TopolearnReal data always has gaps and categories, and both behave in ways that surprise newcomers.
NA marks a missing value. It is contagious: any calculation involving NA gives NA, because the true answer is unknown. mean(c(1, NA, 3)) is NA; pass na.rm = TRUE to most summary functions to ignore missing values.
Test with is.na(x), never x == NA (which is NA itself). sum(is.na(x)) counts missing values and complete.cases(df) finds rows with no NA. There are typed versions (NA_integer_, NA_character_), and NaN (not a number, e.g. 0/0) and NULL (the absence of an object, length 0) are different things.
A factor stores categorical data: an integer vector of codes plus a levels attribute with the category names. Factors control the order of categories in tables, plots and model terms, and models create dummy variables for them automatically.
factor(x, levels = c("low", "mid", "high"))sets the order explicitly; otherwise levels are sorted alphabetically.table(f)counts each level, including levels with zero observations.droplevels()removes unused levels after filtering.
Since R 4.0, data.frame() and read.csv() keep strings as character by default (stringsAsFactors = FALSE), so you create factors deliberately. Convert a numeric-looking factor with as.numeric(as.character(f)); as.numeric(f) returns the internal codes.
Example
x <- c(4, NA, 10)
print(mean(x)) #> [1] NA
print(mean(x, na.rm = TRUE)) #> [1] 7
print(x == NA) #> [1] NA NA NA
print(is.na(x)) #> [1] FALSE TRUE FALSE
size <- factor(c("mid", "low", "mid", "high"),
levels = c("low", "mid", "high"))
print(levels(size)) #> [1] "low" "mid" "high"
print(table(size))
#> size
#> low mid high
#> 1 2 1
print(as.integer(size)) #> [1] 2 1 2 3
f <- factor(c("10", "5", "10"))
print(as.numeric(f)) #> [1] 1 2 1 (codes!)
print(as.numeric(as.character(f))) #> [1] 10 5 10Remember
- NA propagates; use
na.rm = TRUEandis.na() - A factor = integer codes + ordered levels
- Set
levelsexplicitly to control order in tables, plots and models
Watch out for
- Converting a factor of numbers with as.numeric(), which returns level codes instead of values.