How to produce tidy summaries when comparing logical conditions - r

Question

Consider the following data frame:
set.seed(123)
dat <- data.frame(Region = rep(c("a","b"), each=100),
State =rep(c("NY","MA","FL","GA"), each = 50),
Loc = rep(letters[1:20], each = 5),
ID = 1:200,
count1 = sample(4, 200, replace=T),
count2 = sample(4, 200, replace=T))
Individual observations are denoted with a unque ID. There are three grouping variables for the individual observations: Region, State, and Loc. Lets say that I know the following conditions to be true:
- When count1 equals 1 then count2 should equal 2
- When count1 equals 2 then count2 should equal 4
- When count1 equals 3 then count2 should equal 1
- When count1 equals 4 then count2 should equal 3
I want to answer the following types of questions:
1. How many observations that belong to each grouping variable (Region, State, Loc) are in each level of count1 and count2
Which IDs are in which level of count1 and count2(and what grouping variables do these IDs belong to)
How often do the conditions outlined above hold true, and how often do they not hold true
For which grouping variables and IDs do these conditions hold true, and for which do they not hold true
When the conditions do not hold true, what is actually observed (e.g., when count1 equals 1 then count2 should equal 2; so when count1 equals 1 but count2 does not equal 2, then what does count2 equal instead).
How can I specify these conditions and produce tidy summary-like tables to answer these questions?
You can think of the levels of count1 and count2 as being associated with certain characteristics, and I want understand the relationship between those levels with each other, and with the grouping variables. If anyone has any graphical visualization ideas for these types of questions, that would be very helpful as well!

Answer 1

Here's one way to go for questions 1 & 2 although this feels a little involved. I am using tidyr pivot_wider to create columns for each unique value of count1 and count2. The function length in values_fn counts the number of elements in the vectors created by pivot_wider for relevant combinations. As we need answers for count1 and count2 separately I run pivot_wider twice.
Results are then combined with bind_cols and superfluous columns are removed.
All of this can probably be improved on with a bit more thought.
library(dplyr)
library(tidyr)
library(tibble)
set.seed(123)
data <- data.frame(Region = rep(c("a","b"), each=100),
State =rep(c("NY","MA","FL","GA"), each = 50),
Loc = rep(letters[1:20], each = 5),
ID = 1:200,
count1 = sample(4, 200, replace=T),
count2 = sample(4, 200, replace=T))
# 1. How many observations that belong to each grouping variable (Region, State, Loc) are in each level of count1 and count2
level_count1 <- data %>%
select(-count2) %>%
pivot_wider(id_cols = c(Region, State, Loc),
values_from = count1,
values_fn = list(count1 = length),
names_from = count1,
names_prefix = "count1_")
level_count2 <- data %>%
select(-count1) %>%
pivot_wider(id_cols = c(Region, State, Loc),
values_from = count2,
values_fn = list(count2 = length),
names_from = count2,
names_prefix = "count2_")
level_count <- bind_cols(level_count1, level_count2) %>% select(-Region1, -State1, -Loc1)
# 2. Which IDs are in which level of count1 and count2(and what grouping variables do these IDs belong to)
ID_count1 <- data %>%
select(-count2) %>%
pivot_wider(id_cols = ID,
values_from = count1,
values_fn = list(count1 = length),
names_from = count1,
names_prefix = "count1_") %>%
left_join(data %>% select(Region, State, Loc, ID), by = "ID")
ID_count2 <- data %>%
select(-count1) %>%
pivot_wider(id_cols = ID,
values_from = count2,
values_fn = list(count2 = length),
names_from = count2,
names_prefix = "count2_") %>%
left_join(data %>% select(Region, State, Loc, ID), by = "ID")
ID_count <- bind_cols(ID_count1, ID_count2) %>% select(-Region1, -State1, -Loc1, -ID)
Results are like this:
> head(level_count)
# A tibble: 6 x 11
Region State Loc count1_3 count1_1 count1_2 count1_4 count2_4 count2_2 count2_1 count2_3
<fct> <fct> <fct> <int> <int> <int> <int> <int> <int> <int> <int>
1 a NY a 3 2 NA NA 2 1 1 1
2 a NY b 2 1 2 NA NA 2 1 2
3 a NY c NA NA 3 2 NA 2 2 1
4 a NY d 1 3 NA 1 1 3 NA 1
5 a NY e 2 3 NA NA 2 3 NA NA
6 a NY f 1 2 1 1 NA NA 2 3
The value of 3 in the first row of column count1_3 means that the combination of Region == "a", State == "NY" and Loc == "a" occurs 3 times for the value 3 for count1. Likewise, the value 2 in the second row indicates that the value 3 occurs twice in count1 for the combination Region == "a", State == "NY" and Loc == "b".
NA values indicate that 3 does not occur for the given combination of categorical columns. And so on. Is this useful for you?
The approach for ID is similar.