I'm listening to Jon Weisz, marketing director for JMP, talk about data-driven story telling. He's explaining the differences between statistics and data visualization. Statistics are often used to confirm a hypothesis, he says. Data visualization is used for exploration. Jon's defining models as approximations: "We're attempting to experiment and determine factors and effects." Often, when statisticians are working with models, they're attempting to answer the question, "What if ...."
Jon is walking through a simple example in the JMP profiler, to introduce some very basic visualization techniques. The questions he's asking are, "What time will I be home from work? And what affects the arrival times?" With just a few variables to account for changes in arrival times, he ends up showing that his assertion, "I'll be home from work before 6:00," is right only about 78 percent of the time.
Moving on to a more complex example, Jon is looking at U.S. housing data from Freddie Mac. He pulled the public data set directly into JMP and is now using the Graph Builder, and showing rules within the product that can tell you how best to display the data.
He's asking the questions, "Was there really a bubble? And did it really pop?" The graph for California shows a clear peak in prices and a huge drop off. In Michigan, there's less of a bubble and more of a gradual decrease.
When looking at graphs for all 50 states side-by-side, you can see that the real estate markets have behaved very differently in each state. Realizing this, Jon decides to cluster the states based on real estate growth data. Using JMP to determine the optimal number of clusters, he separates the states into six groups. The clusters show clearly that AZ, CA, FL and NV experienced the crash in the housing market most dramatically.
Next, Jon is moving the data to SAS Enterprise Guide. "Enterprise Guide is very complementary to JMP," says Jon. JMP's purpose is all about discovery. What Enterprise Guide presents you with is a flow chart. You have to know the answer to get the flow - which you found in JMP. "Once you know the proper analysis flow, then you go into EG and document it."
The final report Jon shows of the state housing market data clustered into 6 groups is very easy to digest at a quick glance. He asks: wouldn't it be nice for these organizations to provide the data in this format on their Web sites instead of just as raw data files or Excel spreadsheets? Wouldn't it be nice for you to do the same in your company?
