---
title: "Practical R: Factors"
author: Abhijit Dasgupta
---
```{r setup, include=F, child = here::here('slides/templates/setup.Rmd')}
```
```{r setup1, include=FALSE}
library(countdown)
library(fontawesome)
```
class: middle,center
# Categorical variables
---
## What are categorical variables?
Categorical variables are variables that
+ have values defining categories of things
+ typically have a few unique values
+ may or may not be ordered
+ are not interval-scaled, i.e., their differences don't make sense _per se_
---
## What are categorical variables?
.pull-left[
#### Non-ordered
1. Race (White, Black, Hispanic, Asian, Native American)
1. Gender (Male, Female, Other)
1. Geographic regions (Africa, Asia, Europe, North America, South America)
1. Genes/Proteins
]
.pull-right[
#### Ordered
1. Income levels (< $10K, $10K - $25K, $25K - $75K, $75K - $100K)
1. BMI categories (Underweight, Normal, Overweight, Obese)
1. Number of bedrooms in houses (1 BR, 2BR, 3BR, 4BR)
]
---
class: middle, center
# Categorical variables in R
---
## The `factor` data type
R stores categorical variables as type `factor`.
+ You can coerce a character or numeric object into a factor using `as.factor`.
+ You can check if an object is a factor with `is.factor`.
+ You can create a factor with the function `factor`.
---
## The `factor` data type
> `factor(x = character(), levels, labels = levels,
exclude = NA, ordered = is.ordered(x), nmax = NA)`
>
> factor returns an object of class "factor" which has a set of integer codes the length of x with a "levels" attribute of mode character and unique
----
+ Internally, each level of a factor is coded as an integer
+ Each such integer has a corresponding `level` which is a character, describing the level.
+ You can add `labels` to each level to change the printed form of the factor.
---
## The `factor` data type
```{r factors-1}
x <- c('Maryland','Virginia', 'District', 'Maryland','Virginia') # a character vector
xf <- as.factor(x)
xf
```
There are three levels, that by default are in alphabetical order
----
.pull-left[
```{r factors-2}
as.integer(xf)
```
+ District = 1, Maryland = 2, Virginia = 3
]
.pull-right[
```{r factors-3}
as.character(xf)
```
+ Get original characters back
]
---
## The `factor` data type
```{r factors-4}
y <- c(5, 3, 9, 4, 5, 3)
yf <- as.factor(y)
yf
```
Levels are still in alphanumeric order
----
.pull-left[
```{r factors-5}
as.numeric(yf)
```
+ Note, we don't get original integers back!!
+ 3 = 1, 4 = 2, 5 = 3, 9 = 4
]
.pull-right[
```{r factors-6}
as.numeric(as.character(yf))
```
+ This is how you get numbers back
]
---
## The `factor` data type
.pull-left[
```{r factors-7}
x <- c('MD','DC','VA','MD','DC')
xf <- factor(x)
unclass(xf)
```
]
.pull-right[
```{r factors-8}
x <- c('MD','DC','VA','MD','DC')
xf <- factor(x, levels = c('MD','DC','VA'))
unclass(xf)
```
]
----
+ If I change the level designation, the underlying coding changes
+ This is important when a factor is an independent variable in a regression model
---
## The `factor` data type
The `drv` variable in the `mpg` dataset tells us the kind of drive (front, rear or 4-wheel) each car has. However
it's coded as `f`, `r`, and `4`, which is not great for display purposes. We can re-label these levels, but we have to be a bit careful
.pull-left[
```{r factors-9, error=TRUE}
x <- mpg$drv
xf <- factor(x,
levels = c('4-wheel','Front wheel',
'Rear wheel'))
head(xf)
```
]
.pull-right[
```{r factors-10}
x <- mpg$drv
xf <- factor(x,
levels = c('4', 'f','r'), #<<
labels = c('4-wheel','Front wheel',#<<
'Rear wheel'))#<<
head(xf)
```
]
----
Levels have to match what's actually in the original data, but you can re-label the levels.
---
class: middle, center
# Why factors?
---
## Factors are R's discrete data type
+ Factors are interpreted as discrete by R's functions
.pull-left[
```{r factors-11, message=TRUE, warning=TRUE, fig.height=3}
ggplot(mpg,
aes(year, cty))+
geom_boxplot()
```
]
.pull-right[
```{r factors-12, fig.height=3}
ggplot(mpg,
aes(as.factor(year), cty))+
geom_boxplot()
```
]
---
## Dummy variables are automatically created from factors
.pull-left[
```{r factors-13, eval=F}
model.matrix(~species, data = palmerpenguins::penguins)
```
```{r factors-14, echo = F}
palmerpenguins::penguins %>%
group_by(species) %>%
slice_head(n=2) %>%
ungroup() %>%
model.matrix(~species, data = .)
