Missing Data in R Exercises: 10 NA Detection & Imputation Problems

Ten hands-on exercises drill NA detection, removal, and imputation in R — from basic is.na() checks to grouped median imputation — run every solution in your browser.

Introduction

Reading about NA handling is one thing. Applying it under pressure with unfamiliar column names and messy patterns is another. These ten problems close that gap. Each one targets a specific missing-data skill that trips up real analysts.

You will start by counting NAs, then move to filtering incomplete rows, and finish by imputing missing values with column-level and group-level strategies using base R and dplyr. By Exercise 10 you will chain detection, reporting, imputation, and verification into one pipeline.

If is.na(), complete.cases(), and na.omit() are new to you, read the parent Missing Values in R tutorial first. Otherwise, run the setup block and begin. All code runs in one shared R session. Use distinct variable names like ans1, ans2 in your own attempts so you do not overwrite the tutorial datasets.

Setup: The Datasets We Will Use

Before the exercises, we build three small tables with deliberate NA patterns. They are tiny on purpose so you can eyeball every row and verify your answers by hand. Run this block once. Every exercise after this assumes these objects exist.

The sales table has missing values in revenue and units. The survey table has NAs scattered across age, score, and group. The weather_data table has missing temp and rainfall readings.

Look at the NA positions carefully. sales has two NAs in revenue and two in units, but in different rows. survey has three NAs in age and two in score, never in the same row. weather_data has one NA in temp and two in rainfall.

Warm-Up: NA Detection (Exercises 1-3)

These three exercises focus on finding and counting NAs. If you get them right, you understand the detection fundamentals. Expected output values are given so you can verify.

Exercise 1: Count the total number of NAs in a data frame

Use is.na() and sum() to count the total number of NA values in the entire sales table. Save the result to ans1. Expected: 4.

Exercise 2: Count NAs per column

Use colSums() with is.na() to count the number of NA values in each column of survey. Save the result to ans2. Expected: id = 0, group = 0, age = 3, score = 2.

Exercise 3: Identify rows with any missing value

Use complete.cases() to find which rows in weather_data have no NAs. Save the complete rows to ans3. Expected: 2 rows (Oslo-Jan and Lima-Feb).

Core Challenges: Removal (Exercises 4-6)

Detection done. These three exercises remove rows based on different criteria. Pay attention to the difference between dropping all incomplete rows and targeting specific columns.

Exercise 4: Remove all rows with any NA

Use na.omit() to remove every row that contains at least one NA from sales. Save to ans4. Expected: 2 rows.

Exercise 5: Remove rows where a specific column is NA

Use filter() with !is.na() to keep only the rows of survey where score is not missing. Save to ans5. Expected: 6 rows.

Exercise 6: Keep rows complete in selected columns only

Use complete.cases() on just the temp and rainfall columns of weather_data to keep rows where both measurements exist. Save to ans6. Expected: 2 rows.

Advanced: Imputation (Exercises 7-10)

Removing rows wastes data. These four exercises replace NAs with computed values. You will impute with column means, column medians, group-level means, and finally build a full cleaning pipeline.

Exercise 7: Impute NA with column mean (base R)

Replace the NA values in sales$revenue with the mean of the non-missing revenue values. Use base R (no dplyr). Save the modified data frame to ans7. Expected: the two NA revenues become 457.5.

Exercise 8: Impute NA with column median using dplyr

Use mutate() with replace() to replace NA values in survey$age with the median age. Save to ans8. Expected: the three NA ages become 29 (the median of 22, 25, 29, 34, 41).

Exercise 9: Impute NA with group-wise mean

Use group_by() and mutate() to replace NA values in survey$score with the mean score of each respondent's group. Save to ans9. Expected: group A mean score is 77.25 and group B mean score is 82.

Exercise 10: Full cleaning pipeline

Build a complete cleaning pipeline for sales. In one dplyr chain: (1) add a column na_count that counts NAs per row, (2) impute revenue NAs with the column mean, (3) impute units NAs with the column median, (4) verify zero NAs remain. Save the final cleaned table to ans10. Expected: 6 rows, 5 columns, 0 NAs in revenue and units.

Common Mistakes and How to Fix Them

Mistake 1: Using == NA instead of is.na()

Bad:

Good:

Why it is wrong: NA == NA returns NA, not TRUE. The == operator propagates NA because R treats NA as "unknown." Two unknowns might or might not be equal, so the comparison is also unknown. Always use is.na() to test for missingness.

Mistake 2: Forgetting na.rm = TRUE in aggregation

Bad:

Good:

Why it is wrong: Without na.rm = TRUE, functions like mean(), sum(), sd(), and median() return NA if any input is NA. The default is strict on purpose so you notice the missingness. Add na.rm = TRUE when you have already decided that dropping NAs is appropriate for your analysis.

Mistake 3: Imputing before understanding the missingness pattern

Bad:

Why it is wrong: If 80% of a column is NA, the mean is computed from only 20% of the data and is unreliable. If the data is not missing at random (for example, high-income respondents skip the income question), mean imputation introduces bias. Always check the proportion and pattern of missingness first with colSums(is.na(df)) / nrow(df). Columns above 40-50% missing are often better dropped entirely.

Summary

Here is a quick reference for the skill each exercise tests.

FAQ

When should I remove rows versus impute missing values?

Remove rows when the proportion of NAs is small (under 5%) and the missing values are random. Impute when you have a moderate amount of missingness (5-30%) and removal would shrink your dataset too much. If a column has more than 40-50% NAs, consider dropping the column entirely rather than imputing unreliable values.

Does na.rm = TRUE change the original data?

No. The na.rm parameter only affects the current calculation. It tells mean(), sum(), or median() to ignore NAs while computing the result. The original vector or column is unchanged. To actually modify the data, you need assignment (<-) or mutate().

How do I check if imputation worked correctly?

Run sum(is.na(column)) after imputation to confirm zero NAs remain. Also compare the mean and standard deviation before and after imputation. If the standard deviation dropped noticeably, your imputation method may be collapsing the variance (a known issue with mean imputation).

What if an entire group has all NAs?

Group-wise imputation returns NaN (Not a Number) because mean(c(NA, NA), na.rm = TRUE) is NaN. Your column then has NaN instead of NA, which can cause different errors downstream. Guard against this by checking group sizes before imputing or by falling back to the overall column mean when a group mean is NaN.

References

1. R Core Team — is.na() documentation. Link
2. R Core Team — complete.cases() documentation. Link
3. Wickham, H. & Grolemund, G. — R for Data Science, 2nd Edition. Chapter 18: Missing Values. Link
4. UC Business Analytics R Programming Guide — Dealing with Missing Values. Link
5. dplyr documentation — mutate() reference. Link
6. The Epidemiologist R Handbook — Missing Data chapter. Link
7. Tierney, N. — naniar: Data Structures, Summaries, and Visualisations for Missing Data. Link

What's Next?

Now that you can detect, remove, and impute missing values, explore these related tutorials:

• Missing Values in R: Detect, Count, Remove & Impute NA — review the parent tutorial if any exercise stumped you
• dplyr Exercises (15 problems) — practise filtering, grouping, and joining on clean data
• tidyr Reshaping Exercises (10 problems) — reshape data after cleaning it