Small Multiples – Abortion Data 1980-2003

Jorge used some U.S. Census Bureau data (original Excel file) to visualize the abortion ratio as a small multiples chart.

I like this chart as I am a big fan of small multiples displays. They help us to understand the nature of multi-dimensional data.

Jorge asked for suggestions to improve his chart and I came up with these:

  • The abortion ratio is calculated as the number of pregnancies that end in an abortion per 1000 pregnancies. I would put the definition of the ratio in the title or a footnote (Number of abortions per 1,000 live births). Most places where I have seen this data presented tend to do it in this way too, although my preference would be for a % format.
  • I think that the data for Race, Marital Status and Age Group are actually 3 different stories. It took me several scans of the whole chart set to realize this. I would add some white space between the Age group charts, the Race and Martial Status charts, to make it clear that are comparing different variables.
  • Jorge shows an integrated line chart for Marital State and for Race but a small multiples chart for Age Groups. For a more consistent chart reading I would use small multiples charts for all variables.
  • The axes of the Percent Distribution bars are not labeled, and have no scale. This coupled with varying meaning of the color encoding is confusing. The percentage distribution data for the Age < 15 lies around the 1% level and so cannot be seen on this scale even though data does exist. We also cannot encode the way this measure also varies with time over the same period. In fact we miss the fact that for women over 25 years the percentage distribution is actually increasing.
  • I’m not sure if you need to show the all age groups in the Age group charts as light Grey lines. I know that you want the reader to compare the current age group in the context of the other age groups but this exactly what the small multiples are for anyway.

I like Jorge’s idea of integrating the abortion ratio and abortion percentages, but I found having another group of small multiples provides a more consistent view. If we don’t have the percentage distribution numbers then we can easily draw the wrong conclusions about the data.

Here my small multiples chart for the Abortion by Age Groups:

image

 

Update 6/22/2008: Jon Peltier posted on Re: Abortion Ratios 1980-2003 and Interactive Multiple Line Chart some nice Excel techniques to create interactive charts to analyze the Abortion data set.

More Information per Pixel!

In my last post I suggested some chart selection rules as an alternative to Godin’s Silly Rules for Great Graphs. Jerome commented:

[…] on a slide, you want to convey one message. your graph must NOT carry any information that can be interpreted differently than the point you are trying to make. the corollary is that in virtually all cases, you should display as little data points as possible: 1 if possible, 2 but no more than 3. If you need more than 3 data points, use handouts. […]

which is very much in line with what Seth Godin said in his famous post about Chart Rules:

No, the reason you put a chart in a presentation is to tell a story. A single story, one story per chart

Why should a presentation display as little data as possible? Why should a slide contain only one chart? I demand More Information Per Pixel. Why not have a data-rich chart in a slide – no, even a couple of charts to support my message?

My friend Rolf Hichert has a totally different design philosophy.

Components of good presentations slides:

  • A clear message
  • A clear title (should be a complete sentence, including units like K$)
  • Each slide to conveys only one message
  • More tables and charts to support the message
  • Choose the right chart type
  • Use arrows, color etc. to highlight the message

Look at this sample taken from Rolf’s web site:

image

The slide has a clear title that conveys one message: "Further positive Development in Frankfurt, Vienna and Graz – Action needed in Lausanne and Linz".

The slide contains small multiple charts to support the message, where Rolf has chosen a line chart to emphasize trends or patterns. The problematic regions mentioned in the title are colored in red; those that made the CEO happy in green.

Chart Rules, As Simple as Possible, But Not Any Simpler!

In chart design it’s good to make things simple, but you certainly should avoid oversimplification. As Einstein said:

"Things should be made as simple as possible, but not any simpler"

Seth Godin presented in his blog The three laws of great graphs:

1. One Story
2. No Bar Charts
3. Motion

Effective chart design rules are simple, but reducing it to this set of 3 rules certainly is an over-oversimplification.

