Chapter 6 NA, NULL, NaN, Inf

What You’ll Learn:

• The four “missing” value types in R
• Critical differences between NA, NULL, NaN, and Inf
• Common errors with missing data
• How to handle missing values properly
• Testing and detecting special values

Key Errors Covered: 15+ missing data errors

Difficulty: ⭐ Beginner to ⭐⭐ Intermediate

6.1 Introduction

R has four special “non-values” that trip up everyone:

NA    # Not Available
#> [1] NA
NULL  # Nothing
#> NULL
NaN   # Not a Number
#> [1] NaN
Inf   # Infinity
#> [1] Inf

They look similar but behave very differently, and confusing them causes endless errors.

💡 Key Insight: The Four Special Values

# NA - Missing data (most common)
ages <- c(25, 30, NA, 35)  # One age is missing

# NULL - Absence of value (empty)
result <- NULL  # No result yet

# NaN - Invalid math result
0 / 0  # Undefined
#> [1] NaN

# Inf - Infinite value
1 / 0  # Positive infinity
#> [1] Inf
-1 / 0 # Negative infinity
#> [1] -Inf

Think of it this way: - NA: “I don’t know what this is” - NULL: “There’s nothing here” - NaN: “This calculation doesn’t make sense” - Inf: “This is beyond measurement”

6.2 NA: Not Available

6.2.1 What is NA?

# NA is a placeholder for missing data
x <- c(1, 2, NA, 4, 5)
x
#> [1]  1  2 NA  4  5

# It has a type!
typeof(NA)              # logical
#> [1] "logical"
typeof(NA_real_)        # double
#> [1] "double"
typeof(NA_integer_)     # integer
#> [1] "integer"
typeof(NA_character_)   # character
#> [1] "character"

6.3 Error #1: missing values where TRUE/FALSE needed

⭐ BEGINNER 🔢 TYPE

6.3.1 The Error

x <- NA
if (x > 5) {
  print("Large")
}
#> Error in if (x > 5) {: missing value where TRUE/FALSE needed

🔴 ERROR

Error in if (x > 5) { : missing value where TRUE/FALSE needed

6.3.2 What It Means

if() needs TRUE or FALSE, but got NA. R won’t guess which branch to take.

6.3.3 Common Causes

6.3.3.1 Cause 1: Comparison with NA

x <- c(1, 2, NA, 4, 5)

# Comparisons with NA give NA
x > 3
#> [1] FALSE FALSE    NA  TRUE  TRUE

# Using in if()

if (x[3] > 3) {  # x[3] is NA
  print("yes")
}
#> Error in if (x[3] > 3) {: missing value where TRUE/FALSE needed

6.3.3.2 Cause 2: Missing Data in Condition

df <- data.frame(
  age = c(25, NA, 35),
  name = c("Alice", "Bob", "Charlie")
)

# This creates NA in condition
if (df$age[2] > 30) {
  print("Senior")
}
#> Error in if (df$age[2] > 30) {: missing value where TRUE/FALSE needed

6.3.3.3 Cause 3: Function Returns NA

get_value <- function() {
  return(NA)
}

if (get_value() == 10) {
  print("Ten!")
}
#> Error in if (get_value() == 10) {: missing value where TRUE/FALSE needed

6.3.4 Solutions

✅ SOLUTION 1: Test for NA First

x <- NA

# Check for NA before using
if (!is.na(x) && x > 5) {
  print("Large")
} else {
  print("Not large or is NA")
}
#> [1] "Not large or is NA"

# Or handle NA explicitly
if (is.na(x)) {
  print("Missing value")
} else if (x > 5) {
  print("Large")
} else {
  print("Small")
}
#> [1] "Missing value"

