Charts And Diagrams

Ryan Home » PPOL225 » Lectures » Charts And Diagrams

Page Summary
This page ….

Basics

Making quality charts is part art, part science. It is all skill, patience, and experience. Do not underestimate the value of investing in developing the skill. A great chart can kill — though often without people realizing it. On the other hand, most charts are junk: poorly conceived, poorly executed — they do a bad job of communicating, they waste people's time, and they confuse rather than enlighten. But no one really notices, because they are so used to it.

ADVICE: Take the time. Be disciplined. Experiment.

RULE 1. Every chart should stand on its own : be completely self-documenting

Always assume that a chart will be reproduced out of its original context. It should convey the same information ripped from context as it would in its original place on the page. This means

Chart title must be descriptive, accurate, and standardized
All axes must be properly labeled
Dates: as of, when made, last revised
Data sources, other references, and any disclaimers or caveats should be integral to the chart.
Who made this chart?

RULE 2. Know what a chart is trying to say. Most charts have a single message — make sure you know what it is and that it is successfully communicated.

Distinguish, then, the use of charts for data exploration and their deployment as a means for visually conveying quantitative information.
- A sometimes useful heuristic: can you write a straightforward declarative sentence as a caption?
Although we often make charts while exploring data, charts that we make for reports and publications are designed to show something we have found out. In general, our design strategy is to direct attention to what matters and avoid drawing attention to that which does not.
Lying: as a rule, you shouldn't use charts to lie, but if you do, you should be in control of the lies you are telling. See below.

RULE 3. Pick the right chart type for the job.

Proportions that add up to 100 : pie chart.
Frequency distributions of one variable: bar/column histogram.
Correlation between two variables: xy scatterplot
Trend over time: line graph
Three dimensions
- Two IV + frequency : sometimes a "3D column chart"
- Two IV + magnitude : sometimes a "bubble chart"
See [http://extremepresentation.typepad.com/files/choosing-a-good-chart-09.pdf] for a pretty good "all on one page" version of this.

RULE 4. Be in control of the visual communication

Meaning of distance
Meaning of area
- White space
- Solid color space
- Attribute spaces
Zeros, minima, maxima
1+1=3 Problems and other forms of visual noise
CRAP
Colors and other graphic elements should be robust in the face of photocopying whenever possible.
Avoiding chart junk and eye-candy

Background and Review : Types of Variables

Nominal or categorical
Ordinal or ranked categories
Interval — distance means something
Ratio — zero means something

Background and Review : Distribution

Background and Review : Frequency

Chart Types, Indications and Contraindications

http://freshfreestuff.net/wp-content/uploads/2009/09/excel-pp-chart-template.jpg

Bar and Column

When to Use

Counts or levels in categories for nominal (categorical) data. Ordinal, interval or ratio data if grouped into categories or classes.
Relatively small number of categories.
When the story is the shape of a discrete distribution (over categories or time)

When not to Use

Large number of categories
When the story is relative proportions or fractions of a whole (see pie charts)

Proportions vs. Absolute Numbers

Source Dual Career Report

Variants and Specialty Uses

A stacked bar (or column) chart is used to show both the total in a category and the size of the subcategories of which it is composed.

http://www.roseindia.net/tutorialfiles/1305.stackedChart.gif

Population Pyramid

http://thinkgeography.files.wordpress.com/2008/10/population_pyramid.png

Line Charts

A line chart typically shows values of a dependent variable on the vertical axis with an ordinal variable along the horizontal axis. The horizontal variable may be discrete (that is, measured only at specific instants, not continuously), but the connection of the data points typically implies something that can at least be thought of as if it were a continuous process.

http://webgis.wr.usgs.gov/globalgis/images/tutorials/chart_line.gif

Each line on a line chart consists of one data series.

A line chart can show data markers or leave them out, depending on whether we are trying to emphasize individual points or the trend.

Pie Charts

A pie chart is used to show the proportions or parts of something. The area of each slice of the pie is proportional to the share one data value represents vis a vis the total.

The data series for a pie chart must be mutually exclusive and exhaustive of a category or collection. Another way to think of this is that the values have to add up to 100%

Although popular (especially in USA Today 3D exploded view form) in popular media, pie charts are less than stellar tools of visual communication. The human eye/mind is not so good at comparing items of different shapes and less adept at perceiving area than length and we can mentally a set of objects by length better than we can order a set of slices by angle.¹

Pie Chart Example

Let's consider what is good about this pie chart.

The "whole" is easy to conceptualize.
Ordering slices by region turns it into a highly abstracted world map.
Color coding by region, using same colors with region names.
Changing label font sizes relative to size of slices.

http://upload.wikimedia.org/wikipedia/commons/archive/d/de/20080512031830!World_population_pie_chart.JPG

We can also do some interesting combination charts using Pie charts.

http://www.stanford.edu/group/gender/ResearchPrograms/DualCareer/DualCareerFinal.pdf#page=15

How to make this chart with Excel

Scatter Plots

For a scatter plot, aka x-y plot, we use two data series at once. One defines the horizontal position of charted points, the other the vertical.

http://enthusiasm.cozy.org/wp-content/uploads/2009/12/GDPCapitaVSGini.png

http://soc125ppol225.wikispot.org/Excel_Scatterplot

Sometimes it is hard to distinguish a scatter-plot from a line chart. Take this example where we have two sets of x-y data, both of which have the same X-values. Make sure you see this before you continue.