-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathPS4.Rmd
181 lines (96 loc) · 11.9 KB
/
PS4.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
---
title: "PS 4 - Data Visualization 1: Reading Plots"
author: "Elise Hellwig"
date: "`r Sys.Date()`"
output:
pdf_document:
extra_dependencies: ["xcolor", "float"]
html_document:
df_print: paged
---
\definecolor{shadecolor}{RGB}{245,235,220}
```{r setup, include=FALSE}
knitr::opts_chunk$set(fig.pos = "!H", out.extra = "")
library(reticulate)
options(scipen=999)
```
# Introduction
Before doing this problem set, please watch [**Charts Are Like Pasta**](https://www.youtube.com/watch?v=hEWY6kkBdpo) and [**Plots, Outliers, and Justin Timberlake**](https://www.youtube.com/watch?v=HMkllhBI91Y&list=PL8dPuuaLjXtNM_Y-bUAhblSAdWRnmBUcr&index=7&t=100s).
# Types of Visualizations
Images from this section come from [**Material Design**](https://m2.material.io/design/communication/data-visualization.html#types).
There are many types of visualizations, and new ones are being created regularly. However, the most common, and the ones we will focus on, are line plots, bar plots, pie charts, scatterplots, boxplots and histograms.
{width=90%}
We can group these six types of plots based on what type of information each plot is conveying:
* Comparison - comparing data from different categories or at different points in time (line, bar)
* Composition - how much of each part makes up a whole (pie)
* Relationship - how is one variable related to another variable (scatterplot)
* Distribution - how frequent are values from a single variable (boxplot, histogram)
Line plots work well for displaying time series (ex. temperature, daily COVID cases). Bar charts work well for displaying data by category (ex. housing prices by region). Pie chars, and other composition charts, show parts of a whole. This means they are not suited for data where there is no obvious "whole" or when data points can fit into multiple "parts". Scatterplots show the relationship (or correlation) between two (or more) variables. For example, this could be how much candy people eat on Halloween by age, or the average minimum temperature by latitude for cities in the United States. Distribution plots show how spread out data is for a single variable. This could be the height of women in the United States or standardized test scores for high school sophomores in the state of California. Because boxplots are more compact, they can display distributions by group (ex. standardized test scores for CA high school sophomores by county).
# Reading plots
This section will focus on the things that can go wrong with creating a visualization, both intentional and unintentional. This is so you can create accurate and informative visualizations and so you can view any plots you see with a critical eye. Most of the visualizations in this section are on the topic of the COVID-19, mostly because nothing brings out bad data like [**public health**](https://99percentinvisible.org/episode/florence-nightingale-data-viz-pioneer/). However, because the creators of some of the plots below have been shamed into removing them from the internet, many of the images actually come from [**The Good, the Bad, and the Ugly Coronavirus Graphs**](https://www.usu.edu/math/symanzik/talks/2021_SouthwestMichiganChapter.pdf) by Jürgen Symanzik from Utah State University.
Listed below is a (non-exhaustive) list of the common problems you will see with visualizations.
1. Wrong type of visualization (done)
2. Misleading data grouping/binning (done)
3. Cherry-picking data (done)
4. Poorly ordered data (done)
5. Differences in size are not proportional (done)
6. Changes in scales (done)
7. Poor or changing baseline
8. Bad or missing title and labels (done)
9. Low contrast and poor color choice (done)
# 1. Wrong type of chart
{width=60%}
As we discussed above, different types of charts lend themselves to displaying different types of data. Data that is easy to interpret using one chart may be almost incomprehensible using a different type. Be wary of new or more complicated types of visualizations. It is much easier to come to the wrong conclusions if you are unfamiliar with the
Pie charts are the most commonly misused type of visualization. This is because for a pie chart to be accurate, all of the parts/categories must add up to 100% of the whole, and each category must be mutually exclusive from the other categories. If this is not the case, your best option is probably to use a bar plot. Figure 2 below messes up on both counts. It also gets an honorable mention for cherry picking data. I can't imagine those things were the only things people have been worried about during the pandemic.
# 2. Misleading data grouping
Many times it is necessary to group (or bin) data in order to present it on a plot. It would be very hard to see the data points on a COVID-19 death rate graph that had a bar for every age from 0 to 100 (or however the oldest person is now). It also wouldn't be very useful, because while there is a big difference between COVID-19 death rates in between people who are 25 and people who are 65, there isn't much of a difference in deaths between people who are 25 and 26.
