How to Randomly Replace Values in a Matrix to NAs

NAs in R
Learn how to randomly replace values in a matrix to nas with this comprehensive R tutorial. Includes practical examples and code snippets.
Published

August 16, 2022

Introduction

Randomly replacing values with NAs in a matrix is a common technique used in data science for testing missing data handling methods, creating realistic datasets with missing values, or simulating data collection problems. This approach allows you to control the proportion and pattern of missing data in your analysis.

Getting Started

library(tidyverse)
set.seed(123)  # For reproducible results

Example 1: Basic Random NA Replacement

The Problem

We need to randomly introduce missing values into a complete matrix to simulate real-world data collection scenarios where some observations are unavailable.

Step 1: Create a Sample Matrix

First, let’s create a simple numeric matrix to work with.

# Create a 5x4 matrix with numbers 1-20
sample_matrix <- matrix(1:20, nrow = 5, ncol = 4)
colnames(sample_matrix) <- c("A", "B", "C", "D")
print(sample_matrix)

This creates a complete matrix with no missing values that we can use for demonstration.

Step 2: Generate Random Positions

We need to identify which positions in the matrix will become NA values.

# Calculate total number of elements
total_elements <- nrow(sample_matrix) * ncol(sample_matrix)
# Choose 30% of positions randomly
na_count <- round(total_elements * 0.3)
na_positions <- sample(total_elements, na_count)

This randomly selects 30% of the matrix positions to convert to NA values.

Step 3: Replace Selected Values with NAs

Now we’ll replace the selected positions with NA values.

# Create a copy and replace values
matrix_with_nas <- sample_matrix
matrix_with_nas[na_positions] <- NA
print(matrix_with_nas)

The matrix now contains randomly distributed missing values while preserving the original structure.

Example 2: Practical Application with Real Data

The Problem

Let’s apply this technique to the mtcars dataset, converting it to a matrix and introducing missing values to test how different analysis methods handle incomplete data.

Step 1: Prepare the Data Matrix

We’ll select numeric columns from mtcars and convert them to a matrix format.

# Select key numeric variables and convert to matrix
car_matrix <- mtcars |>
  select(mpg, hp, wt, qsec) |>
  as.matrix()

head(car_matrix)

This creates a matrix with four important car characteristics that we can work with.

Step 2: Create a Targeted NA Pattern

Instead of completely random replacement, let’s create a more realistic pattern where missing values are more likely in certain ranges.

# Get positions of high horsepower cars (hp > 150)
high_hp_rows <- which(car_matrix[, "hp"] > 150)
# Randomly select 50% of these positions for NA replacement
target_positions <- sample(high_hp_rows, length(high_hp_rows) * 0.5)

This targets specific rows based on a condition, simulating real scenarios where certain types of observations are more likely to have missing data.

Step 3: Apply Selective NA Replacement

Now we’ll introduce NAs in multiple columns for the selected rows.

# Create copy of matrix
car_matrix_na <- car_matrix
# Replace weight values for selected high-HP cars
car_matrix_na[target_positions, "wt"] <- NA
# Also introduce some random NAs in mpg (20% chance)
random_mpg <- sample(nrow(car_matrix), nrow(car_matrix) * 0.2)
car_matrix_na[random_mpg, "mpg"] <- NA

This creates a more realistic missing data pattern with both systematic and random missing values.

Step 4: Verify the Missing Data Pattern

Let’s examine the pattern of missing values we’ve created.

# Check missing data summary
na_summary <- car_matrix_na |>
  is.na() |>
  colSums()
print(na_summary)

# Calculate percentage missing per column
na_percentages <- round(na_summary / nrow(car_matrix_na) * 100, 1)
print(na_percentages)

This shows us exactly how many and what percentage of values are missing in each column.

Summary

  • Basic random replacement: Use sample() to select random positions and replace with NA for uniform missing data distribution
  • Targeted replacement: Apply conditions to create more realistic missing data patterns that reflect real-world scenarios
  • Multiple column approach: Introduce different missing data rates across columns to simulate complex data collection issues
  • Verification step: Always check the resulting pattern to ensure it matches your intended missing data structure

  • Reproducibility: Use set.seed() to make your random NA replacement reproducible for testing and validation