✅ SOLUTION 2: Use isTRUE()

x <- NA

# isTRUE() converts NA to FALSE
if (isTRUE(x > 5)) {
  print("Large")
}

# Equivalent to:
if (!is.na(x) && x > 5) {
  print("Large")
}

✅ SOLUTION 3: Use %in% for Comparisons

x <- NA

# == with NA gives NA
x == 5  # NA
#> [1] NA

# %in% handles NA better
x %in% c(5)  # FALSE
#> [1] FALSE

# But still need to check:
5 %in% c(NA, 5)  # TRUE
#> [1] TRUE
NA %in% c(NA, 5) # TRUE
#> [1] TRUE

⚠️ Common Pitfall: && vs &

x <- c(NA, 2, 3)

# Single & returns vector with NA
x > 1 & x < 5
#> [1]   NA TRUE TRUE

# Double && errors on vector

if (x > 1 && x < 5) {  # Error!
  print("yes")
}
#> Error in x > 1 && x < 5: 'length = 3' in coercion to 'logical(1)'

For if(): Use && but check for NA first

For vectorized ops: Use & and handle NAs appropriately

6.4 Error #2: missing values and NaN's not allowed

⭐ BEGINNER 🧮 MATH

6.4.1 The Error

x <- c(1, 2, NA, 4, 5)
var(x)  # Without na.rm
#> [1] NA

🔴 ERROR

Error in var(x) : missing values and NaN's not allowed if 'na.rm' is FALSE

6.4.2 What It Means

Some functions refuse to work with NA unless you explicitly tell them how to handle it.

6.4.3 Common Functions With This Error

x <- c(1, 2, NA, 4, 5)

# These error without na.rm:
var(x)
#> [1] NA
sd(x)
#> [1] NA
cov(x, x)
#> [1] NA
cor(x, x)
#> [1] NA

6.4.4 Solutions

✅ SOLUTION 1: Use na.rm = TRUE

x <- c(1, 2, NA, 4, 5)

# Remove NAs for calculation
var(x, na.rm = TRUE)
#> [1] 3.333333
sd(x, na.rm = TRUE)
#> [1] 1.825742
mean(x, na.rm = TRUE)
#> [1] 3
sum(x, na.rm = TRUE)
#> [1] 12

✅ SOLUTION 2: Remove NAs First

x <- c(1, 2, NA, 4, 5)

# Complete cases only
x_complete <- x[!is.na(x)]
var(x_complete)
#> [1] 3.333333

# Or use na.omit()
x_complete <- na.omit(x)
var(x_complete)
#> [1] 3.333333

✅ SOLUTION 3: Replace NAs

x <- c(1, 2, NA, 4, 5)

# Replace with mean
x_filled <- x
x_filled[is.na(x_filled)] <- mean(x, na.rm = TRUE)
var(x_filled)
#> [1] 2.5

# Replace with median (more robust)
x_filled <- x
x_filled[is.na(x_filled)] <- median(x, na.rm = TRUE)
var(x_filled)
#> [1] 2.5

6.5 Error #3: 'x' contains missing values

⭐ BEGINNER 🧮 MATH

6.5.1 The Error

x <- c(1, 2, NA, 4)
y <- c(2, 3, 4, 5)
t.test(x, y)
#> 
#>  Welch Two Sample t-test
#> 
#> data:  x and y
#> t = -1.0675, df = 3.9593, p-value = 0.3465
#> alternative hypothesis: true difference in means is not equal to 0
#> 95 percent confidence interval:
#>  -4.213398  1.880065
#> sample estimates:
#> mean of x mean of y 
#>  2.333333  3.500000

🔴 ERROR

Error in t.test.default(x, y) : 'x' contains missing values

6.5.2 What It Means

Statistical tests often require complete data and won’t work with NAs.

