Geek Answers Handbook

How to repeat the Grubbs test and indicate emissions

I want to reapply the Grubbs test to a dataset until it stops detecting outliers. I want the outliers to be labeled rather than deleted so that I can display the data as a histogram with a different color. I used grubbs.test from the outliers package to manually identify outliers, but I can’t figure out how to get through them and successfully record them. The type of output I'm aiming for is as follows:

X Outlier 152.36 Yes 130.38 Yes 101.54 No 96.26 No 88.03 No 85.66 No 83.62 No 76.53 No 74.36 No 73.87 No 73.36 No 73.35 No 68.26 No 65.25 No 63.68 No 63.05 No 57.53 No
+6
source share
2 answers

It looks like you will need a short function to do what you want:

 library(outliers) library(ggplot2) X <- c(152.36,130.38,101.54,96.26,88.03,85.66,83.62,76.53, 74.36,73.87,73.36,73.35,68.26,65.25,63.68,63.05,57.53) grubbs.flag <- function(x) { outliers <- NULL test <- x grubbs.result <- grubbs.test(test) pv <- grubbs.result$p.value while(pv < 0.05) { outliers <- c(outliers,as.numeric(strsplit(grubbs.result$alternative," ")[[1]][3])) test <- x[!x %in% outliers] grubbs.result <- grubbs.test(test) pv <- grubbs.result$p.value } return(data.frame(X=x,Outlier=(x %in% outliers))) }

Here's the conclusion:

 grubbs.flag(X) X Outlier 1 152.36 TRUE 2 130.38 TRUE 3 101.54 FALSE 4 96.26 FALSE 5 88.03 FALSE 6 85.66 FALSE 7 83.62 FALSE 8 76.53 FALSE 9 74.36 FALSE 10 73.87 FALSE 11 73.36 FALSE 12 73.35 FALSE 13 68.26 FALSE 14 65.25 FALSE 15 63.68 FALSE 16 63.05 FALSE 17 57.53 FALSE

And if you need a histogram with different colors, you can use the following:

 ggplot(grubbs.flag(X),aes(x=X,color=Outlier,fill=Outlier))+ geom_histogram(binwidth=diff(range(X))/30)+ theme_bw()

Outlier histogram

+15
source

Sam Dickson's answer is excellent, but it will throw an error if you reach the point where all but two values ​​are marked as outliers or if you just started with three values ​​in the first place (grubbs.test () will not return a p-value if input vector there are only two values).

I have added a breakpoint for the while loop for this unforeseen situation, and it will also issue a warning if this happens. In addition, it will generate an informative error at startup with less than two input values.

 grubbs.flag <- function(x) { outliers <- NULL test <- x grubbs.result <- grubbs.test(test) pv <- grubbs.result$p.value # throw an error if there are too few values for the Grubb test if (length(test) < 3 ) stop("Grubb test requires > 2 input values") while(pv < 0.05) { outliers <- c(outliers,as.numeric(strsplit(grubbs.result$alternative," ")[[1]][3])) test <- x[!x %in% outliers] # stop if all but two values are flagged as outliers if (length(test) < 3 ) { warning("All but two values flagged as outliers") break } grubbs.result <- grubbs.test(test) pv <- grubbs.result$p.value } return(data.frame(X=x,Outlier=(x %in% outliers))) }

Of course, of course, it makes no sense to do outlier tests if you have only three data points, but I don’t know your business.

+8
source

More articles:


All Articles