What are levels in R?

Question

I understand this is a very basic question but I don't understand what levels mean in R.

For reference, I have done a simple script to read CSV table, filter on one of the fields, pass this on to a new variable and clear the memory allocated for the first variable. If I call unique() on the field on which I filtered, I see that the results were indeed filtered but there is one additional line showing 'Levels' corresponding to data that is in the original dataset.

Example:

df = read.csv(path, sep=",", header=TRUE)
df_intrate = df[df$AssetClass == "ASSET CLASS A", ]

rm(df)
gc()

unique(df_intrate$AssetClass)

Results:

[1] ASSET CLASS A
Levels: ASSET CLASS E ASSET CLASS D ASSET CLASS C ASSET CLASS B ASSET CLASS A

Is the structural information from df somehow preserved in df_intrate despite R studio showing that df_intrate is indeed the expected number of rows for ASSET CLASS A ?

Answer 1

Is the structural information from df somehow preserved in df_intrate despite R studio showing that df_intrate is indeed the expected number of rows for ASSET CLASS A ?

Yes. This is how categorical variables, called factors, are stored in R - both the levels, a vector of all possible values, and the actual values taken, are stored:

x = factor(c('a', 'b', 'c', 'a', 'b', 'b'))
x
# [1] a b c a b b
# Levels: a b c

y = x[1]
# [1] a
# Levels: a b c

You can get rid of unused levels with droplevels(), or by re-applying the factor function, creating a new factor out of only what is present:

droplevels(y)
# [1] a
# Levels: a

factor(y)
# [1] a
# Levels: a

You can also use droplevels on a data frame to drop all unused levels from all factor columns:

dat = data.frame(x = x)
str(dat)
# 'data.frame': 6 obs. of  1 variable:
#  $ x: Factor w/ 3 levels "a","b","c": 1 2 3 1 2 2

str(dat[1, ])
# Factor w/ 3 levels "a","b","c": 1

str(droplevels(dat[1, ]))
# Factor w/ 1 level "a": 1

---

Though unrelated to your current issue, we should also mention that factor has an optional levels argument which can be used to specify the levels of a factor and the order in which they should go. This can be useful if you want a specific order (perhaps for plotting or modeling), or if there are more possible levels than are actually present and you want to include them. If you don't specify the levels, the default will be alphabetical order.

x = c("agree", "disagree", "agree", "neutral", "strongly agree")
factor(x)
# [1] agree         disagree      agree         neutral       strongly agree
# Levels: agree disagree neutral strongly agree
## not a good order

factor(x, levels = c("disagree", "neutral", "agree", "strongly agree"))
# [1] agree          disagree       agree          neutral        strongly agree
# Levels: disagree neutral agree strongly agree
## better order

factor(x, levels = c("strongly disagree", "disagree", "neutral", "agree", "strongly agree"))
# [1] agree          disagree       agree          neutral        strongly agree
# Levels: strongly disagree disagree neutral agree strongly agree
## good order, more levels than are actually present

You can use ?reorder and ?relevel (or just factor again) to change the order of levels for an already created factor.