Feature request: making tbl_regression() compatible with svymultinom() / svy_vglm() outputs

[alexandrayas]

I have been using tbl_regression() to produce tables displaying multinom() outputs. However I need to weight my analysis and have been using svymultinom() (from CMAverse: https://github.com/BS1125/CMAverse) to do this. I would love to produce tables for these outputs too, however I don't think think this is supported currently.

Similarly I can produce tables for a vglm() (from VGAM) output, but not svy_vglm (from svyVGAM).

Is there a work around for this?

Thank you

[larmarange]

There are two things to consider:

1. Do we have a tidy() method for CMAverse::svymultinom(). I know that svrepmisc::svymultinom() implementation (using the same aproach) does have it.
2. tbl_regression() needs to bee more generic about multicomponent models, cf. #1540

[alexandrayas]

Thanks for your quick reply. I had another query I was about to post on stackoverflow but actually this relates to the same issue.

I've also been using the VGAM::vglm() to fit multinomial logit models as I thought vglm() models were compatible with tbl_regression() but I don't think it works in this case. I'd like an output similar to that produced when using multinom(), as you suggested on the other thread.

> multLogitVGLM <- vglm(cbind(e0, e1, e2, e3) ~ timeInt + sex + ethnicity + hhincome_11 + parent_dailsmk_12, family = multinomial(refLevel = 1), data = dfimpLong)

> multLogitVGLM

Call:
vglm(formula = cbind(e0, e1, e2, e3) ~ timeInt + sex + ethnicity + 
    hhincome_11 + parent_dailsmk_12, family = multinomial(refLevel = 1), 
    data = dfimpLong)


Coefficients:
         (Intercept):1          (Intercept):2          (Intercept):3              timeInt:1              timeInt:2 
           -1.07285969            -4.66330565            -2.44357706            -0.06668069             0.47041720 
             timeInt:3            sexFemale:1            sexFemale:2            sexFemale:3       ethnicityBAME*:1 
            0.23124890             0.17373448             0.80105971            -0.33116889             0.12846803 
      ethnicityBAME*:2       ethnicityBAME*:3   hhincome_11430-559:1   hhincome_11430-559:2   hhincome_11430-559:3 
           -0.71503275            -0.58453115            -0.21189162            -0.16407150            -0.25399205 
  hhincome_11240-429:1   hhincome_11240-429:2   hhincome_11240-429:3      hhincome_11<240:1      hhincome_11<240:2 
            0.12482068            -0.03994199             0.40814847            -0.24953751            -0.92927210 
     hhincome_11<240:3 parent_dailsmk_12Yes:1 parent_dailsmk_12Yes:2 parent_dailsmk_12Yes:3 
           -0.33789652            -0.55539843            -0.44771708             0.16109892 

Degrees of Freedom: 7119 Total; 7095 Residual
Residual deviance: 4173.659 
Log-likelihood: -2086.829 

This is a multinomial logit model with 4 levels

> tbl_regression(multLogitVGLM, tidy_fun = broom.helpers::tidy_parameters)
Error in `tibble::tibble()`:
! Tibble columns must have compatible sizes.
• Size 24: Existing data.
• Size 8: Column `variable`.
ℹ Only values of size one are recycled.
Backtrace:
  1. gtsummary::tbl_regression(multLogitVGLM, tidy_fun = broom.helpers::tidy_parameters)
 10. broom.helpers::tidy_identify_variables(., quiet = quiet)
 12. broom.helpers:::model_identify_variables.default(model)
 20. tibble::tibble(term = colnames(model_matrix), variable = variables[assign])

# below works fine

> multLogitVGLM <- vglm(e0 ~ timeInt + sex + ethnicity + hhincome_11 + parent_dailsmk_12, family = multinomial(refLevel = 1), data = dfimpLong)

> multLogitVGLM

Call:
vglm(formula = e0 ~ timeInt + sex + ethnicity + hhincome_11 + 
    parent_dailsmk_12, family = multinomial(refLevel = 1), data = dfimpLong)


Coefficients:
         (Intercept)              timeInt            sexFemale       ethnicityBAME*   hhincome_11430-559 
          0.90504708          -0.08317794          -0.07338119           0.11953831           0.21457882 
  hhincome_11240-429      hhincome_11<240 parent_dailsmk_12Yes 
         -0.19683651           0.33837123           0.30666168 

Degrees of Freedom: 2373 Total; 2365 Residual
Residual deviance: 2926.447 
Log-likelihood: -1463.224 

This is a multinomial logit model with 2 levels

> tbl_regression(multLogitVGLM, tidy_fun = broom.helpers::tidy_parameters)

[larmarange]

Once again, for a proper support, we need first to fix #1540 and then to update broom.helpers accordingly

Best regards