Optimizing selection of color palettes

# Optimizing selection of color palettes
### Computation Skills Workshop

---

# What to color by

---

# Categorical color scales

---

# Sequential color scales

---

# Diverging color scales

---

# Highlighting/de-emphasizing

---

# Choosing a color scale

- Emphasis on interpretability and accessibility
- Default palettes are less than desirable
- Variables may require transformations

---

# Default palette in `ggplot2`

---

# Suboptimal default choices

---

# Common forms of color vision deficiency

### Red-green

- Deuteranomaly
- Protanomaly
- Protanopia and deuteranopia

### Blue-yellow

- Tritanomaly
- Tritanopia

### Complete color vision deficiency

- Monochromacy

---

# Inspecting for color vision deficiency

---

# Inspecting for color vision deficiency

```r
library(colorblindr)
cvd_grid(plot = pen_fig)
```

---

# Inspecting for color deficiency

---

# Inspecting for color deficiency

---

# Inspecting for color deficiency

---

# When to use quantitative or qualitative color scales?

---

# Use hues for nominal variables

---

# Use hues for nominal variables

---

# Quantitative `$\neq$` continuous

---

# Shades to emphasize order

---

# Overcomplicating graphs

---

# Overcomplicating graphs

---

# Correcting the graph

---

# Encoding a new variable

---

# Double-encoded line chart

---

# Shades to distinguish subcategories

---

# Stick to shades of one hue

---

# Implementing optimal color palettes in R

- [`RColorBrewer`](https://colorbrewer2.org/)
- [`viridis`](https://sjmgarnier.github.io/viridis/)

---

# Color Brewer

## Sequential palettes

---

# Color Brewer

## Qualitative palettes

---

# `gss`

```
## # A tibble: 1,974 × 221
##     year    id wrkstat   wrkslf  wrkgvt  marital  sibs childs   age educ  degree
##    <dbl> <dbl> <fct>     <fct>   <fct>   <fct>   <dbl>  <dbl> <dbl> <fct> <fct> 
##  1  2012     1 WORKING … SOMEON… PRIVATE Never …     1      0    22 4 ye… Bache…
##  2  2012     2 WORKING … SOMEON… PRIVATE Never …     2      0    21 12th… HS    
##  3  2012     3 WORKING … SOMEON… PRIVATE Married     1      2    42 12th… HS    
##  4  2012     4 <NA>      SOMEON… PRIVATE Married     2      2    49 1 yr… HS    
##  5  2012     5 RETIRED   SOMEON… GOVERN… Separa…     0      3    70 4 ye… Bache…
##  6  2012     6 <NA>      SOMEON… PRIVATE Widowed     4      2    50 7 ye… Bache…
##  7  2012     7 KEEPING … SOMEON… PRIVATE Married     2      2    35 3 ye… Junio…
##  8  2012     8 KEEPING … <NA>    <NA>    Separa…     2      3    24 11th… <HS   
##  9  2012     9 KEEPING … <NA>    <NA>    Married     2      2    28 9th … <HS   
## 10  2012    10 WORKING … SOMEON… PRIVATE Never …     0      0    28 5 ye… Bache…
## # … with 1,964 more rows, and 210 more variables: sex <fct>, race <fct>,
## #   polviews <fct>, partyid <fct>, mobile16 <fct>, born <fct>, income06 <fct>,
## #   rincom06 <fct>, region <fct>, size <dbl>, vote08_coded <fct>, pres08 <fct>,
## #   natspac <fct>, natenvir <fct>, natheal <fct>, natcity <fct>,
## #   natcrime <fct>, natdrug <fct>, nateduc <fct>, natrace <fct>, natarms <fct>,
## #   nataid <fct>, natfare <fct>, natroad <fct>, natsoc <fct>, natmass <fct>,
## #   natpark <fct>, natchld <fct>, natsci <fct>, natenrgy <fct>, eqwlth <dbl>, …
```

---

# `scale_*_brewer()`

```r
ggplot(data = gss_polviews, mapping = aes(x = polviews, fill = polviews)) +
  geom_bar() +
  scale_fill_brewer(type = "seq")
```

---

# `scale_*_brewer()`

```r
ggplot(data = gss_polviews, mapping = aes(x = polviews, fill = polviews)) +
  geom_bar() +
  scale_fill_brewer(type = "seq", guide = FALSE)
```

---

# Sequential palette

```r
ggplot(data = gss_polviews, mapping = aes(x = polviews, fill = polviews)) +
  geom_bar() +
  scale_fill_brewer(palette = "OrRd", guide = FALSE)
```

