_CA.UTF-8/en_CA.UTF-8/en_CA.UTF-8/C/en_CA.UTF-8/en_CA.UTF-8

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

other attached packages:
[1] lme4_1.1-35.2      Matrix_1.6-4       modelsummary_2.0.0 devtools_2.4.5    
[5] usethis_2.1.6     

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.11         lattice_0.20-45     listenv_0.9.1       ps_1.7.5           
 [5] utf8_1.2.4          digest_0.6.33       mime_0.12           parallelly_1.36.0  
 [9] R6_2.5.1            backports_1.4.1     reprex_2.0.2        evaluate_0.23      
[13] pillar_1.9.0        rlang_1.1.2         rstudioapi_0.14     minqa_1.2.6        
[17] data.table_1.15.4   miniUI_0.1.1.1      performance_0.11.0  callr_3.7.3        
[21] nloptr_2.0.3        urlchecker_1.0.1    checkmate_2.3.1     effectsize_0.8.6   
[25] rmarkdown_2.25      splines_4.2.2       stringr_1.5.1       htmlwidgets_1.6.4  
[29] shiny_1.7.4         compiler_4.2.2      httpuv_1.6.7        xfun_0.43          
[33] pkgconfig_2.0.3     parameters_0.21.6   pkgbuild_1.4.3      clipr_0.8.0        
[37] globals_0.16.3      htmltools_0.5.7     insight_0.19.10     tibble_3.2.1       
[41] codetools_0.2-18    fansi_1.0.6         future_1.33.2       withr_3.0.0        
[45] later_1.3.0         tables_0.9.25       MASS_7.3-58.1       grid_4.2.2         
[49] nlme_3.1-160        xtable_1.8-4        lifecycle_1.0.4     magrittr_2.0.3     
[53] bayestestR_0.13.2   datawizard_0.10.0   future.apply_1.11.2 cli_3.6.2          
[57] stringi_1.8.3       cachem_1.0.8        fs_1.6.3            promises_1.2.0.1   
[61] remotes_2.4.2       ellipsis_0.3.2      generics_0.1.3      vctrs_0.6.5        
[65] boot_1.3-28         tools_4.2.2         tinytable_0.2.1     glue_1.6.2         
[69] purrr_1.0.2         processx_3.8.3      yaml_2.3.8          pkgload_1.3.4      
[73] parallel_4.2.2      fastmap_1.1.1       sessioninfo_1.2.2   memoise_2.0.1      
[77] knitr_1.45          profvis_0.3.7      

[mccarthy-m-g]

I'm new to using modelsummary so I'm still familiarizing myself with it, but I believe the issue here is that the residual SD and other variance components have different group identifiers, which prevents the desired reordering.

get_estimates(sleepstudy_fit)
                          term     estimate std.error conf.level   conf.low conf.high
1                  (Intercept) 251.40510485  6.824597       0.95 237.935457  264.8748
2                         Days  10.46728596  1.545790       0.95   7.416374   13.5182
3       SD (Intercept Subject)  24.74065800        NA       0.95         NA        NA
4            SD (Days Subject)   5.92213766        NA       0.95         NA        NA
5 Cor (Intercept~Days Subject)   0.06555124        NA       0.95         NA        NA
6            SD (Observations)  25.59179572        NA       0.95         NA        NA
  statistic df.error      p.value effect    group s.value
1 36.838090      174 4.372791e-84  fixed            276.9
2  6.771481      174 1.882719e-10  fixed             32.3
3        NA       NA           NA random  Subject      NA
4        NA       NA           NA random  Subject      NA
5        NA       NA           NA random  Subject      NA
6        NA       NA           NA random Residual      NA

Note that changing the shape argument to use different group identifiers will make the desired reordering possible (e.g., shape = term + effect + statistic ~ model). So I guess the issue at hand is whether or not this could be the default for mixed models, rather than the present default. Not exactly a bug, but still unexpected behaviour that appears bug-like at first.

[vincentarelbundock]

Yep, I think you're diagnostic is correct. You can use group_map to change the order of groups.

[vincentarelbundock]

That said, I do agree that this is counterintuitive. I pushed a commit to Github. If you install it an restart R, you’ll get this:

library(lme4)
library(modelsummary)
sleepstudy_fit <- lmer(Reaction ~ Days + I(Days^2) + (Days | Subject), sleepstudy)

modelsummary(
  sleepstudy_fit,
  coef_map = c("SD (Observations)", "(Intercept)", "Days")
)

SD (Observations)	25.534
(Intercept)	255.449
	(7.514)
Days	7.434
	(2.819)
Num.Obs.	180
R2 Marg.	0.280
R2 Cond.	0.800
AIC	1756.8
BIC	1779.2
ICC	0.7
RMSE	23.30

[mccarthy-m-g]

Thanks Vincent, great package by the way!

[vincentarelbundock]

Glad you like it!