Error with multicategorical mediator

[olivierISCIII]

Dear medflex Team,

I tried to use the weighting-based approach with a multicategorical mediator variable and got the following error:

expData <- neWeight(cut(negaff,3) ~ att + gender + educ , family = multinomial(), data = UPBdata, FUN = vgam)
## Error in pred[i, as.character(x[i])] : incorrect number of dimensions

The version of medflex that I have installed is 0.6-6.
Best regards.

[martaczc]

I have a similar problem.
Here is a working example:

library(tidyverse)
library(medflex)
library(VGAM) 

my_data <- tibble(
  mediator = c("a", "b", "c", "b", "c", "c", "a", "a", "b", "b"),
  treatment = c(1, 0, 1, 1, 0, 1, 0, 0, 0, 1),
  outcome = c(0, 0, 1, 1, 1, 0, 1, 0, 1, 0)
)

model_w <- vglm(
  mediator ~ factor(treatment),
  data = my_data,
  family = multinomial
)

my_data_exp <- neWeight(model_w)
# Error in pred[i, as.character(x[i])] : incorrect number of dimensions

I use medflex 0.6-7 and VGAM 1.1-5

[jmpsteen]

Thanks for your interest in medflex! This issue should be fixed now in the latest development release 0.6-8 available from github (by a simple , :-)). The working example used to work in an older version of the package, but due to some changes along with insufficient testing, the updated code no longer seemed to be compatible with multinomial regression using vglm.

@olivierISCIII Please note that the first working example provided does not work due to use of the cut function in the formula argument. Medflex only handles the functions factor and some other convenience functions for modelling (e.g. splines or polynomials in neImpute). So please first recode this variable as below. With these changes your code should work.

UPBdata$negaffcat <- cut(UPBdata$negaff, 3)
expData <- neWeight(negaffcat ~ att + gender + educ , family = multinomial(), data = UPBdata, FUN = vgam)

Please close this issue if you agree it has been fixed and the working example runs without errors.

[anhthyngo]

Thanks for your interest in medflex! This issue should be fixed now in the latest development release 0.6-8 available from github (by a simple , :-)). The working example used to work in an older version of the package, but due to some changes along with insufficient testing, the updated code no longer seemed to be compatible with multinomial regression using vglm.

@olivierISCIII Please note that the first working example provided does not work due to use of the cut function in the formula argument. Medflex only handles the functions factor and some other convenience functions for modelling (e.g. splines or polynomials in neImpute). So please first recode this variable as below. With these changes your code should work.

UPBdata$negaffcat <- cut(UPBdata$negaff, 3)
expData <- neWeight(negaffcat ~ att + gender + educ , family = multinomial(), data = UPBdata, FUN = vgam)

Please close this issue if you agree it has been fixed and the working example runs without errors.

Hi I'm still getting 'Error in pred[i, as.character(x[i])] : incorrect number of dimensions' when running this line of code

UPBdata$negaffcat <- cut(UPBdata$negaff, 3)
expData <- neWeight(negaffcat ~ att + gender + educ , family = multinomial(), data = UPBdata, FUN = vgam)

[olivierISCIII]

It works fine for me, using version 0.6-8. Thanks.