如何用R中卡方检验高效处理多分类变量的相关性检验

Hey there! I totally get it—running 14×14 pairwise chi-squared tests manually is tedious and error-prone. Let’s walk through a couple of efficient ways to automate this process so you can get all your results in one go.

Method 1: Base R (No Extra Packages Needed) #

If you prefer sticking to base R, we can use combn to generate all variable pairs and lapply to run the tests across each pair. Here’s how:

# Step 1: Extract your 14 categorical variables from your data frame
# Replace the variable names below with your actual 14 variables
cat_vars <- DATA_BASE[, c("TYPE_PEAU", "SENSIBILITE", "VAR3", "VAR4", ...)]

# Step 2: Generate all unique pairwise combinations of variables
var_pairs <- combn(names(cat_vars), 2, simplify = FALSE)

# Step 3: Define a function to run chi-squared tests and format results
run_chi_sq <- function(pair) {
  # Create contingency table for the pair
  tbl <- table(cat_vars[[pair[1]]], cat_vars[[pair[2]]])
  # Run the test
  test_out <- chisq.test(tbl)
  # Return a tidy data frame with key results
  data.frame(
    Variable_1 = pair[1],
    Variable_2 = pair[2],
    Chi_Squared = round(test_out$statistic, 3),
    Degrees_of_Freedom = test_out$parameter,
    P_Value = round(test_out$p.value, 4),
    stringsAsFactors = FALSE
  )
}

# Step 4: Apply the function to all pairs and combine results
all_chi_results <- do.call(rbind, lapply(var_pairs, run_chi_sq))

# View the final results
print(all_chi_results)

Method 2: Use the rstatix Package (Simpler, Tidy Output) #

If you prefer a more streamlined workflow with tidy results, the rstatix package has a built-in function for pairwise chi-squared tests. It handles formatting automatically:

# Install the package if you haven't already
# install.packages("rstatix")
library(rstatix)

# Extract your categorical variables (same as Method 1)
cat_vars <- DATA_BASE[, c("TYPE_PEAU", "SENSIBILITE", "VAR3", ...)]

# Run pairwise chi-squared tests in one line
all_chi_results_rstatix <- pairwise_chisq_test(cat_vars, vars = everything())

# View the tidy results (includes adjustments for multiple comparisons if needed)
print(all_chi_results_rstatix)

Important Note: Handling Small Expected Frequencies #

Chi-squared tests rely on the assumption that most expected frequencies are ≥5. If more than 20% of your contingency table cells have expected values <5, the test results might be unreliable. We can modify our base R function to check for this and fall back to Fisher’s exact test when needed:

run_chi_or_fisher <- function(pair) {
  tbl <- table(cat_vars[[pair[1]]], cat_vars[[pair[2]]])
  expected_vals <- chisq.test(tbl)$expected
  # Check if >20% of cells have expected values <5
  low_expected <- sum(expected_vals < 5) / length(expected_vals) > 0.2
  
  if (low_expected) {
    warning(paste("Pair", paste(pair, collapse = " & "), "has too many small expected frequencies—using Fisher's exact test instead."))
    test_out <- fisher.test(tbl)
    return(data.frame(
      Variable_1 = pair[1],
      Variable_2 = pair[2],
      Test_Type = "Fisher's Exact",
      P_Value = round(test_out$p.value, 4),
      stringsAsFactors = FALSE
    ))
  } else {
    test_out <- chisq.test(tbl)
    return(data.frame(
      Variable_1 = pair[1],
      Variable_2 = pair[2],
      Test_Type = "Pearson's Chi-Squared",
      Chi_Squared = round(test_out$statistic, 3),
      Degrees_of_Freedom = test_out$parameter,
      P_Value = round(test_out$p.value, 4),
      stringsAsFactors = FALSE
    ))
  }
}

# Run the adjusted function
all_results_with_check <- do.call(rbind, lapply(var_pairs, run_chi_or_fisher))

This way, you’ll get reliable results without having to manually check each pair.

Hope this saves you a ton of time! If you hit snags with specific variable pairs or need help interpreting the output, feel free to ask.

内容的提问来源于stack exchange，提问作者Rprogrammer