JMP 13 Preview: Association analysis for analyzing sparse categorical data
Aug 24, 2016 10:57 AM
Association analysis is a powerful way to look at categorical data that has a lot of sparsity, and it’s used not only by market researchers but also reliability engineers.
Do you find it fascinating to look at the items in other people’s grocery carts at the store? I do. In mine, you will often see orange juice, whole milk, Fuji apples and blackberries. And if I’m buying grape tomatoes, it’s likely I’m also buying parmesan cheese and basil (to make pasta with pesto for dinner that night).
Wouldn’t it be fun to analyze this kind of data – what people tend to buy together? For the first time, you can do this in JMP.
Sometimes called market basket analysis, it was originally used for analyzing scanner data from cashiers. In JMP 13, it’s called association analysis, and it was a highly requested feature.
How would you use the results of association analysis? It can help you to assist customers in finding what they may need and to sell more through effective placement of items.
“If you see that someone buys A, B and C, then the results can tell you that you should make it easier for that person to find X, Y and Z,” explains Melinda Thielbar, the research statistician developer who worked on this new capability in JMP.
Association analysis is a standard technique used in market research. But it is increasingly being used by reliability engineers to find out such things as what parts in a machine tend to break down together.
“Association analysis is a really good way to look at categorical data, especially if it has a lot of sparsity,” she says. “You may have a machine with thousands of parts, and there may a few clusters of parts that tend to break down at the same time. That’s a difficult pattern to discover without a statistical model.”
The analysis produces an SVD plot, which Melinda says she has not seen elsewhere. “It allows you to sort through all of the items that appear together as if you were doing a principal components analysis on a very large data set that describes which items appear together,” she notes.
There’s been a lot of excitement about the entire consumer research platform from early adopters, with particular interest in MDS (multidimensional scaling), association analysis and choice analysis.
“I’d like to see our customers using consumer research more, and in more areas,” Melinda says.
One of the best parts of working on this area of JMP is that she gets to collaborate with a lot of other developers. Melinda will be at Discovery Summit next month presenting a poster on new features in choice modeling in JMP and JMP Pro, as well as talking with customers informally about how they use the software to solve their consumer research problems.
For more information on what's coming in JMP 13 and JMP Pro 13, stop by our preview site.
Can you guess which of the Topics above closely resembles Melinda Thielbar’s own grocery cart when her significant other is out of town? I’m not telling!