6.5.3 Common Functions

x <- c(1, 2, NA, 4, 5)
y <- c(2, 3, 4, 5, 6)

# These need complete data:
t.test(x, y)
#> 
#>  Welch Two Sample t-test
#> 
#> data:  x and y
#> t = -0.86603, df = 6.0472, p-value = 0.4195
#> alternative hypothesis: true difference in means is not equal to 0
#> 95 percent confidence interval:
#>  -3.820108  1.820108
#> sample estimates:
#> mean of x mean of y 
#>         3         4
wilcox.test(x, y)
#> Warning in wilcox.test.default(x, y): cannot compute exact p-value with ties
#> 
#>  Wilcoxon rank sum test with continuity correction
#> 
#> data:  x and y
#> W = 6.5, p-value = 0.4568
#> alternative hypothesis: true location shift is not equal to 0
chisq.test(x, y)
#> Warning in chisq.test(x, y): Chi-squared approximation may be incorrect
#> 
#>  Pearson's Chi-squared test
#> 
#> data:  x and y
#> X-squared = 12, df = 9, p-value = 0.2133

6.5.4 Solutions

✅ SOLUTIONS

1. Remove NAs from both vectors:

x <- c(1, 2, NA, 4, 5)
y <- c(2, 3, 4, 5, 6)

# Find complete cases
complete <- complete.cases(x, y)
x_clean <- x[complete]
y_clean <- y[complete]

t.test(x_clean, y_clean)
#> 
#>  Welch Two Sample t-test
#> 
#> data:  x_clean and y_clean
#> t = -0.7746, df = 6, p-value = 0.468
#> alternative hypothesis: true difference in means is not equal to 0
#> 95 percent confidence interval:
#>  -4.15895  2.15895
#> sample estimates:
#> mean of x mean of y 
#>         3         4

2. Use na.action:

# For functions that support it
df <- data.frame(x = c(1, 2, NA, 4), y = c(2, 3, 4, 5))
t.test(x ~ 1, data = df, na.action = na.omit)
#> 
#>  One Sample t-test
#> 
#> data:  x
#> t = 2.6458, df = 2, p-value = 0.1181
#> alternative hypothesis: true mean is not equal to 0
#> 95 percent confidence interval:
#>  -1.461250  6.127916
#> sample estimates:
#> mean of x 
#>  2.333333

3. Impute missing values (advanced):

library(mice)
imputed <- mice(data, m = 5, method = "pmm")

6.6 NULL: The Absence of Value

6.6.1 What is NULL?

# NULL means "nothing"
x <- NULL
x
#> NULL

# Length zero
length(NULL)
#> [1] 0

# NULL disappears in vectors
c(1, 2, NULL, 4)
#> [1] 1 2 4

# Different from NA
c(1, 2, NA, 4)
#> [1]  1  2 NA  4

6.7 Error #4: argument is of length zero

⭐⭐ INTERMEDIATE 📏 LENGTH

6.7.1 The Error

x <- NULL
if (x > 5) {
  print("yes")
}
#> Error in if (x > 5) {: argument is of length zero

🔴 ERROR

Error in if (x > 5) { : argument is of length zero

6.7.2 What It Means

You’re trying to use NULL in a context that needs at least one value.

6.7.3 Common Causes

6.7.3.1 Cause 1: Uninitialized Variable

# Forgot to assign
result <- NULL

# Try to use
if (result == 100) {
  print("success")
}
#> Error in if (result == 100) {: argument is of length zero

6.7.3.2 Cause 2: Empty Subset

df <- data.frame(x = 1:5, y = 6:10)

# Filter returns nothing
subset_val <- df$x[df$x > 10]  # numeric(0)
subset_val <- NULL  # Or explicitly NULL

# Try to use
if (subset_val > 0) {
  print("positive")
}
#> Error in if (subset_val > 0) {: argument is of length zero

6.7.3.3 Cause 3: Function Returns NULL

get_value <- function(x) {
  if (x > 10) {
    return(x * 2)
  }
  # Implicitly returns NULL when x <= 10
}

result <- get_value(5)  # NULL

if (result > 0) {
  print("positive")
}
#> Error in if (result > 0) {: argument is of length zero

6.7.4 Solutions

✅ SOLUTION 1: Check for NULL

x <- NULL

# Check before using
if (!is.null(x) && x > 5) {
  print("Large")
} else {
  print("NULL or not large")
}
#> [1] "NULL or not large"

