utils_stats: Summary statistics utilities
In phytomosaic/ecole: ecole: School of Ecology Package

Description Usage Arguments Details Value See Also Examples

Vectorized summary statistics, including geometric mean, harmonic mean, sample standard error (SE), coefficient of variation (CV), root mean square error (RMSE), mean absolute error (MAE), sensitivity, and robust z-scores.

geom_mean(x, na.rm = TRUE, zero.rm = FALSE, ...)

harm_mean(x, na.rm = TRUE, zero.rm = FALSE, ...)

sem(x, na.rm = TRUE, ...)

cv(x, na.rm = TRUE, ...)

rmse(x, y, na.rm = TRUE, ...)

mae(x, y, stdz = FALSE, na.rm = TRUE, ...)

zscr(x, robust = TRUE, na.rm = TRUE, ...)

`x`	vector of values to evaluate
`na.rm`	logical, should NA values in `x` be removed before calculating?
`zero.rm`	logical, should zeros in `x` be removed before calculating harmonic or geometric means?
`...`	further arguments passed to other methods
`y`	vector of 'predicted' values to compare against `x`
`stdz`	logical, standardize output by range of `x`?
`robust`	logical, should robust z-scores be calculated?

For vectors including at least one zero, results of geom_mean and harm_mean are always 0 by definition, unless zero.rm=TRUE.

Like sd, sem uses n-1 in denominator to correct for small-sample bias.

rmse is one way to assess prediction accuracy.

mae gives a measure of sensitivity when stdz=TRUE.

zcsr gives robust z-scores based on median (not mean) and median absolute deviation (not standard deviation).

These functions return NA when NAs present and na.rm=TRUE.

Numeric values.

sd

# test data
xx <- c(-1, 0, 1, 4, 77, NA)

# harmonic mean
harm_mean(xx, na.rm=TRUE, zero.rm=FALSE)     # 0 by definition
harm_mean(xx, na.rm=TRUE, zero.rm=TRUE)      # 15.20988

# geometric mean
### NOT RUN:
# geom_mean(xx, na.rm=TRUE, zero.rm=FALSE))  # fails for neg vals
### END NOT RUN
xx <- xx[-1]                                 # remove negative values
geom_mean(xx, na.rm=TRUE, zero.rm=FALSE)     # 0 by definition
geom_mean(xx, na.rm=TRUE, zero.rm=TRUE)      # 6.753313

# standard error of the mean
sem(xx)                       # 21.76899

# coefficient of variation
cv(xx)                        # 183.9268

# root mean squared error
set.seed(23)
xx <- c(-1, 0, 1, 4, 77, NA)
yy <- xx+rnorm(length(xx), 10)
rmse(xx, yy)                  # 10.71919
rmse(yy, xx)                  # same, order invariant

# mean absolute error
mae(xx, yy, stdz=FALSE)       # 10.69236

# range-standardized mean absolute error (aka sensitivity)
mae(xx, yy, stdz=TRUE)        # 0.1370815
mae(yy, xx, stdz=TRUE)        # 0.135684 -- order matters!

# robust z-scores not so influenced by extreme values
x <- c(-99, -9, 0, 9, 99)
plot(zscr(x, robust=FALSE), zscr(x, robust=TRUE), asp=1)
abline(0,1)