Innovative, yes. But as useless as an ashtray on a motorbike without a training course of at least a week. BTW, all pictures can be made with Excel, albeit it will take some time to do it. Only the grey connection lines and these dots on the "circle graph" (what's the proper name for this ? Even pie charts are more clear than this) can only be hand picked - but they are confusing. Frankly, Zach, I don't see the point you want to make. But if this is your fiercest competitor, can I buy shares in Juice Analytics? [wink]

Zach:

Thanks for an interesting post.

Ms. Shen-Hsieh's states "In the most simpliest terms, if you think of Excel, you’ve got an x-axis and y-axis, so you basically have two dimensions of data that you are looking at at any one time. A complex problem is going to have lots of dimensions hat you need to consider. That’s what visualization gets you. It gives you a much broader context besides a flat, tabular, x-y".

A few points:

1. All of the Visual I/O examples shown in your post are 2-D. These plots are "flat". Does Ms. Shen-Hsieh think that makes them bad?

2. Ms. Shen-Hsieh makes some breath taking leaps when she jumps from: 1) talking about Excel being 2 dimensional to 2) "..complex problems.." having "...lots of dimensions that you need to consider" to 3) "that's what visualization gets you". This leap from Excel is only 2D to 'visualization" is the answer needs to be parsed to really see what she is claiming.

3. She is partly right, most of our problems are multi-dimensional. The dimensions, however, are in the data, not in some vauge "visualization" concept. The solution is to do effective multi-dimensional analysis.

4. She has made 8 standard 2-D plots of the demonstration data set. These are 8 data slices out of potentially dozens. Are these the most important, the best for discovering the underlying data structure? We don't really know.

5. As Ms. Shen-Hsieh shows in the example, we need to view our multivariable data through multiple 2D windows. It's not the 2D plot tool that is the problem, it's how we are using the 2D tool.

6. Why blame Excel? Do we blame the carpenter or the saw for poor woodworking?

7. Effective data visualization requires the user to query, slice, dice data into multiple 2D windows to see/ understand the underlying structure in the data. Relying on just one 2D plot won't work.

8. Excel, like other charting tools, displays data in 2 dimensions at a time. However, we can add additional factors by using techniques like Trellis - lattice displays. S-Plus, R use mutiple 2D plot arrays (trellis / lattice analogy) to systematically display the relationship between three-four factors at a time. Factors 1 and 2 are "conditioned" by Factors 3 /4 and plotted in a 2D plot array. While each plot is 2D, the multiple plot array allows the user to see the role of Factors 3 - 4 on Factors 1 and 2..

9. Excel users can create vertical and/or horizontal panel charts to simulate trellis like displays that can help in multi-variable analysis.

10. The problem is not in 2D charts or Excel. The problem is how users Excel (or any other tool) and 2D plots to investigate a multi-dimensional problem.

11. An Excel user could do the same thing as Ms. Shen-Hsieh if they have their data well structured and have good charting skills, and use their imagination to ask the right questions of how does X relate to Y, Z, Q T and other factors? Those who ony look at X versus Y won't benefit with any tool.

12. An Excel user can make Trellis like (Cleveland, Robbins) or small multiple like(Tufte, Few) charts. I have a number of examples of horizontal and vertical panel charts on my site ( http://processtrends.com) that your readers can download and try out on their own.

13. A skilled investigator (good statistical and data visualization background) with Excel can beat an unskilled investigator with a "'powerful tool" 90 out of 100 times.

14. Let's focus on improving investigators' skills by showing users how to address multi-dimensional problems. Training in muti-variable analysis will be useful no matter what "visualization tool" is used.

In the carpenter example, let's make sure he has the fundamental skills (training if necessary) before we buy him expensive that new (costly) power tool.

Kelly O'Day
http://processtrends.com