StATS: Seminar notes, Creating More Effective Graphics (March 24, 2006).
I attended a web seminar, "Creating More Effective Graphs," taught by Michael O'Connell and Naomi Robbins and sponsored by Insightful Software, the makers of S-plus.
Michael O'Connell opened the talk with a brief description of Trellis graphics. It's hard to capture exactly what Trellis graphics are in words, but there is a nice web page by the developers of this product at Bell Labs (stat.bell-labs.com/project/trellis).
Dr. O'Connell also mentioned Graphlets. Graphlets are small java programs created by S-plus to produce highly interactive graphics that you can place on your web page. You can see some examples at
- S-plus Graphlets Gallery. Insightful Software. Accessed on 2006-03-24. (Model, Descriptive) [Excerpt] Here are just a few examples of S-PLUS Graphlets. Some of the examples include the S source code used to create the Graphlet, which you can use as a basis for creating your own Graphlets. www.insightful.com/products/graphlets/gallery/graphlet_gallery.asp
Naomi Robbins talked about effective graphs. She is the author of a textbook, Creating More Effective Graphs, and has a web site,
- NBR Offers Businesses Consulting and Training on Better Charts and Graphs. NBR Consulting. Accessed on 2006-03-24. (Model, Descriptive) [Excerpt] NBR, a leader in consulting and training for graphical data presentation. www.nbr-graphs.com
The choice of a graph form has a tremendous impact on your understanding of the data as well as your audience's understanding. One graph is more effective than another if the reader can extract quantitative information more rapidly. You should think about this definition of "effective" because it is not what most people would use. A commonly held conception is that a graphic is more effective if it grabs your attention more rapidly. In some contexts (e.g., advertising) this might make sense but I believe that Dr. Robbins definition is preferred in most applications.*
Let me insert a personal comment here: Leland Wilkinson has an interesting take on this. He splits the types of tasks in a graph into two different categories, pattern recognition (discovering trends and associations), and table lookup (computing a numeric value from a graph). Some of the things that help with table lookup (grid lines, for example) will interfere with pattern recognition.
Dot plots are an improvement over traditional bar charts, and this improvement is even more apparent when you use a clustered bar chart. By placing the dots side by side on the same scale, you see patterns more easily than with side by side bars. The panel display, placing several related graphs in a rectangular grid, is also an improvement over clustered and/or stacked bar charts. I don't have access to the graphs she used, but I will try to find or create comparable plots for this and all the other examples described here.
Dr. Robbins talked about a "month plot," also commonly called a cycle plot. It uses the S-plus command, monthplot. It is very useful way to display time series where you want to look at patterns across two different time scales. For example, you want to see the trend in sales over months, but you also are interested in variations in sales across days of the week.
She showed a polar plot for a library survey where respondents were asked how the library was doing on a range of scales, what they thought was the minimum acceptable level and what was a desired level. An important consideration is when the perceived performance was lower than the minimum acceptable level or higher than the desired level. The polar plot used a cryptic color system (red means you failed to meet minimum acceptable level on average, for example). By plotting the minimum and desired values as a range, and placing a dot at the perceived level, you could easily see when the library was deficient (perceived level was lower than what was considered minimally acceptable, on average) and when the library was exceeding expectations (perceived level was higher than the desired level, on average).
She also had a nice display of a 2^4 factorial design. I'm not sure what the name is for the type of plot she displayed, but I will find an example when I have time.
The choice of color in graphics is important, and you need to keep in mind that some of your readers are color blind. There is a nice test for color blindness, and it includes a negative control (an image that even color blind patients can discern). I don't have Dr. Robbins' image, but I think I found a copy of it on the web (www.geocities.com/heartland/8833/coloreye.html)
Dr. Robbins used these two colors in most of her plots because this choice of contrasting colors that will work well even for colorblind individuals.
One of the questions at the end of the seminar was about how to best display data that has a three way interaction. It turns out that Richard Heiberger has a book which discusses interaction plots in Chapters 12-13. I'm guessing that the title of the book is Statistical Analysis and Data Display [BookFinder4U link], which was co-authored with Burt Holland, but I don't have this book in my library, so I can't say for sure.
[Update, April 27, 2006] I got a nice email from Dr. Robbins. She offered a minor correction (the title of her book is Creating More Effective Graphs, not Creating Effective Graphs) which I just fixed and confirmed that the Heiberger book "Statistical Analysis and Data Display" is indeed the book she was referring to.
This page was written by Steve Simon while working at Children's Mercy Hospital. Although I do not hold the copyright for this material, I am reproducing it here as a service, as it is no longer available on the Children's Mercy Hospital website. Need more information? I have a page with general help resources. You can also browse for pages similar to this one at Category: Descriptive statistics.