Particularly rule 2 is flawed. Seth details rule 2:

"NO BAR CHARTS
Bar charts are dramatically overrated, primarily because they’re the first choice in many graphing programs.

The problem with bar charts is that they should either be line/area charts (when graphing a change over time, like unemployment rates) or they should be a simple pie chart (when comparing two or three items at the same scale).

Jorge, Kaiser and Jon already wrote some critical posts about this rule, where Jon suggested to replace rule 2 with

Choose Chart Types Intelligently

We are working tightly together with Stephen Few on a new product that helps business users creating effective charts with Excel and are therefore we are quite familiar with Stephen’s design principles.

Its an easy to learn set of rules

1. Determine the relationship you want to display

Relationship Sample

Value Comparison

Sales in different regions

Ranking

Best selling products

Time-Series

Sales in the last 12 months

Part-to-Whole

Market shares

Deviation

Revenue Actual vs Budget in the last 12 months

Distribution

Support response times

Correlation

Relationship between employee’s heights in inches and their salary

 

2. Determine if you want to emphasize individual values or the overall pattern
3. Determine the chart type

Relationship Encoding Method 

Value Comparison

Bars and Columns

image

Ranking

Bars and Columns

image

Time-Series

Lines to emphasize the overall trends or pattern

image

 

Points connected by lines to slightly emphasize individual values

image

 

Columns to emphasize and support comparisons between individual values

image

Part-to-Whole

Bars and Columns

image

Deviation

Lines to emphasize the overall shape of the data

image

 

Points connected by lines to slightly emphasize individual data points

image

 

Bars and Columns to emphasize individual values

image

Distribution

Columns to emphasize individual values

image

 

Lines to emphasize the overall shape of he data

image

Correlation

Points and a trend line in the form of a scatter plot

image

 

Armed with this set of rules you would rule out Seth’s pie chart, and use the bar chart in the appropriated business context.

Win Lose Charts – English Premier League 2007 – 2008

With hindsight, it’s easy to look at it with hindsight” – Glen Hoddle

Little did we suspect midway through the season that the Premier League would turn into a 2 horse race. Poor starts by both Man Utd. and Chelsea were transformed into a consistent run of good form which would extend the drama to the very last day of the season.

With one game left to play, the top two were level on points with Manchester having a superior goal difference. If Chelsea could achieve a better result than Manchester, then they would clinch the league in the final game…

The English Premier league currently has 20 teams requiring a total of 380 games per season. The results of sporting leagues are usually displayed in a league table format. During the course of a season we see teams occupy different table positions. The standard table format however leaves out the important historic story of the league.

This is how soccerstats.com shows the final league table.

image

 

 

 

 

 

 

 

 

 

 

 

 

 

The final 8 games are summarized as a colors encoded Win Lose chart. The table found at the web site above provides helpful links to each team’s performance and a snapshot in time regarding table position for each month of the league. It is also possible to rank the teams with respect to their home or away performance and overall is a very powerful tool for analyzing statistics of the league. Some historical information is also given with the results for the last 8 games being displayed in the last column.

Statistics are like miniskirts; they give you good ideas but hide the important things.” – Ebbe Skovdahl

However, there are a number of things about the table in that I think could be improved and all of them are centred on the “Last 8” column. The table author is attempting to describe win-lose information graphically and has dedicated approximately one quarter of the width of the table to it. With a quarter of the table dedicated to it, we should expect more than 8 out of 38 games to be described. The author has chosen to invent a chart type which encodes the results using colored squares with more empty space than data. The whole season can easily fit into this width if we choose an appropriate display technique such as a win-lose chart.

