1. I don’t recall seeing survey questions that have positive responses before (left of) negative responses. It seems backwards. But definitely when plotting numerical values it is customary that negatives are to the left and positives to the right. Disagree is negative and agree positive, so by this logic, it seems to me…
2. I’ve seen several tutorials (not just diverging stacked bars) that try to overlay two charts. It seems not worth the trouble, because preserving alignment of the two charts and their axis scales can be problematic. Using one chart for this exercise is as difficult as multiplying some values by -1.
3. I posted a quick sample chart at
http://peltiertech.com/images/2013-09/SampleDivergingBars.png
The vertical axis may help the reader see where positives and negatives diverge.
4. The chart I posted was made using a prototype of a new Excel add-in I’ve been working on. It uses similar formulas to split the neutral responses and arrange the data so the legend entries are in the correct order. Smoke and mirrors are the biggest part of Excel charting.
5. As a survey taker, I don’t like having to choose between Agree and Disagree when I don’t care. That’s why I said a survey should have a neutral option. I don’t see why you would need more than 5 responses, though. We’re measuring subjective feelings, not atomic properties.
6. The prevalence of this topic in the blogosphere lately inspired me last week to write up some thoughts on presentation of survey results, including these diverging bar charts:
http://peltiertech.com/WordPress/charting-survey-results/

Jon,
Thanks for reading!
1. In other fields, yes, I’ve definitely seen positive responses to the right instead of to the left. Jozsef Hornyik arranged his entry this way (http://bl.ocks.org/CodeXmonk/6262477). I put positive responses to the left because that’s how I structure surveys (the check boxes for strongly agree are to the left). I follow Dillman’s (2008) suggestions for survey design. Although as I write this, I realize 2008 was a while ago, so perhaps there are newer suggestions for survey design? I should see if Dillman has a new book out…
2. Yep, you’re right, these charts can be made with the floating white bars technique too. I had already shared my spreadsheet for that technique here (http://emeryevaluation.com/2013/08/07/chart-colors/) and Stephanie Evergreen blogged her solutions to the dataviz challenge here (http://stephanieevergreen.com/diverging-stacked-bars/). So, the purpose of this post is to share an alternate solution.
3. I’m not sure what you mean about the vertical axis.
4. Yep, if I had a neutral category, I could split it. A bunch of the contestants did this, and David Bonachea even shared a screenshot of his Excel formula for splitting the neutral category (http://annkemery.tumblr.com/post/60407970083/check-out-david-bonacheas-entry-to-dataviz).
We’re going to have to agree to disagree that “any good survey” has a neutral category. The “how many points on a scale are best? 4, 5, 7, 9?” debate has been going on forever in evaluation and I have no desire to jump into that today, ha! 😉
Thanks! Ann

David,
This is why I love blog discussions. You’ve brought up excellent points.
1) Yes, absolutely, every single adjustment to a chart changes what part of the data is emphasized or de-emphasized. This is part of the evaluator’s job as the storyteller, right? We guide stakeholders through a sea of numbers and help them pick out the most important information. In a traditional stacked bar chart, the reader can make part-to-whole comparisons; that is, strongly agree vs. agree, or strongly agree vs. agree/disagree/strongly disagree lumped together, or strongly agree/agree vs. disagree/strongly disagree, etc. In contrast, diverging stacked bar charts intentionally focus the reader’s attention on collapsed categories; that is, just the strongly agree/agree vs. disagree/strongly disagree finding.
2) Again, yes! The reader’s comprehension is central. That’s exactly what this evaluation and dataviz work is all about. I’ve had great success with clients understanding diverging stacked bar charts. The captions above the charts are crucial. I used to label charts as “chart 1” or “chart 2” but clients really benefit from the captions. (Unfortunately, not all charts are successful… Social network maps haven’t gone over as well: http://www.visualisingdata.com/index.php/2013/05/conquering-the-dusty-shelf-report-data-visualization-for-evaluation/)
My goal in facilitating these dataviz challenges is to expose my readers to a variety of chart types and help them think through the pros and cons of each style, as you’ve done. Stay tuned for a future post, “1 Dataset, 10 Different Charts,” that shows how the same results can be shown at least 10 different ways and discusses the advantages and disadvantages of each chart type.
Ann