如何为twoway包编写长表转宽表的公式方法

Got it, let's get that formula method working for your twoway package! The main hurdle here is dynamically translating the formula components into a reshape operation that converts your long-form data into the wide format expected by twoway.default(). Here's a complete, working implementation with explanations:

Key Fixes & Explanation #

• Correctly extract formula components: We'll pull the response variable, row variable, and column variable directly from the parsed terms object to avoid confusion.
• Dynamically build the reshape formula: Since we can't pass raw variable names directly to reshape functions in the function environment, we'll construct a formula string and convert it to a formula object.
• Clean up the wide data: Ensure row names match the row variable values, and drop any extra ID columns before passing to the default method.

Complete Working Code #

#' Formula method for twoway analysis of long-form data
#'
#' Analyze a two-way table from long-form data using a formula of the form
#' \code{response ~ row + column}
#'
#' @param formula A formula of the form \code{response ~ rowvar + colvar}
#' @param data The data frame containing the variables in the formula
#' @param subset An expression to subset the data (unused)
#' @param na.action What to do with NAs? (unused)
#' @param ... other arguments, passed down to twoway.default()
#' @importFrom stats terms
#' @importFrom reshape2 dcast
#'
twoway.formula <- function(formula, data, subset, na.action, ...) {
  # Validate formula input
  if (missing(formula) || !inherits(formula, "formula"))
    stop("'formula' is missing or not a valid formula object")
  if (length(formula) != 3L)
    stop("Formula must have both a left-hand (response) and right-hand (row + column) side")
  
  # Parse terms to extract variables
  tt <- terms(formula, data = data)
  term_labels <- attr(tt, "term.labels")
  
  # Ensure no interactions and exactly two RHS variables
  if (any(attr(tt, "order") > 1))
    stop("Interactions are not allowed in the formula")
  if (length(term_labels) != 2)
    stop("Formula must include exactly two right-hand side variables (row + column)")
  
  # Extract variable names
  response_var <- as.character(formula[[2]])
  row_var <- term_labels[1]
  col_var <- term_labels[2]
  
  # Get the data (handle subset/na.action if needed later)
  edata <- eval(match.call()$data, parent.frame())
  
  # Dynamically build the dcast formula: row_var ~ col_var
  reshape_formula <- as.formula(paste(row_var, "~", col_var))
  
  # Reshape long to wide
  wide_data <- dcast(data = edata, formula = reshape_formula, value.var = response_var)
  
  # Set row names to the row variable values, drop the ID column
  rownames(wide_data) <- wide_data[[row_var]]
  wide_matrix <- as.matrix(wide_data[, -1, drop = FALSE])
  
  # Call the default method with the wide matrix
  twoway.default(wide_matrix, ...)
}

How to Test It #

Using your example data:

# Create the long-form data as you did
library(reshape2)
data(taskRT)
long <- melt(as.matrix(taskRT))
colnames(long) <- c("task", "topic", "RT")

# Now call the formula method
twoway(RT ~ task + topic, data = long)

This should produce the exact same output as twoway(taskRT).

Alternative: Using tidyr::pivot_wider (Modern Approach) #

If you prefer using tidyr over reshape2, replace the dcast section with:

library(tidyr)
wide_data <- pivot_wider(edata, 
                         id_cols = all_of(row_var), 
                         names_from = all_of(col_var), 
                         values_from = all_of(response_var))

Just make sure to add @importFrom tidyr pivot_wider to the function's roxygen comments.

内容的提问来源于stack exchange，提问作者user101089