```
]
.pull-right[
+ If a factor has _n_ levels, you get _n-1_ dummy variables
+ The level corresponding to integer code 1 is omitted as the reference level
]
----
Changing the base level (integer code 1) changes model interpretation since
it changes the reference level against which all other levels are compared.
---
class: middle, center
# Manipulating factors
The `forcats` package (part of `tidyverse`)
---
## Effect in models
.pull-left[
```{r factors-15}
library(palmerpenguins)
m <- lm(body_mass_g ~ species, data = penguins)
broom::tidy(m)
```
Compare with Adele
]
.pull-right[
```{r factors-16}
p1 <- penguins %>%
mutate(species = fct_relevel(species, 'Gentoo'))
m1 <- lm(body_mass_g ~ species, data=p1 )
broom::tidy(m1)
```
Compare with Gentoo
]
----
Providing only one level to `fct_relevel` makes that the base level (integer code 1).
You can also fully specify all the levels in order, or partially specify them. If you
partially specify them, the remaining levels will be put in alphabetical order after the
ones you specify.
---
## Effect in plots
.pull-left[
```{r factors-17, fig.height=4}
ggplot(penguins,
aes(x = species))+
geom_bar()
```
]
.pull-right[
```{r factors-18, fig.height=4}
ggplot(p1,
aes(x = species))+
geom_bar()
```
]
Changes the order in which bars are plotted
---
## Extra levels
.pull-left[
```{r factors-19}
x <- factor(str_split('statistics', '')[[1]],
levels = letters)
x
```
```{r factors-20}
p1 <- penguins %>% filter(species != 'Gentoo')
fct_count(p1$species)
```
]
.pull-right[
```{r factors-21}
fct_drop(x)
```
```{r factors-22}
p1 <- p1 %>% mutate(species = fct_drop(species))
fct_count(p1$species)
```
]
Getting rid of extra levels
Sometimes levels with no data show up in summaries or plots
---
# Ordering levels by frequency
.pull-left[
```{r factors-23, fig.height=4}
ggplot(penguins,
aes(x = species))+
geom_bar()
```
]
.pull-right[
```{r factors-24, fig.height=4}
ggplot(penguins,
aes(x = fct_infreq(species)))+
geom_bar()
```
]
Ordering levels from most to least frequent
---
## Ordering levels by values of another variable
.pull-left[
```{r factors-25, fig.height=4}
ggplot(penguins,
aes(x = species,
y = bill_length_mm))+
geom_boxplot()
```
]
.pull-right[
```{r factors-26, fig.height=4}
ggplot(penguins,
aes(x = fct_reorder(species, bill_length_mm,
.fun=median, na.rm=T),
y = bill_length_mm))+
geom_boxplot() + labs(x = 'species')
```
]
`fct_reorder` is useful for ordering a plot by ascending or descending levels. This
makes the plot easier to read.
---
## Ordering levels by values of another variable
```{r factors-27, echo=FALSE}
if('USArrests' %in% ls()) rm(USArrests)
if(!('State' %in% names(USArrests))) USArrests <- USArrests %>% rownames_to_column('State')
```
```{r factors-28, eval=F}
USArrests <- USArrests %>% rownames_to_column('State')
```
.pull-left[
```{r factors-29, fig.height=5}
ggplot(USArrests, aes(x=State, y = Murder))+
geom_bar(stat = 'identity') +
theme(axis.text = element_text(size=6))+
coord_flip()
```
]
.pull-right[
```{r factors-30, fig.height=5}
ggplot(USArrests, aes(
x = fct_reorder(State, Murder),
y = Murder))+
geom_bar(stat = 'identity')+
theme(axis.text = element_text(size=6))+
coord_flip()
```
]
---
## Order levels based on last values when plotting 2 variables
The level ordering also shows up in the order of levels in the legends of plots.
Suppose you are plotting two variables, grouped by a factor.
.pull-left[
```{r factors-31, fig.height=4}
ggplot(iris, aes(
x = Sepal.Length,
y = Sepal.Width,
color = Species))+
geom_smooth(se=F)
```
]
.pull-right[
```{r factors-32, fig.height=4}
ggplot(iris, aes(
x = Sepal.Length,
y = Sepal.Width,
color = fct_reorder2(Species,
Sepal.Length, Sepal.Width)))+
geom_smooth(se=F) + labs(color = 'Species')
```
]
---
## Further exploration
1. [forcats cheatsheet](https://github.com/rstudio/cheatsheets/raw/master/factors.pdf)
1. [Chapter 15](https://r4ds.had.co.nz/factors.html) of R4DS