如何用purrr的map函数逐行执行prop.test并将结果添加到数据框？

Hey there! I’ve dealt with this exact problem before, so let me walk you through two straightforward approaches to get those p-values and confidence intervals (CI) added to your dataframe. We’ll cover both base R and tidyverse methods so you can pick what fits your workflow best.

First, Let’s Create Sample Data #

To make this concrete, let’s start with a dummy dataframe that matches your structure (successes + total trials):

set.seed(123) # Ensures reproducible results
df <- data.frame(
  successes = sample(1:50, 10),
  trials = sample(50:100, 10)
)

Approach 1: Base R with apply() #

This method uses base R tools, no extra packages needed. We’ll write a custom function to process each row, then merge the results back into your original dataframe.

Step 1: Define a Row-Processing Function #

This function takes a row from your dataframe, runs the proportion test, and extracts the metrics you care about:

process_row <- function(row) {
  # Pull success count and total trials from the row
  success_num <- row["successes"]
  total_trials <- row["trials"]
  
  # Run prop.test (swap with binom.test if you need exact binomial tests)
  test_output <- prop.test(x = success_num, n = total_trials, correct = FALSE)
  # Note: `correct=FALSE` turns off continuity correction; adjust based on your needs
  
  # Extract key metrics
  p_val <- test_output$p.value
  ci_lower <- test_output$conf.int[1]
  ci_upper <- test_output$conf.int[2]
  # Format CI as a readable string (e.g., "(0.32, 0.58)")
  ci_string <- sprintf("(%.2f, %.2f)", ci_lower, ci_upper)
  
  # Return results as a named vector
  return(c(p_value = p_val, ci = ci_string, ci_low = ci_lower, ci_high = ci_upper))
}

Step 2: Apply the Function and Merge Results #

# Apply the function to every row, transpose the output to a dataframe
test_results <- as.data.frame(t(apply(df, 1, process_row)))

# Merge with original dataframe
final_df <- cbind(df, test_results)

# View the result
head(final_df)

Approach 2: Tidyverse with dplyr + broom #

If you prefer the tidyverse workflow, using dplyr::rowwise() and the broom package (which converts statistical outputs to tidy data frames) makes this super clean.

Step 1: Install/Load Required Packages #

install.packages(c("dplyr", "broom")) # Run once if not installed
library(dplyr)
library(broom)

Step 2: Process Rows and Extract Results #

final_df_tidy <- df %>%
  rowwise() %>% # Treat each row as an individual group
  mutate(
    # Run the test and store the result as a list column
    test_result = list(prop.test(x = successes, n = trials, correct = FALSE)),
    # Extract p-value using broom::tidy()
    p_value = tidy(test_result)$p.value,
    # Extract CI bounds
    ci_low = tidy(test_result)$conf.low,
    ci_high = tidy(test_result)$conf.high,
    # Format CI string
    ci = sprintf("(%.2f, %.2f)", ci_low, ci_high)
  ) %>%
  ungroup() %>% # Reset row-wise grouping
  select(-test_result) # Optional: Remove the raw test result column

# View the tidy result
head(final_df_tidy)

Key Notes #

• Switching to binom.test: Just replace prop.test with binom.test in either method—both functions return objects with the same structure for p-values and confidence intervals.
• Continuity Correction: The correct=FALSE argument in prop.test turns off continuity correction. If you want it enabled (the default), just remove that argument.
• Additional Metrics: If you want to include the estimated proportion, add estimate = tidy(test_result)$estimate (tidyverse) or estimate = test_output$estimate (base R) to your output.

内容的提问来源于stack exchange，提问作者Luise