# Or use default
if (is.null(x)) {
  x <- 0  # Default value
}

if (x > 5) {
  print("Large")
}

✅ SOLUTION 2: Check Length

x <- NULL

# Check length instead
if (length(x) > 0 && x > 5) {
  print("Large")
}

# Works for both NULL and empty vectors
y <- numeric(0)
if (length(y) > 0 && y[1] > 5) {
  print("Large")
}

✅ SOLUTION 3: Use Default Values

get_value_safe <- function(x) {
  if (x > 10) {
    return(x * 2)
  } else {
    return(0)  # Default instead of NULL
  }
}

result <- get_value_safe(5)
if (result > 0) {
  print("positive")
}

6.8 NaN: Not a Number

6.8.1 What is NaN?

# Result of invalid math
0 / 0
#> [1] NaN
Inf - Inf
#> [1] NaN
Inf / Inf
#> [1] NaN

# NaN is a special type of NA
is.na(NaN)   # TRUE
#> [1] TRUE
is.nan(NA)   # FALSE
#> [1] FALSE

6.9 Error #5: NaNs produced

⭐ BEGINNER 🧮 MATH

6.9.1 The Warning

sqrt(-1)
#> Warning in sqrt(-1): NaNs produced
#> [1] NaN
log(-1)
#> Warning in log(-1): NaNs produced
#> [1] NaN
0/0
#> [1] NaN

🟡 WARNING

Warning message:
In sqrt(-1) : NaNs produced

6.9.2 What It Means

You performed a mathematically invalid operation. Result is NaN.

6.9.3 Common Causes

6.9.3.1 Cause 1: Negative Square Root

# Real square root of negative number doesn't exist
sqrt(-1)
#> Warning in sqrt(-1): NaNs produced
#> [1] NaN
sqrt(c(1, 4, -9, 16))
#> Warning in sqrt(c(1, 4, -9, 16)): NaNs produced
#> [1]   1   2 NaN   4

6.9.3.2 Cause 2: Log of Negative/Zero

# Log of negative or zero
log(0)   # -Inf
#> [1] -Inf
log(-1)  # NaN
#> Warning in log(-1): NaNs produced
#> [1] NaN

# Log of vector with negatives
values <- c(1, 10, -5, 100)
log(values)
#> Warning in log(values): NaNs produced
#> [1] 0.000000 2.302585      NaN 4.605170

6.9.3.3 Cause 3: Invalid Division

# 0/0 is undefined
0 / 0
#> [1] NaN

# Inf operations
Inf - Inf
#> [1] NaN
Inf / Inf
#> [1] NaN

6.9.4 Solutions

✅ SOLUTION 1: Check Input Before Operation

safe_sqrt <- function(x) {
  if (any(x < 0, na.rm = TRUE)) {
    warning("Negative values found, returning NA for those")
  }
  result <- sqrt(x)
  return(result)
}

safe_sqrt(c(1, 4, -9, 16))
#> Warning in safe_sqrt(c(1, 4, -9, 16)): Negative values found, returning NA for
#> those
#> Warning in sqrt(x): NaNs produced
#> [1]   1   2 NaN   4

✅ SOLUTION 2: Handle NaN After Operation

x <- c(1, -4, 9, -16)
result <- sqrt(x)
#> Warning in sqrt(x): NaNs produced

# Replace NaN with NA (more standard)
result[is.nan(result)] <- NA
result
#> [1]  1 NA  3 NA

# Or with a specific value
result <- sqrt(x)
#> Warning in sqrt(x): NaNs produced
result[is.nan(result)] <- 0
result
#> [1] 1 0 3 0

