Color

class: middle, center, title-slide

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

---

## Color in Visualizations

How do we choose colors appropriately?

We need to pay attention to data type!

.pull-left[
### Sequential Palette

- imply order among values

### Qualitative Palette

- imply no ordering
]

.pull-right[
### Diverging Palette

- emphasis on low, mid, and high
]

---

## Natural Color Mappings

Sometimes, the type of data implies a natural color mapping, such as temperature, geographical features, or political affiliation.

<div class="figure" style="text-align: center">
<img src="images/europe-forest.png" alt="Forest cover in Europe (www.eea.europa.eu)." width="70%" />
<p class="caption">Forest cover in Europe (www.eea.europa.eu).</p>
</div>

---

## ggplot2

ggplot2 tries to **guess** which color palette to use:

- ordinal and numerical variables are mapped to a **sequential** palette
- unordered categorical variables are mapped to a **qualitative** palette

but ggplot2 never knows when to map a variable a **diverging** palette, or which
direction of values indicate strongest intensity.

### Applying a Color Palette in ggplot2

- `scale_colour_*` for color mappings: 
  - `geom_point()`, 
  - `geom_line()`,
  - etc
- `scale_fill_*` for fill mappings 
  - `geom_col()`, 
  - `geom_tile()`,
  - etc
  
---

## Sequential Palettes

A sequential palette
- is typically used for data that is ordered from low to high.
- consists of colors with varying lightness, progressing from light to dark.

**Examples:** population count, quality rating, size

```r
ggplot(diamonds, aes(price, fill = cut)) +
  geom_histogram(binwidth = 1000) +
* scale_fill_ordinal()
```

---

## Diverging Palettes

Diverging palettes place equal emphasis on the middle, low, and high ends of the scale.

In ggplot2:
- `scale_fill_gradient2()`
- `scale_color_gradient2()`

**Examples:** temperature (celsius), budget balance, correlations

.pull-left[

```r
# see source for dataset
mtcars_cor %>%
  ggplot(aes(var1, 
             var2, 
             fill = cor)) +
  geom_tile() +
* scale_fill_gradient2() +
  labs(x = NULL, y = NULL)
```
]

.pull-right[
<div class="figure" style="text-align: center">
<img src="color_files/figure-html/unnamed-chunk-7-1.png" alt="Heatmap of correlation matrix." width="324" />
<p class="caption">Heatmap of correlation matrix.</p>
</div>

]

---

## Qualitative Palettes

Colors should be as distinct as possible and should not signal differences in magnitude.

**Examples:** political party affiliation, gender

```r
ggplot(msleep, aes(bodywt, sleep_total, color = vore)) +
  geom_point() +
  scale_x_log10() +
* scale_color_discrete()
```

---

## Color Blindness

.pull-left[
red-green color blindness affects 8% of men and 0.5% of females<sup>1</sup>

Several different types: 
- protanotopia
- deuteranopia
- protanomaly 
- deuteranomaly
- tritanopia
- tritanomaly
]

.pull-right[
<div class="figure" style="text-align: center">
<img src="images/ishihara.png" alt="Ishihara color test. What number do you see? (en.wikipedia.org/wiki/Color_blindness)" width="455" />
<p class="caption">Ishihara color test. What number do you see? (en.wikipedia.org/wiki/Color_blindness)</p>
</div>

]

.footnote[
<sup>1</sup> Of people of Northern European descent.
]

---

### Simulating Color Blindness

```r
library(colorBlindness)

p <- ggplot(mpg, aes(hwy, fill = drv)) + geom_density(alpha = 0.5)
*cvdPlot(p)
```

---

## ColorBrewer

ColorBrewer, based on work on color use in maps <a name=cite-harrower2003></a>([Harrower and Brewer, 2003](https://www.tandfonline.com/doi/abs/10.1179/000870403235002042)), is applicable to most types of plots.

- sequential, diverging, and qualitative palettes, many tailored to individuals
with color blindness

- requires [RColorBrewer](https://CRAN.R-project.org/package=RColorBrewer) package

.pull-left[
- `scale_color_brewer()`: discrete (integer, categorical) data

- `scale_color_distiller()`: continuous data

```r
p + 
  scale_fill_brewer(
    palette = "Accent"
  )
```

]

.pull-right[
<img src="color_files/figure-html/unnamed-chunk-11-1.png" width="345.6" style="display: block; margin: auto;" />

]

---

<div class="figure" style="text-align: center">
<img src="color_files/figure-html/unnamed-chunk-12-1.png" alt="All the palettes in ColorBrewer." width="720" />
<p class="caption">All the palettes in ColorBrewer.</p>
</div>

---

## Viridis

**Sequential** palettes created by Stéfan van der Walt and Nathaniel Smith for the Python matplotlib library

- are perceptually uniform and great for individuals with color blindness.

- requires [viridis](https://CRAN.R-project.org/package=viridis) 
(or [viridisLite](https://CRAN.R-project.org/package=viridisLite))

.pull-left[
- `scale_viridis_d`: discrete data
- `scale_viridis_c`: continuous data

```r
tibble(x = rnorm(1e4), 
       y = rnorm(1e4)) %>%
  ggplot(aes(x, y)) +
  geom_hex() + 
* scale_fill_viridis_c() +
  coord_fixed()
```
]

.pull-right[
<img src="color_files/figure-html/unnamed-chunk-13-1.png" width="345.6" style="display: block; margin: auto;" />
]

---

<div class="figure" style="text-align: center">
<img src="color_files/figure-html/unnamed-chunk-14-1.png" alt="Palettes from the viridis package." width="576" /><img src="color_files/figure-html/unnamed-chunk-14-2.png" alt="Palettes from the viridis package." width="576" /><img src="color_files/figure-html/unnamed-chunk-14-3.png" alt="Palettes from the viridis package." width="576" /><img src="color_files/figure-html/unnamed-chunk-14-4.png" alt="Palettes from the viridis package." width="576" /><img src="color_files/figure-html/unnamed-chunk-14-5.png" alt="Palettes from the viridis package." width="576" />
<p class="caption">Palettes from the viridis package.</p>
</div>

---

## Other Considerations

### 1. Greyscale

Color visualizations are still sometimes printed in grayscale; therefore, it may be a good idea to use palettes that vary in lightness.

<div class="figure" style="text-align: center">
<img src="color_files/figure-html/unnamed-chunk-15-1.png" alt="The same plot in original colors and greyscale." width="720" />
<p class="caption">The same plot in original colors and greyscale.</p>
</div>

---

### 2. Aesthetics

Aesthetics are **subjective** and **fickle**, yet they remain important to consider.

Beautifully crafted visualizations will always attract more attention.

<div class="figure" style="text-align: center">
<img src="color_files/figure-html/unnamed-chunk-16-1.png" alt="Which would you prefer?" width="720" />
<p class="caption">Which would you prefer?</p>
</div>

---

## References

<a name=bib-harrower2003></a>[Harrower, M. and C. A.
Brewer](#cite-harrower2003) (2003). "ColorBrewer.Org: An Online Tool
for Selecting Colour Schemes for Maps". In: _The Cartographic Journal_
40.1, pp. 27-37. ISSN: 0008-7041. DOI:
[10.1179/000870403235002042](https://doi.org/10.1179%2F000870403235002042).
URL:
[https://www.tandfonline.com/doi/abs/10.1179/000870403235002042](https://www.tandfonline.com/doi/abs/10.1179/000870403235002042)
(visited on Sep. 16, 2020).