---

# Qualitative palette

```r
ggplot(data = gss_relig, mapping = aes(x = relig, fill = relig)) +
  geom_bar() +
  scale_fill_brewer(guide = FALSE)
```

---

# Qualitative palette

```r
ggplot(data = gss_relig, mapping = aes(x = relig, fill = relig)) +
  geom_bar() +
  scale_fill_brewer(type = "qual", guide = FALSE)
```

---

# Quantitative color brewer palettes

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_distiller()
```

---

# Quantitative color brewer palettes

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_distiller(palette = "YlGn")
```

---

# `viridis`

---

# Comparison of palettes

---

# Green-Blind (Deuteranopia)

---

# Desaturated

---

# `viridis` - continuous

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_viridis_c()
```

---

# `viridis` - continuous

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_viridis_c(option = "magma")
```

---

# `viridis` - discrete

```r
ggplot(data = gss_polviews, mapping = aes(x = polviews, fill = polviews)) +
  geom_bar() +
  scale_fill_viridis_d(guide = FALSE)
```

---

# Exercise on implementing palettes

---

# When to use sequential or diverging color scales?

---

# Except when you shouldn't!

- Meaningful middle point
- Emphasize extremes
- See more differences

---

# Meaningful middle point

- Zero
- 50%
- Average
- Agreed threshold
- A target

---

# WTF?

---

# Emphasize highest values

---

# Emphasize extremes

---

# Increased range in scale

---

# Increased range in scale

---

# Implementing diverging palettes

```r
ggplot(data = gss_polviews, mapping = aes(x = polviews, fill = polviews)) +
  geom_bar() +
  scale_fill_brewer(type = "div", guide = FALSE)
```

---

# Implementing diverging palettes

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_distiller(type = "div")
```

---

# Implementing diverging palettes

```r
ggplot(data = unemp_20, mapping = aes(fill = unemp)) +
  geom_sf() +
  scale_fill_viridis_c(option = "turbo")
```

---

# Exercise on diverging palettes

---

# When to use classed or unclassed color scales?

---

# Use a classed scale if the data is classed

---

# Communicating statistical brackets

---

# Communicating general patterns

---

# Increased classes for more nuance

---

# Assisting with interpretations

---

# Directly encoding values

---

# Choosing an interpolation

- Linear
- Quartiles, quantiles, deciles

---

# Linear

---

# Linear

---

# Quartiles

---

# Quintiles

---

# Deciles

---

# Natural breaks

---

# Implementing breaks

- Automatically with `scale_*_viridis_b()`
- Manually with `cut_number()`

---

# Quartiles

```r
ggplot(data = unemp_oct21,
       mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_viridis_b(n.breaks = 3,
                       nice.breaks = FALSE)
```

]

]

---

# Quintiles

```r
ggplot(data = unemp_oct21,
       mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_viridis_b(n.breaks = 4,
                       nice.breaks = FALSE)
```

]

]

---

# Deciles

```r
ggplot(data = unemp_oct21,
       mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_viridis_b(n.breaks = 9,
                       nice.breaks = FALSE)
```

]

]

---

# Manually with `cut_number()`

```r
unemp_oct21 %>%
  mutate(unemp = cut_number(unemp, n = 4)) %>%
  ggplot(mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_brewer(type = "seq")
```

]

]

---

# Manually with `cut_number()`

```r
unemp_oct21 %>%
  mutate(unemp = cut_number(unemp, n = 5)) %>%
  ggplot(mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_brewer(type = "seq")
```

]

]

---

# Manually with `cut_number()`

```r
unemp_oct21 %>%
  mutate(unemp = cut_number(unemp, n = 10)) %>%
  ggplot(mapping = aes(fill = unemp)) +
  geom_sf(color = "white", size = .1) +
  scale_fill_brewer(type = "seq")
```

]

]

---

# Natural breaks

```r
library(classInt)

# calculate intervals
unemp_classes <- classIntervals(unemp_oct21$unemp,
                                n = 5,
                                style = "jenks")

unemp_oct21 %>%
  # apply to variable
  mutate(unemp_cat = kimisc::cut_format(
    x = unemp,
    breaks = unemp_classes$brks,
    include.lowest = TRUE,
    format_fun = scales::percent
  )) %>%
  ggplot(mapping = aes(fill = unemp_cat)) +
  geom_sf(color = "white", size = .1)
```

]

]

---

# Exercises on interpolating continuous variables