How To Find Outliers in Boxplots [via R Programming]
Finding outliers in Boxplots via Geom_Boxplot in R Studio
In the first boxplot that I created using GA data, it had ggplot2 + geom_boxplot to show google analytics data summarized by day of week.
While the min/max, median, 50% of values being within the boxes [inter quartile range] were easier to visualize/understand, these two dots stood out in the boxplot.
If the whiskers from the box edges describes the min/max values, what are these two dots doing in the geom_boxplot?
Using a dplyr package to provide summary stats for geom_boxplot
After asking around, I found out a dplyr package that could provide summary stats for the boxplot [while I still haven't figured out how to add the data labels to the boxplot, the summary table seems like a good start].
This bit of the code creates a summary table that provides the min/max and inter-quartile range.
gadata %>% group_by(as.factor(dayofWeek)) %>% summarise(Min = min(sessions), Max = max(sessions), Median = median(sessions), IQRange = IQR(sessions))
If you download the Xlsx dataset and then filter out the values where dayofWeek =0, we get the below values:
3, 5, 6, 10, 10, 10, 10, 11,12, 14, 14, 15, 16, 20
N = 14
Central values = 10, 11 [50% of values are above/below these numbers]
Median = (10+11)/2 or 10.5 [matches with the table above]
Smallest value = 3
Largest value = 20
Lower Quartile Value [Q1]: = (7+1)/2 = 4th value [below median range]= 10
Upper Quartile Value [Q3]: (7+1)/2 = 4th value [above median range] = 14
Inter Quartile Range = 14-10 = 4
If we want to know whether the first value  is an outlier here,
Lower outlier limit = Q1 - 1.5 * IQR = 10 - 1.5 *4
Lower outlier limit = 4
Upper outlier limit = Q3 + 1.5 *IQR = 14 + 1.5*4
Upper outlier limit = 20
As 3 is below the outlier limit, the min whisker starts at the next value ,
As all the max value is 20, the whisker reaches 20 and doesn't have any data value above this point.
And there's the geom_boxplot explained. Kinda cool it does all of this automatically!
Hope you found this blog post helpful
#run the GA library library(googleAnalyticsR) #authentication token ga_auth() #setting up the query gadata <- google_analytics(viewId = 55290473, date_range = c(Sys.Date()-100, Sys.Date()-1), metrics = c("users", "sessions", "pageviews"), dimensions = c("date", "dayofWeek"), anti_sample = TRUE) #examining the first 6 rows as a preview head(gadata) #looking at the structure of the data str(gadata) #creating boxplot boxplotchart <- ggplot(gadata, aes(x=dayofWeek, y = sessions)) + geom_boxplot() boxplotchart + ylab("Sessions on that day") + xlab("Day of Week; 0 = Sun, 6 = Sat") boxplotchart <- ggplot(gadata, aes(x = as.factor(dayofWeek), y = sessions)) + geom_boxplot() boxplotchart #table of boxplot data with summary stats gadata %>% group_by(as.factor(dayofWeek)) %>% summarise(Min = min(sessions), Max = max(sessions), Median = median(sessions), IQRange = IQR(sessions)) install.packages("xlsx") library("xlsx") write.xlsx(gadata, file = "C:\\Users\\KhanAd\\Dropbox\\blog content\\2018\\052018\\20180526 Day of week boxplot with outlier.xlsx", sheetName = "Day Of Week", append = FALSE)