✅ SOLUTION 3: Use Complex Numbers for Sqrt

# For genuinely needing complex results
x <- -1
sqrt(as.complex(x))  # 0+1i
#> [1] 0+1i

# Vector
x <- c(1, -4, 9, -16)
sqrt(as.complex(x))
#> [1] 1+0i 0+2i 3+0i 0+4i

6.10 Inf: Infinity

6.10.1 What is Inf?

# Division by zero (non-zero)
1 / 0   # Inf
#> [1] Inf
-1 / 0  # -Inf
#> [1] -Inf

# Very large calculations
exp(1000)  # Inf
#> [1] Inf

# Inf in comparisons
Inf > 1e100  # TRUE
#> [1] TRUE
-Inf < -1e100  # TRUE
#> [1] TRUE

6.11 Error #6: infinite or missing values in 'x'

⭐⭐ INTERMEDIATE 🧮 MATH

6.11.1 The Error

x <- c(1, 2, Inf, 4)
y <- c(2, 3, 4, 5)
cor(x, y)
#> [1] NaN

🔴 ERROR

Error in cor(x, y) : infinite or missing values in 'x'

6.11.2 What It Means

Some functions can’t handle infinite values and need finite numbers.

6.11.3 Solutions

✅ SOLUTIONS

1. Check for and remove Inf:

x <- c(1, 2, Inf, 4, -Inf, 5)

# Find finite values
is.finite(x)
#> [1]  TRUE  TRUE FALSE  TRUE FALSE  TRUE

# Keep only finite
x_finite <- x[is.finite(x)]
y_finite <- y[is.finite(x)]
cor(x_finite, y_finite)
#> [1] NA

2. Replace Inf with large number:

x <- c(1, 2, Inf, 4, -Inf, 5)

# Replace Inf with max/min of finite values
x_fixed <- x
x_fixed[x == Inf] <- max(x[is.finite(x)]) * 10
x_fixed[x == -Inf] <- min(x[is.finite(x)]) * 10
x_fixed
#> [1]  1  2 50  4 10  5

3. Check before calculation:

safe_cor <- function(x, y) {
  finite_both <- is.finite(x) & is.finite(y)
  
  if (sum(finite_both) < 2) {
    stop("Need at least 2 finite pairs")
  }
  
  cor(x[finite_both], y[finite_both])
}

6.12 Testing for Special Values

🎯 Best Practice: Comprehensive Testing

# Test functions
x <- c(1, NA, NaN, Inf, -Inf, 0)

# Individual tests
is.na(x)       # TRUE for NA and NaN
#> [1] FALSE  TRUE  TRUE FALSE FALSE FALSE
is.nan(x)      # TRUE only for NaN
#> [1] FALSE FALSE  TRUE FALSE FALSE FALSE
is.infinite(x) # TRUE for Inf and -Inf
#> [1] FALSE FALSE FALSE  TRUE  TRUE FALSE
is.finite(x)   # TRUE for normal numbers
#> [1]  TRUE FALSE FALSE FALSE FALSE  TRUE

# NULL is different
y <- NULL
is.null(y)     # TRUE
#> [1] TRUE
is.na(y)       # logical(0) - no value to test
#> logical(0)

# Combined checks
is_valid <- function(x) {
  !is.na(x) & !is.nan(x) & is.finite(x)
}

is_valid(x)
#> [1]  TRUE FALSE FALSE FALSE FALSE  TRUE

Decision Tree:

if (is.null(x)) {
  # Handle NULL - no value at all
} else if (is.nan(x)) {
  # Handle NaN - invalid math result
} else if (is.infinite(x)) {
  # Handle Inf/-Inf - beyond limits
} else if (is.na(x)) {
  # Handle NA - missing data
} else {
  # Normal value
}

6.13 Handling Missing Data Strategies

💡 Key Insight: Missing Data Strategies

1. Complete Case Analysis (Listwise Deletion)

df <- data.frame(
  x = c(1, 2, NA, 4, 5),
  y = c(10, NA, 30, 40, 50)
)

# Keep only complete rows
complete.cases(df)
#> [1]  TRUE FALSE FALSE  TRUE  TRUE
df_complete <- df[complete.cases(df), ]
df_complete
#>   x  y
#> 1 1 10
#> 4 4 40
#> 5 5 50

