How to group, pivot and count in R - r

Question

Using the GermanCredit dataset from the caret library.
library("caret")
data(GermanCredit)
After filtering this down a bit
credit.all <- GermanCredit[,c(10, 1:9, 11:13, 16:19)]
attach(credit.all)
names(credit.all)
We have these names
[1] "Class" "Duration"
[3] "Amount" "InstallmentRatePercentage"
[5] "ResidenceDuration" "Age"
[7] "NumberExistingCredits" "NumberPeopleMaintenance"
[9] "Telephone" "ForeignWorker"
[11] "CheckingAccountStatus.lt.0" "CheckingAccountStatus.0.to.200"
[13] "CheckingAccountStatus.gt.200" "CreditHistory.ThisBank.AllPaid"
[15] "CreditHistory.PaidDuly" "CreditHistory.Delay"
[17] "CreditHistory.Critical"
What I need to do is pivot at summarize on two of these columns, something I know how to do in SQL like this.
SELECT
Class
, SUM(CASE WHEN `CreditHistory.Critical` = 1 THEN 1 ELSE 0 END) AS Critical
, SUM(CASE WHEN `CreditHistory.Critical` = 0 THEN 1 ELSE 0 END) AS NotCritical
, SUM(CASE WHEN `CreditHistory.Critical` = 1 THEN 1 ELSE 0 END) / COUNT(*) AS PctCritical
FROM `credit.all`
GROUP BY
Class
Which would produce something like this
However, I am struggling mightily to get a foothold in R, using books and Google, it seems I should use reshape2 melt and dcast to achieve something like this. What I have tried are basically variants of this:
library(reshape2)
credit.melted <- melt(credit.all[,c(1,17)], ID=c("name", "Class"))
dcast(credit.melted, Class~CreditHistory.Critical, nrow, fill=0)
But all of my attempts with these functions have produced errors too cryptic and too common to understand what I am doing wrong.
Error in vapply(indices, fun, .default) : values must be length 1,
but FUN(X[[1]]) result is length 0
Sometimes my random permutations of the function calls produce slightly different error output, but nothing that points me in the right direction.
Question: How can I do the pivoted summary similar to the SQL result using R?

Answer 1

I wouldn't really consider this a pivot. You're not trying to use a pivot command in SQL. You can use dplyr to follow the exact same method as your SQL:
library(dplyr)
credit.all %>%
group_by(Class) %>%
summarize(Critical = sum(CreditHistory.Critical == 1),
NotCritical = sum(CreditHistory.Critical == 0),
PctCritical = mean(CreditHistory.Critical == 1))
# # A tibble: 2 x 4
# Class Critical NotCritical PctCritical
# <fct> <int> <int> <dbl>
# 1 Bad 50 250 0.167
# 2 Good 243 457 0.347
Since it's a binary column the == 1 isn't really necessary, but I leave it in because (a) it's more similar to your SQL code, and (b) if there were other values but you wanted the count of 1s, this would be the way to do it. However, you could get the same results a little more simply like this:
credit.all %>%
group_by(Class) %>%
summarize(Critical = sum(CreditHistory.Critical),
NotCritical = n() - Critical,
PctCritical = Critical / n())
If you really want a pivot approach, we can go that route, it just seems less straightforward. Your data is already in a long format, so we don't need to melt, we can just cast:
pivot = dcast(Class ~ CreditHistory.Critical, data = credit.all)
pivot
# Using CreditHistory.Critical as value column: use value.var to override.
# Aggregation function missing: defaulting to length
# Class 0 1
# 1 Bad 250 50
# 2 Good 457 243
You could then rename the columns and calculate the percentages:
names(pivot)[2:3] = c("NotCritical", "Critical")
pivot$PctCritical = with(pivot, Critical / (Critical + NotCritical)