Nonetheless, improper binning can be used to mislead or obfuscate the underlying data. Even worse than using bad binning techniques, is using different binning schemes in the same plot, like the Georgia Health Department did in Figure 3. Their color choice also contributes to the deceptive nature of the plot.
{width=60%}
# 3. Cherry-picking data
One common tactic for creating misleading plots, and in fact misleading research, is cherry-picking which data is actually included. This can mean leaving relevant data out or including non-relevant data. In Figure 4, the map creator showed COVID-19 case rates over time noting the presence or absence of mask mandates for each time series. No justification is given for why these particular countries were picked, and others (Russia, Switzerland, and the Netherlands) were left out. Additionally, while it is true that these countries different mask mandate statuses at this time during the pandemic, it is certainly not the only or even most notable difference in the countries responses to the pandemic. This plot also gets a honorable mention for poor labeling, as it is impossible to tell which line corresponds to which country.
{width=60%}
# 4. Poorly ordered data
Figure 5 is an example of poorly ordered data at its best/worst. The Georgia Health Department changes the order of the counties changes between different dates, making it hard to compare case counts by county. Even worse, the dates themselves are not in order, so we have no idea whether cases are increasing or decreasing over time. This means at first glance, it seems like cases are improving, when as we know, the opposite was happening. It also gets an honorable mention for wrong type of chart, low contrast, and hard to read labeling.
{width=90%}
# 5. Differences in aize are not proportional to differences in value
Another common way people create misleading visualizations is by having differences in the size of an object be independent of the number that object represents. Figure 6 is from Russia early on in the Pandemic. If you don't look too closely, it seems like case counts are leveling off. If you look at the numbers instead of the bars, they are getting worse, much worse. This also gets an honorable mention for poor labeling by not even having a Y axis labeled.
{width=60%}
# 6. Changes in axis scales
While it may seem obvious, the scale of the X and Y axes should not change within a given plot, and ideally within a given visualization (if it contains multiple plots).
In Figure 7, neither the X and Y axes are consistently scaled. On the X axis, the same amount of space represents 9, 5, and 4 days on the scale. On the Y axis the same distance can represent 5 thousand cases or 15 thousand cases.
{width=60%}
# 7. Poor or changing baseline
When creating a plot, especially a comparison plot, values need to be plotted relative to a reasonable (and stable) baseline. This means that axes for numeric visualizations should include 0, or an appropriate substitution based on the situation (ex. 7 for pH). It also means that stacked plots should be avoided unless the bar are labeled in the plot, like in Figure 8.
{width=50%}
Figure 9, is a particularly egregious example of choosing a misleading baseline. Because the scale starts at 34\% and not 0\%, the size of the bar for the higher tax rate is several times larger than the bar for the lower tax rate, even though it is only a 13\% difference in value.
{width=60%}
# 8. Poor or missing title and labels
All plots should have a plot title as well as axes labels that tell you what is being plotted, at what scale, and what the units are (if applicable). Tick marks on the axes should also have clear labels. No overlapping or too small to read text, and numbers should be formatted so they are easy to interpret. For example, \$7 million or 7,000,000 is better than 7000000 and $2.3\cdot 10^{-5}$ is better than 0.000023. Figure 10 is missing the y axis entirely and its x axis is barely labeled.
{width=60%}
# 9. Low contrast and poor color choice
Contrast refers to the difference in brightness between two colors. To make plots easy to read, the contrast between the text/graphic and the background should be high. This means if the background is light the text should be dark and visa versa. [**Webaim**](https://webaim.org/resources/contrastchecker/) has a contrast checker that allows you to see if your foreground and background colors are sufficiently different so as to to be easy to read.
Even if the foreground and background colors are high contrast, you also need to make sure colors used in the plot are distinguishable from each other. This is especially important if you are using color to communicate data, as in Figure 11. If you aren't sure what colors to use, or you want to make sure that the colors are visually distinct for people who are colorblind you can use the [**Color Brewer**](https://colorbrewer2.org/) website for color palette recommendations.
{width=50%}
# Questions
1. Categorize the following plot types as Comparison, Composition, Relationship, Distribution, or a combination of them.
a) bubble plot
b) stacked bar plot
c) density plot
d) radar plot
e) area plot
2. Give an example of a time unequal sized groups or bins that would help people better understand the data, instead of misleading them.
3. List two color palettes, one for July 2 and one for July 20, that would make the maps from Figure 3 less misleading.
4. List 3 reasons, other than presence of mask mandate, why the countries in Figure 4 might have the COVID-19 case rates they do?
5. For Figures labeled with Question 5, list the type of plot and all of the problems each visualization has.
{width=80%}
{width=80%}
