performance (version 0.3.0)

check_outliers: Check for influential observations

Description

Checks for and locates influential observations (i.e., "outliers") via several distance methods.

Usage

check_outliers(x, ...)
# S3 method for default
check_outliers(x, method = c("cook", "mahalanobis",
  "ics"), threshold = NULL, ...)

Arguments

x

A model object.

...

When method = "ics", further arguments in ... are passed down to ICSOutlier::ics.outlier().

method

The method to calculate the distance, at which value a point is considered as "outlier".

threshold

The threshold indicating at which distance an observation is considered as outlier. Possible values are "cook" (Cook's Distance), "mahalanobis" (Mahalanobis Distance) or "ics" (Invariant Coordinate Selection, Archimbaud et al. 2018). May be abbreviated. See 'Details'.

Value

Check (message) on whether outliers were detected or not, as well as a data frame (with the original data that was used to fit the model), including information on the distance measure and whether or not an observation is considered as outlier.

Details

Performs a distance test to check for influential observations. Those greater as a certain threshold, are considered outliers. This relatively conservative threshold is useful only for detection, rather than justificaiton for automatic observation deletion.

Choosing Default Thresholds

Cook's Distance

When method = "cook", threshold defaults to 4 divided by numbers of observations.

Mahalanobis Distance

When method = "mahalanobis", the default for threshold is based on a weird formula (floor(3 * sqrt(sum(cov(predictors)^2)) / nobs(x))), which is limted to values between 3 and 10, to account for different variation in the data depending on the number of observations. There is no "rule of thumb" for the threshold regarding the Mahalanobis Distance, most studies use a value between 3 and 10. It is most likely better to define own, sensible thresholds.

Invariant Coordinate Selection

If method = "ics", the threshold is determined by ICSOutlier::ics.outlier(). Refer to the help-file of that function to get more details about this procedure. Note that method = "ics" requires both ICS and ICSOutlier to be installed, and that it takes a bit longer to compute the results.

References

• Cook, R. D. (1977). Detection of influential observation in linear regression. Technometrics, 19(1), 15-18.

• Archimbaud, A., Nordhausen, K., & Ruiz-Gazen, A. (2018). ICS for multivariate outlier detection with application to quality control. Computational Statistics & Data Analysis, 128, 184<U+2013>199. 10.1016/j.csda.2018.06.011

Examples

# NOT RUN {
# select only mpg and disp (continuous)
mt1 <- mtcars[, c(1, 3, 4)]
# create some fake outliers and attach outliers to main df
mt2 <- rbind(mt1, data.frame(mpg = c(37, 40), disp = c(300, 400), hp = c(110, 120)))
# fit model with outliers
model <- lm(disp ~ mpg + hp, data = mt2)

check_outliers(model)
plot(check_outliers(model))

check_outliers(model, method = "m")

# }
# NOT RUN {
# This one takes some seconds to finish...
check_outliers(model, method = "ics")
# }
# NOT RUN {
# }