Error: Custom function that has ~ (tilde) and/or $ (dollar sign) in the body in r

Question

I am trying to simplify my code by using custom/user-defined functions in r.

I want to turn the following into functions:

shapiro <- shapiro.test(data$count)

if(shapiro$p.value > 0.05) {
  require(tidyverse)
  bart <- bartlett.test(count ~ species, data=data)
} 

if ((shapiro$p.value > 0.05) && exists("bart")){
  if(bart$p.value > 0.05) { 
    OneWay <- aov(count ~ species, data = data) 
    oneway <- summary(OneWay) #normal distibution and equal varaince
  } else  {
    welch <- oneway.test(count ~ species, data = data) #Welch ANOVA, normal distribution, unequal variance
  }
} else {
  fligner <- fligner.test(count ~ species, data = data)
   #if not normal distribution, perform fligner-killen homogeneity of variance test
} 

The first set I have tried:

perform_shapiro <- function(df, contvar) {
   shapiro.test(df$contvar)
}
perform_shapiro(df = data, contvar = count)

Which returns the error: Error in shapiro.test(df$contvar) : is.numeric(x) is not TRUE.

The second set is successful with:

perform_barlett <- function(df, contvar, catvar) {
  if(shapiro$p.value > 0.05) {
    require(tidyverse)
    bart <- bartlett.test(contvar ~ catvar, data=df)
    return(bart)
  }
}
perform_bartlett(data, count, species)

But the third also has a new error:

perform_oneway_welch_fligner <- function(df, contvar, catvar){
      if ((shapiro$p.value > 0.05) && exists("bart")){
      if(bart$p.value > 0.05) { 
        OneWay <- aov(contvar ~ catvar, data = df) 
        oneway <- summary(OneWay) #normal distibution and equal varaince
        print(oneway)
      } else  {
        welch <- oneway.test(contvar ~ catvar, data = df) #Welch ANOVA, normal distribution, unequal variance
        print(welch)
      }
    } else {
      fligner <- fligner.test(contvar ~ catvar, data = df)
       #if not normal distribution, perform fligner-killen homogeneity of variance test
      print(fligner)
    } 
}
perform_oneway_welch_fligner(data, count, species)

Error in eval(predvars, data, env) : object 'count' not found.

Data

data <- iris
# rename columns
names(data)[names(data) == "Sepal.Length"] <- "count"

FYI shapiro is < 0.05, so bart should not exist, hence fligner should be returned. Contvar = continuous variable, catvar = categorical variable.

I thought the errors at first were due to $ and ~, but the second function works well and it has a tilde formula inside, so now I am unsure. Thanks!

Answer 1

Here is a full solution, answering to the OP's comments.

perform_shapiro <- function(df, contvar) {
  contvar <- as.character(substitute(contvar))
  shapiro.test(df[[contvar]])
}
perform_bartlett <- function(df, contvar, catvar) {
  contvar <- as.character(substitute(contvar))
  catvar <- as.character(substitute(catvar))
  fmla <- reformulate(catvar, contvar)
  shapiro <- shapiro.test(df[[contvar]])
  if(shapiro$p.value > 0.05) {
    bartlett.test(fmla, data = df)
  }
}
perform_oneway_welch_fligner <- function(df, contvar, catvar){
  contvar <- as.character(substitute(contvar))
  catvar <- as.character(substitute(catvar))
  fmla <- reformulate(catvar, contvar)
  shapiro <- shapiro.test(df[[contvar]])
  bart <- bartlett.test(fmla, data = df)
  if ((shapiro$p.value > 0.05) && !is.null(bart)){
    if(bart$p.value > 0.05) { 
      OneWay <- aov(fmla, data = df) 
      summary(OneWay) #normal distibution and equal varaince
    } else  {
      # Welch ANOVA, normal distribution, unequal variance
      oneway.test(fmla, data = df) 
    }
  } else {
    #if not normal distribution, perform fligner-killen homogeneity of variance test
    fligner.test(fmla, data = df)
  } 
}

data <- iris
names(data)[names(data) == "Sepal.Length"] <- "count"

perform_shapiro(df = data, contvar = count)
#> 
#>  Shapiro-Wilk normality test
#> 
#> data:  df[[contvar]]
#> W = 0.97609, p-value = 0.01018

# returns NULL invisiby
perform_bartlett(data, count, Species)

perform_oneway_welch_fligner(data, count, Species)
#> 
#>  Fligner-Killeen test of homogeneity of variances
#> 
#> data:  count by Species
#> Fligner-Killeen:med chi-squared = 11.618, df = 2, p-value = 0.003

^{Created on 2023-01-06 with reprex v2.0.2}

Answer 2

The first function could be solved using an anonymous function

perform_shapiro <- function(x) {
   shapiro.test(x)
}
perform_shapiro(data$count)

(sorry after playing around a bit I figure out that this works! It did not work in the third function (i.e. replacing the formulas with x))