Calculating percentages is a fairly common operation, right? However, doing it without leaving the pipeflow always force me to do some bizarre piping such as double grouping and summarise.

I am using again the nuclear accidents dataset, and trying to calculate the percentage of accidents that happened in Europe each year.

nuclear_accidents <- read_csv("https://query.data.world/s/47s7katrhxxd674ulus425k42l5u4s") 

nuclear_accidents <- nuclear_accidents %>% 
 select(-Description) %>% 
 mutate(Year = Date %>% mdy() %>% year(),
 In_Europe = if_else(Region %in% c("EE", "WE"), T, F) %>% as.factor()) %>% 
 filter(Year %>% between(1989, 2016))

nuclear_accidents %>% head(4) %>% kable()

This can be achieved by several ways. One common path would be:

nuclear_accidents %>% 
 group_by(Year, In_Europe) %>% 
 summarise(N = n()) %>% 
 group_by(Year) %>% 
 mutate(Total_per_year = sum(N), 
 Ratio = round(N/Total_per_year, 2)) %>% 
 head(4) %>% 
 kable()

Another one more bizarre would be totalizing first, then grouping including that amount (to avoid being dropped) and then summarise.

nuclear_accidents %>% 
 group_by(Year) %>% 
 mutate(Total_per_year = n()) %>% 
 group_by(Year, In_Europe, Total_per_year) %>% 
 summarise(N = n()) %>% 
 mutate(Ratio = round(N/Total_per_year, 2)) %>% 
 head(4) %>% 
 kable()

Kind of weird. However, there is a much simpler way:

nuclear_accidents %>% 
 group_by(Year, In_Europe) %>%
 summarize(N = n()) %>%
 mutate(Ratio = round(N / sum(N), 2)) %>% 
 head(4) %>% 
 kable()

The first time I saw this result I didn’t understand it because if you have your dataframe grouped by Year and In_Europe then sum(N) should be equal to N. What is going on? This behaviour has to do with a tricky funcionality of summarise.

Take a closer look of the grouping variables at the console output. Before the summarise function the dataframe seems grouped normally and the operation will be performed within each group:

nuclear_accidents %>% 
 group_by(Year, In_Europe) %>% 
 head(4)
# A tibble: 4 x 9
# Groups: Year, In_Europe [4]
# Date Location `Cost (millions ~ INES `Smyth Magnitud~ Region Fatalities Year In_Europe
# <chr> <chr> <dbl> <dbl> <dbl> <chr> <dbl> <dbl> <fct> 
# 1 3/11/~ Fukushima ~ 166089 7 7.5 A 573 2011 FALSE 
# 2 12/8/~ Tsuruga, J~ 15500 NA NA A 0 1995 FALSE 
# 3 12/19~ Vandellòs,~ 931. 3 NA WE 0 1989 TRUE 
# 4 2/1/2~ Vernon, Ve~ 809. NA NA NA 0 2010 FALSE 

However, once the dataframe is summarized, the resulting dataframe is no longer grouped by the same original variables:

nuclear_accidents %>% 
 group_by(Year, In_Europe) %>%
 summarize(N = n()) %>% 
 head(4)
# A tibble: 4 x 3
# Groups: Year [3]
# Year In_Europe N
# <dbl> <fct> <int>
# 1 1989 FALSE 4
# 2 1989 TRUE 2
# 3 1990 FALSE 1
# 4 1991 FALSE 3

Actually, the default behaviour of summarise is to drop the last group. The reason behind that is that, once the operation is performed you should have only one obervation per group, and it has no sense to grouping by it anymore. That’s why the last example I show above works. Now you can take advantage of it too!

Furthermore, you can learn more about the dplyr 1.0.0 last minute additions which include an explicit message to highlight the behaviour we have talked about here.