Visualizing Uncertainty in R

# Visualizing Uncertainty in R
## hurricanes, spaghetti, and hypothetical outcome plots
### Will Bonnell
### 2018/04/26

---

.remark-code, .remark-inline-code {
  background: #f0f0f0;
}
.remark-code {
  font-size: 12px;
}
.huge .remark-code { /*Change made here*/
  font-size: 200% !important;
}
.tiny .remark-code { /*Change made here*/
  font-size: 70% !important;
}
</style>

1. Action-accuracy gap<sup>1</sup>

1. Status quo of easily misinterpretable dataviz

1. Misconveyed info leads to loss of credibility of researchers

### What can we do

1. Frequency framing

1. Express potential outcomes

]

<br>

![an image](https://upload.wikimedia.org/wikipedia/commons/thumb/4/4e/Irma_2017-09-06_1745Z.jpg/800px-Irma_2017-09-06_1745Z.jpg)

]

---
class: inverse, center, middle

#The Status Quo

---

##Cones of Uncertainty

![an image](https://mikehaskew.files.wordpress.com/2008/09/ike.jpg)

Image credit: [Cornell Risk Communication Research Group](https://www.risk.comm.cornell.edu/research_briefs/the-cone-of-uncertainty/)

---

###Reimagined

```r
map <- get_map(location = c(lon = -72, lat = 24), zoom = 5, maptype = "satellite")

g <- ggmap(map, extent = "device") +
  geom_path(data = hurricane1, aes(x = lon, y = lat, group = hurricane),  colour = "#ff4500", size = 1, alpha = 0.6) +
  geom_point(data = hurricane1, aes(x = lon, y = lat), size = 2, alpha = 0.9, colour = "#ff0800") +
  geom_text(data = hurricane1[1,], aes(x = lon, y = lat), size = 10, label = emoji("cyclone")) +
  coord_fixed(ylim=c(16, 34))
g
```

---

#Error bars and violins

![](index_files/figure-html/plot1-1.svg)

]

![](index_files/figure-html/plot-1.svg)

]

---

#Potential Replacement

---

##Hypothetical Outcome Plots

1. Animated plots to show potential values or probabilities

1. Allow individuals to think in finite terms (counts) instead of inifinite terms (probabilities)

1. Outperformed error bars and violin plots in assessing respondents ability to predict probability that B > A<sup>2</sup>

---

Plotting potential real values

```r
ggplot(mtv_hyp, aes(age, probability)) + 
  geom_errorbar(aes(ymin = probability, ymax = probability)) + 
  theme_minimal() +
  theme(axis.title.y = element_text(margin = margin(r = 20))) +
  # its animated yo
  transition_states(
    simulation,
    transition_length = 2,
    state_length = 1) +
  enter_fade() + 
  exit_shrink() +
  ease_aes('sine-in-out') +
  shadow_mark(past = TRUE, future = TRUE, color = "#d3d3d3") + 
  labs(x = "Age", y = "Percentage")
```

---

###Distribution example

```r
mb <- microbenchmark(x, y, n = 50)

ggplot2::autoplot(mb) +
  theme_minimal
```

![an image](distribution.png)

---

###Hypothetical runtimes

```r
ggplot(mb1, aes(expr, time)) + 
  geom_errorbar(aes(ymin = time, ymax = time)) + 
  theme_minimal() +
  theme(axis.title.y = element_text(margin = margin(r = 20))) +
  transition_states(frame, transition_length = 2, state_length = 1) +
  enter_fade() + 
  exit_shrink() +
  ease_aes('sine-in-out') +
  shadow_mark(past = TRUE, future = TRUE, color = "#d3d3d3") + 
  labs(x = "Function", y = "Time to run")
```

---

# Thanks!

Slides created via the R package [**xaringan**](https://github.com/yihui/xaringan).

Twitter: [@_willdebras](https://twitter.com/_willdebras)

---

##References:

[Hullman, et al. 2015](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0142444)