My least favorite aspect of the “Last 8” column is the non-standard use of a 1 dimensional plot. I come from a physical science background and so I am used to diagrams telling me certain things in certain ways. One of these things is (in the absence of an axis or some other visual guide) that time should go from left to right! In this 1D plot the knowledge of which way time is going is fundamental. The links at the top of the same table used to break the table down into months, go in chronological order from left to right. This sets up my brain to expect the rest of the table to behave in the same way. So why don’t the last 8 games do the same? Unless you actually have some knowledge about how the season ended, you might not actually realize that time is going from right to left. I happened to know that Liverpool and Arsenal didn’t lose their final games of the season. This made me double check what the table was actually telling me. Without some knowledge of how the Premiership ended I would have interpreted the information presented wrongly. The fact that time is going from right to left isn’t wrong. The fact that the rest of the table is telling me to expect it to go from left to right is.

“Well, Clive, it’s all about the two M’s. Movement and positioning” – Ron Atkinson

So what improvements could we make to the standard table format to get more information into the table? To tell the story of the 2007-2008 league we need to include the historical context.

The Excel table below uses sparklines to summarizes the season for each team in terms of relative positions in the table and a win lose chart for the entire season.

image

 

Now we get a feeling for the true drama that occurred during the season. The Position column shows how the each teams position changed over the course of the league. Those of us that followed the 2007-2008 Premiership will be able to look at the performance curves and remember the situation for a given time. For example:

  • The lowest point of the season for Man Utd ,in every sense, was when they lost to Man City
  • The departure of Jose Mourinho after 8 games was a catalyst for change in Chelsea’s fortunes

“I never make predictions, and I never will” – Paul Gascoigne

It is interesting to directly compare the performance of teams together. Adding an interactive Bumps Chart in allows us to see how Man Utd and Chelsea faired over the season. To compare two team click the sparklines in the ranking table or click the data label on the Bumps Charts.

image

 

In the next post I will talk about the techniques used to produce the ineractive ranking table within Excel. Until then why not check out a live web version, published from Excel to our website.

Information visualization: take the red pill

“To clarify, add detail”, says Edward Tufte.

A richer, more detailed picture, is a solid foundation for your decision-making processes. But to add detail you need a higher resolution display device (be that a computer screen or a sheet of paper), as we saw in the previous post.

Now, regarding the use of resolution, you have a “take the red/blue pill” kind of choice:

  • You take the red pill, accept Tufte’s advice and you’ll get more insights from your data
    • You take the blue pill, buy the stuff most vendors want you to buy and you stay under the illusion of the “professional looking chart”;

Let me detail the blue pill option. According to a large majority of vendors, we should get higher and higher resolutions, yes, but only to admire how eye-catching their products are, how well rendered, even if they display less and less actual data points. For the untrained eye, they may look like a Ferrari, but there’s a Tata underneath.

In reality, vendors and knowledgeable users have different agendas. Users want higher screen resolution to accommodate more data, while vendors want it because it makes they products look… “cool”? Apparently, in the mass market, form and function are strange to each other.
Let me exemplify the problem with a typical pie chart. I already gave my two cents for the never ending discussion around the sins and virtues of pie charts, so I will not do it again soon.

What I want to emphasize is that you can’t have more than five or six data points in a pie chart, but if you add texture to make it glow you will need to remove some data points and enlarge the chart. You need more space (= larger charts) in order for texture to be noticed, and there goes better (for efficient) information visualizations.

Unlike scientific visualization (that usually creates digital models of objects), information visualization focus on abstract concepts, like “inflation rate” or “market share”. You can’t add texture to market share. A chart is a “metaphoric space” where some objects (points, lines, rectangles) stand for an abstract concept, and we infer something from their relative positions in space.

So, you have a large, high-resolution computer monitor and also a high end color printer. You have the option between texture and detail. You can’t have both. Choosing detail you are focusing on the data and how to squeeze the juice out of it. Choosing texture you are adopting a marketing posture whereby you are not selling insights, you are selling yourself (it is an option, and some times you’ll need it). Or worse, in your naivety, you believe that information visualization is just a glowing 3D pie chart. Believe me, it is not.

So, what color is your pill?