Quartoblog - Long to wide (and vv) transformations

Introduction

In this session, Sam Langton demonstrated long to wide (and wide to long) transformations using functions available in the tidyr package using data from the London Fire Brigade (Langton, 2022).

Preparation

library(tidytuesdayR)
library(dplyr)
library(tidyr)
library(forcats)
library(ggplot2)

Identify datasets available (fire service data).

tt_df <- tidytuesdayR::tt_load("2021-06-29")


    Downloading file 1 of 1: `animal_rescues.csv`

Pull out the data.

fire_df <- tt_df$animal_rescues

Save for train working.

Harrie: don't know what this means
save.image("data/train_working.Rdata")
load("data/train_working.Rdata")

Basic cleaning.

fire_clean_df <- fire_df %>% 
  filter(cal_year < 2021) %>% 
  mutate(animal_group_broad = fct_lump(animal_group_parent, n = 5))

Transformations

Create some wide data. Freq counts of incidents for each animal by year, then wide. 2021 is not complete. Note animal category is just for the demo!

Take care, this is a crude recode.

fire_wide1_df <- fire_clean_df %>% 
  filter(cal_year < 2021) %>% 
  select(incident_number, cal_year, animal_group_broad) %>% 
  group_by(cal_year, animal_group_broad) %>% 
  summarise(yearly_count = n()) %>% 
  pivot_wider(names_from = cal_year, values_from = yearly_count)

Non-longitudinal example. Note the fiddly way of keeping zero counts. Requires an ungroup() and then complete(). There might be a better way of doing this!

fire_wide2_df <- fire_clean_df %>% 
  filter(cal_year == 2009) %>% 
  select(incident_number, animal_group_broad, property_category)  %>% 
  group_by(animal_group_broad, property_category) %>% 
  summarise(yearly_count = n()) %>% 
  ungroup() %>% 
  complete(animal_group_broad, property_category, fill = list(yearly_count = 0)) %>% 
  pivot_wider(names_from = property_category, values_from = yearly_count)

Nested data / multilevel example.

fire_wards_df <- fire_clean_df %>%
  group_by(borough_code, ward_code, cal_year) %>% 
  summarise(yearly_count = n()) %>% 
  ungroup() %>% 
  filter(ward_code != "NULL") %>% 
  arrange(borough_code, ward_code, cal_year)

Make wide for example, fill in zeros and arrange. Why interesting? Because it’s nested levels, longitudinal and ‘year’ in the var names.

fire_wide3_df <- fire_wards_df %>% 
  arrange(cal_year) %>% 
  pivot_wider(names_from = cal_year, values_from = yearly_count, names_prefix = "year_") %>% 
  mutate_if(is.numeric, ~replace_na(., 0)) %>% 
  arrange(ward_code)

Remove things that might confuse people.

rm(fire_clean_df, fire_df, tt_df, fire_wards_df)

References

Langton, S. (2022, April). NSC-R Workshops: NSC-R Tidy Tuesday. NSCR. Retrieved from https://nscrweb.netlify.app/posts/2022-01-25-nsc-r-tidy-tuesday/