Categorical Data

class: middle, center, title-slide

.title[
# Categorical Data
]
.subtitle[
## Data Visualization
]
.author[
### Johan Larsson
]
.author[
### Behnaz Pirzamanbein
]
.institute[
### The Department of Statistics, Lund University
]

---

## Visualizing Categorical Data

Usually, the real aim is to visualize **proportions**.

.pull-left[
### Example:

happiness data from the US in 2006

```r
library(tidyverse)
data(
  "happiness",
  package = "wooldridge"
)

# see slide source code for 
# data wrangling code
```

]

.pull-right[
### Bar Charts

simple and readable

```r
ggplot(happiness, aes(happy)) +
* geom_bar()
```

<img src="categorical-data_files/figure-html/unnamed-chunk-3-1.png" width="345.6" style="display: block; margin: auto;" />
]

---

## Grouped Bar Charts

Grouped bar chart is good 
- when counts are what matters or 
- when you have small data sets

```r
ggplot(happiness, aes(gender, fill = happy)) +
* geom_bar(position = "dodge", col = "black")
```

---

## Stacked Bar Charts

Stacked bar chart is compact way to visualize counts, but seldom preferable over grouped bar chart.

```r
ggplot(happiness, aes(gender, fill = happy)) +
* geom_bar(col = "black") # position = "stack" is the default
```

---

## Proportional Grouped Bar Charts

Proportional grouped bar chart is 
- good default choice for bar charts (when you have enough data),
- but need to summarize data before plotting!

```r
happiness_props <- group_by(happiness, gender, happy) %>%
  summarize(n = n()) %>%
  mutate(prop = n / sum(n))

ggplot(happiness_props, aes(gender, prop, fill = happy)) +
* geom_col(position = "dodge", col = "black")
```

---

## Proportional Stacked Bar Charts

Proportional stacked bar chart is 
- compact way to visualize proportions.
- slightly more intuitive than grouped proportional bar charts, 
- but harder to read.

```r
ggplot(happiness, aes(gender, fill = happy)) +
* geom_bar(position = "fill", col = "black") +
  labs(y = "proportion")
```

---

## Mosaic Plots

Mosaic plot is a type of stacked bar chart that maps category size to the width of bars.

- need to use another package for this:
[ggmosaic](https://CRAN.R-project.org/package=ggmosaic)

```r
library(ggmosaic)
ggplot(happiness) +
* geom_mosaic(aes(product(happy, gender), fill = happy)) +
  guides(fill = "none") # remove redundant ink
```

---

## What About Pie Charts?

Pie chart is OK in a few instances, but avoid as a thumb rule.

.pull-left[
<img src="categorical-data_files/figure-html/unnamed-chunk-9-1.png" width="288" style="display: block; margin: auto;" />
]

.pull-right[
<img src="categorical-data_files/figure-html/unnamed-chunk-10-1.png" width="345.6" style="display: block; margin: auto;" />
]

---

## Waffle Plots

Waffle plot is suitable when there are large differences between categories.

- again need to use a new package:
[waffle](https://CRAN.R-project.org/package=waffle)

<div class="figure" style="text-align: center">
<img src="categorical-data_files/figure-html/unnamed-chunk-11-1.png" alt="A waffle plot of the happiness data. Every square represents approximately 10 people. The code is quite complicated (see the source code)." width="504" />
<p class="caption">A waffle plot of the happiness data. Every square represents approximately 10 people. The code is quite complicated (see the source code).</p>
</div>

---

## Pay Attention to Mappings

The choice of mappings is critical when dealing with categorical data!

```r
*ggplot(happiness, aes(happy, fill = gender)) +
  geom_bar(position = "fill", col = "black") +
  ylab("Proportion")
```

---

## Ordering

- if the variable is ordinal, format it as an ordered factor.
- if the variable is **not** ordinal, consider ordering it manually!

```r
# no ordering
happiness %>%
  count(region) %>%
  ggplot(aes(region, n)) +
  geom_col() +
  guides(x = guide_axis(n.dodge = 2)) # avoid label overlap
```

---

## Ordering

- if the variable is ordinal, format it as an ordered factor.
- if the variable is **not** ordinal, consider ordering it manually!

```r
happiness %>%
  count(region) %>%
* mutate(region = fct_reorder(region, n, .desc = TRUE)) %>%
  ggplot(aes(region, n)) +
  geom_col() +
  guides(x = guide_axis(n.dodge = 2)) # avoid label overlap
```