Error when interactionR_boot() requires recoding

[ejvalencia]

Hello!

In implementing the interactionR_boot() function I encounter an error when a recode of the exposure/modifier variables is required. Consider the following example, which works without issue:

# Loading interactionR
if (!require("interactionR")) install.packages("interactionR")
library(interactionR)

# Loading RHC data
df <-
  
  ## Importing data & subsetting columns
  read.csv("https://hbiostat.org/data/repo/rhc.csv", header = TRUE) |>

  ## Subsetting columns
  subset(select = c(death, swang1, sex, adld3p)) |>
  
  ## Recoding binary values to 0/1
  transform(death = ifelse(death == 'Yes', 1, 0)) |>
  transform(swang1 = ifelse(swang1 == 'RHC', 1, 0)) |>
  transform(sex = ifelse(sex == 'Male', 1, 0))

# Implementing a logistic regression model and interactionR_boot()

  ## The logistic regression model
  mod.x <- glm(death ~ sex * swang1, family = binomial, data = df)
  
  ## Summarizing the model estimates
  cbind(Est. = exp(coef(mod.x)), exp(confint.default(mod.x)))
  
  ## THE ABOVE CODE RETURNS THE FOLLOWING OUTPUT
  ##
  ##                  Est.     2.5 %   97.5 %
  ## (Intercept) 1.6150160 1.4617940 1.784298
  ## sex         1.1008802 0.9603613 1.261960
  ## swang1      1.2954362 1.0915900 1.537349
  ## sex:swang1  0.9358933 0.7452079 1.175372
  
  ## Implementing interactionR::interactionR_boot()
  interactionR_boot(mod.x, em = TRUE, recode = TRUE, s = 100)$dframe
  
  ## THE ABOVE CODE RETURNS THE FOLLOWING OUTPUT
  ##
  ##                         Measures   Estimates      CI.ll    CI.ul            p
  ## 1                          OR00  1.00000000         NA       NA           NA
  ## 2                          OR01  1.29543621  1.0915900 1.537349 0.0030448253
  ## 3                          OR10  1.10088016  0.9603613 1.261960 0.1677553438
  ## 4                          OR11  1.33469612  1.1438060 1.557444 0.0002691867
  ## 5 OR(swang1 on outcome [sex==0]  1.29543621  1.0915900 1.537349 0.0030448253
  ## 6 OR(swang1 on outcome [sex==1]  1.21239002  1.0431779 1.409050 0.0119902673
  ## 7          Multiplicative scale  0.93589326  0.7452079 1.175372 0.5687153854
  ## 8                          RERI -0.06162025 -0.3162320 0.211492           NA
  ## 9                            AP -0.04616800 -0.2494047 0.155587           NA

In the above example, there is no need to recode either swang1 or sex because they are risk factors for the outcome, i.e., their corresponding ORs are >1.

However, when a recode is indeed required, as shown below:

# Implementing a logistic regression model and interactionR_boot()

  ## The logistic regression model, subset to complete case observations
  mod.x <- glm(death ~ sex * swang1, family = binomial, data = na.omit(df))
  
  ## Summarizing the model estimates
  cbind(Est. = exp(coef(mod.x)), exp(confint.default(mod.x)))
  
  ## THE ABOVE CODE RETURNS THE FOLLOWING OUTPUT
  ##
  ##                 Est.     2.5 %   97.5 %
  ## (Intercept) 0.8390805 0.7011339 1.004168
  ## sex         1.2651876 0.9917632 1.613994
  ## swang1      0.9852055 0.6728274 1.442613
  ## sex:swang1  1.0430454 0.6434684 1.690749

Then the following code:

  ## Implementing interactionR::interactionR_boot()
  interactionR_boot(mod.x, em = TRUE, recode = TRUE, s = 100)$dframe

Yields the following error:
Error in eval(mf, parent.frame()) : object 'dat.ir' not found

I have found a work-around whereby I manually specify dat.ir <- model.frame(mod.x) and then implement library(car), which seems to resolve the problem. However, I don't think this is the intended behavior or solution.

Note, this issue does not arise with interactionR::interactionR().