2. Available Case Analysis (Pairwise Deletion)

# Use all available data for each calculation
cor(df, use = "pairwise.complete.obs")
#>   x y
#> x 1 1
#> y 1 1

3. Imputation (Replacing with Estimates)

# Replace with mean (simple)
df$x[is.na(df$x)] <- mean(df$x, na.rm = TRUE)

# Replace with median (robust to outliers)
df$y[is.na(df$y)] <- median(df$y, na.rm = TRUE)

4. Keep as NA (Most Honest)

# Just handle NAs in analysis
mean(df$x, na.rm = TRUE)
#> [1] 3

When to use each: - Complete case: When data missing completely at random (MCAR) - Pairwise: When you want to use all available information - Imputation: When you have good reason to estimate missing values - Keep NA: When missingness is informative

6.14 Summary

Key Takeaways:

1. Four special values: NA (missing), NULL (nothing), NaN (invalid math), Inf (infinite)
2. NA vs NULL: NA is a placeholder in a vector, NULL is absence of vector
3. Test before using: Always check is.na(), is.null(), is.finite()
4. Use isTRUE(): For conditions that might be NA
5. na.rm = TRUE: Most statistical functions need this with NAs
6. NaN from invalid math: sqrt(-1), 0/0, etc.
7. Inf from overflow: 1/0, exp(1000), etc.

Quick Reference:

Value	Test	Meaning	Example
NA	is.na()	Missing data	Survey non-response
NULL	is.null()	No value	Uninitialized variable
NaN	is.nan()	Invalid math	0/0
Inf	is.infinite()	Infinite	1/0

In if() statements:

# ❌ Dangerous
if (x > 5) { }  # Errors if x is NA or NULL

# ✅ Safe
if (!is.na(x) && x > 5) { }
if (isTRUE(x > 5)) { }
if (length(x) > 0 && !is.na(x) && x > 5) { }

With functions:

# ❌ May error
mean(x)
var(x)
cor(x, y)

# ✅ Handle missing
mean(x, na.rm = TRUE)
var(x[is.finite(x)])
cor(x, y, use = "complete.obs")

6.15 Exercises

📝 Exercise 1: Identify the Type

What are these and why?

# A
x <- c()
typeof(x)

# B  
y <- sqrt(-1)
class(y)

# C
z <- 1/0
is.finite(z)

# D
w <- c(1, 2, NA, 4)
w == NA

📝 Exercise 2: Fix the Code

Debug these:

# Problem 1
data <- c(1, 2, NA, 4, 5)
if (mean(data) > 3) {
  print("High average")
}

# Problem 2
get_score <- function(x) {
  if (x > 10) {
    return(x * 2)
  }
}
score <- get_score(5)
if (score > 10) {
  print("High score")
}

# Problem 3
values <- c(10, 20, 0, 30)
log_values <- log(values)
mean(log_values)

📝 Exercise 3: Robust Function

Write robust_mean(x) that: 1. Handles NA, NULL, NaN, Inf 2. Reports how many of each were found 3. Calculates mean of valid values 4. Returns list with mean and diagnostics

📝 Exercise 4: Data Cleaning

You have survey data:

survey <- data.frame(
  age = c(25, NA, 35, -999, 40),  # -999 = missing
  income = c(50000, 75000, 0, 80000, NA),  # 0 = refused
  satisfaction = c(5, 3, NA, 4, 99)  # 99 = invalid
)

Clean it: 1. Convert -999 to NA 2. Convert 0 in income to NA 3. Convert 99 in satisfaction to NA 4. Calculate complete case statistics

6.16 Exercise Answers

Click to see answers

Exercise 1:

# A - Empty vector
x <- c()
typeof(x)  # "logical" (default empty type)
#> [1] "NULL"

# B - NaN from invalid math
y <- sqrt(-1)
#> Warning in sqrt(-1): NaNs produced
class(y)  # "numeric"
#> [1] "numeric"
is.nan(y)  # TRUE
#> [1] TRUE

