Mode (most Frequent Value(s))

Calculate the mode, the most frequent value, of a numeric or character vector x.

Usage

modeX(x, na.rm = FALSE)

Arguments

x

a non-empty numeric vector of data values

na.rm

logical. Should missing values be removed? Defaults to FALSE.

Value

the most frequent value as a number or character, depending on class(x). If there is more than one, all are returned in a vector.
The modal frequency is attached as attribute named "freq".

Details

The mode is mainly useful for qualitative data, sometimes still for integer vectors.

For numeric vectors, the interest lies less in central tendency than in conspicuous accumulation points, which can indicate data errors. desc() therefore reports it within the numeric description once the frequency of a single value exceeds a threshold, since ties are generally unexpected in numeric data unless the generating process explains them.

NA handling follows the package standard: a single NA yields NA. This is conservative, as the mode is sometimes determined unambiguously despite missing values. For x = c(1,1,1,1,2,2,NA) the mode is 1 whatever the missing value is, and the modal frequency lies between 4 and 5. Exploiting this is left to the user.

The mode is elsewhere often obtained by tabulating every element and returning the most frequent. This function uses a dedicated C++ data structure and determines only the most frequent element, making it orders of magnitude faster, especially for large numeric vectors with many distinct values.

Note: There are other approaches for determining the mode, e.g. one might use
density(x)$x[which.max(density(x)$y)]
for quantitative data, resp. hist().
Another interesting idea for a more robust estimation of the mode:

 peak <- optimize(function(x, model)
  predict(model, data.frame(x = x)),
    c(min(x), max(x)), maximum = TRUE, model = y.loess)
  points(peak$maximum, peak$objective)

Note

Great Rcpp part contributed by Joseph Wood and Ralf Stubner.

References

rcpp-fast-statistical-mode

See also

Other location: gmean(), hmean(), hodgesLehmann(), huberM(), meanX(), medianX(), tukeyBiweight()

Examples

# normal mode
modeX(c(0:5, 5))
#> [1] 5
#> attr(,"freq")
#> [1] 2

modeX(5)
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(NA)
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(c(NA, NA))
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(c(NA, 0:5))
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(c(NA, 0:5), na.rm=TRUE)
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(c(NA, 0:5, 5), na.rm=TRUE)
#> [1] 5
#> attr(,"freq")
#> [1] 2

# returns all encountered modes, if several exist
modeX(c(0:5, 4, 5, 6))
#> [1] 4 5
#> attr(,"freq")
#> [1] 2

modeX(Pizza$driver)
#> [1] NA
#> attr(,"freq")
#> [1] NA
modeX(Pizza$driver, na.rm=TRUE)
#> [1] Carpenter
#> attr(,"freq")
#> [1] 272
#> Levels: Butcher Carpenter Carter Farmer Hunter Miller Taylor
modeX(as.character(Pizza$driver), na.rm=TRUE)
#> [1] "Carpenter"
#> attr(,"freq")
#> [1] 272

# use sapply for evaluating data.frames (resp. apply for matrices)
sapply(Pizza[,c("driver", "temperature", "date")], modeX, na.rm=TRUE)
#>      driver temperature        date 
#>         2.0        51.3     16137.0