Search code examples
rfunctionpackage

Question regarding error: $ operator is invalid for atomic vectors

I have created this function to automate multinomial regression:

multinom<-function(form,ref,db){
  library(VGAM)
  mod<-vglm(formula(form),multinomial(refLevel = ref),data = db)
  t1<-exp(cbind(OR = coef(mod), confint(mod)))
  p_val<-coef(summary(mod))[,4]
  x<-data.frame(cbind(t1,p_val))
  x$prob<-x[,1]/(1+x[,1])
  colnames(x)<-c("OR","CI_min","CI_max","P_val","Probablity")
  return(x)
}

When this function is used in a script, it works just fine. I have also added this function to a package I am writing:

##############################
#' Statistical analysis: Multinomial Regression
#'
#'The purpose of this function is to calculate the Odds Ratio for a statistical analysis that involves a categorical dependent variable, with more than two unordered levels or classes, and different independent variables of any kind.
#'The Odds Ratio is obtained using a multinominal logit regression. This is an extension of the binomial logistic regression to allow for a categorical dependent variable with more than two unordered categories.
#'In multinomial regression we need to define a reference variable category. The model will determine several binomial distribution parameters with respect to the reference category.
#' @param form model formula. For example: `dep_var~conf_var1+conf_var2+conf_var3`.
#' @param ref reference category from the dependent variable. If its a character variable the reference level must be included in `""`.
#' @param df D.B_Final database obtained from the data incorporation step.
#' @return A table with the Odds ratio, its 95% CI and the associated p-value.
#' @export
#' @examples
#' # Basic Usage of Multinomial Regression.
#' multinom(cov_2~cov_3+var_con,"A",D.B_Final)
#'
multinom<-function(form,ref,db){
  library(VGAM)
  mod<-vglm(formula(form),multinomial(refLevel = ref),data = db)
  t1<-exp(cbind(OR = coef(mod), confint(mod)))
  p_val<-coef(summary(mod))[,4]
  x<-data.frame(cbind(t1,p_val))
  x$prob<-x[,1]/(1+x[,1])
  colnames(x)<-c("OR","CI_min","CI_max","P_val","Probablity")
  return(x)
}

When this function is used inside the corresponding package, specifically as zoombedo::multinorm, I encounter the error message: Error in object$coefficients: $ operator is invalid for atomic vectors.

Can anyone help me to understand what is going on here?

Solution

  • I don't have a perfect solution but I think I can explain the problem. The error is almost certainly coming from your coef(summary(mod))[,4]. This should be calling the coef() method for an object of class summary.vglm:

    selectMethod("coef", "summary.vglm")
Method Definition:

function (object, ...) 
object@coef3
<bytecode: 0x5617c6d9d8e8>
<environment: namespace:VGAM>

    This is apparently not found in the package context. Instead, what you get is stats:::coef.default():

    function (object, complete = TRUE, ...) 
{
    cf <- object$coefficients
    if (complete) 
        cf
    else cf[!is.na(cf)]
}

    Since the object doesn't have an $coefficient element, you get the error.

    The hacky way to solve this would be to access the @coef3 slot directly.

    Using importMethodsFrom("VGAM", "summary") in your NAMESPACE file(or @importMethodsFrom if you're using roxygen) and maybe importClassesFrom("VGAM", "summary.vglm") might solve the problem (see the R Extensions document), but I haven't tested.