# C - Inf from division by zero
z <- 1/0
is.finite(z)  # FALSE (Inf is not finite)
#> [1] FALSE

# D - Comparison with NA gives NA
w <- c(1, 2, NA, 4)
w == NA  # All NA! Use is.na() instead
#> [1] NA NA NA NA
is.na(w)  # Correct way
#> [1] FALSE FALSE  TRUE FALSE

Exercise 2:

# Problem 1 - Need na.rm
data <- c(1, 2, NA, 4, 5)
mean_val <- mean(data, na.rm = TRUE)
if (!is.na(mean_val) && mean_val > 3) {
  print("High average")
}

# Problem 2 - Check for NULL
get_score <- function(x) {
  if (x > 10) {
    return(x * 2)
  } else {
    return(0)  # Return 0 instead of NULL
  }
}
score <- get_score(5)
if (!is.null(score) && score > 10) {
  print("High score")
}

# Problem 3 - Handle -Inf from log(0)
values <- c(10, 20, 0, 30)
log_values <- log(values)
log_values[is.infinite(log_values)] <- NA
mean(log_values, na.rm = TRUE)
#> [1] 2.899838

Exercise 3:

robust_mean <- function(x) {
  # Initialize diagnostics
  diagnostics <- list(
    total = length(x),
    null = is.null(x),
    na = 0,
    nan = 0,
    inf = 0,
    valid = 0
  )
  
  # Check for NULL
  if (is.null(x)) {
    return(list(mean = NULL, diagnostics = diagnostics))
  }
  
  # Count special values
  diagnostics$na <- sum(is.na(x) & !is.nan(x))
  diagnostics$nan <- sum(is.nan(x))
  diagnostics$inf <- sum(is.infinite(x))
  
  # Find valid values
  valid <- x[!is.na(x) & !is.nan(x) & is.finite(x)]
  diagnostics$valid <- length(valid)
  
  # Calculate mean
  if (length(valid) == 0) {
    mean_val <- NA
  } else {
    mean_val <- mean(valid)
  }
  
  # Return
  list(
    mean = mean_val,
    diagnostics = diagnostics
  )
}

# Test
robust_mean(c(1, 2, NA, NaN, Inf, 5, -Inf))
#> $mean
#> [1] 2.666667
#> 
#> $diagnostics
#> $diagnostics$total
#> [1] 7
#> 
#> $diagnostics$null
#> [1] FALSE
#> 
#> $diagnostics$na
#> [1] 1
#> 
#> $diagnostics$nan
#> [1] 1
#> 
#> $diagnostics$inf
#> [1] 2
#> 
#> $diagnostics$valid
#> [1] 3

Exercise 4:

survey <- data.frame(
  age = c(25, NA, 35, -999, 40),
  income = c(50000, 75000, 0, 80000, NA),
  satisfaction = c(5, 3, NA, 4, 99)
)

# Clean data
clean_survey <- survey

# Convert -999 to NA in age
clean_survey$age[clean_survey$age == -999] <- NA

# Convert 0 to NA in income
clean_survey$income[clean_survey$income == 0] <- NA

# Convert 99 to NA in satisfaction
clean_survey$satisfaction[clean_survey$satisfaction == 99] <- NA

# Complete case analysis
clean_survey_complete <- clean_survey[complete.cases(clean_survey), ]

# Statistics
list(
  n_complete = nrow(clean_survey_complete),
  mean_age = mean(clean_survey_complete$age),
  mean_income = mean(clean_survey_complete$income),
  mean_satisfaction = mean(clean_survey_complete$satisfaction)
)
#> $n_complete
#> [1] 1
#> 
#> $mean_age
#> [1] 25
#> 
#> $mean_income
#> [1] 50000
#> 
#> $mean_satisfaction
#> [1] 5