64-Bit JMP for Linux

The world's largest privately owned software company, SAS, was cofounded in 1976 by Dr James Goodnight and John Sall. They continue to run the company as CEO and Executive Vice President. Sall is also chief architect of SAS's statistical discovery software called JMP (pronounced “jump”), which he invented for the Macintosh in the late 1980s. It is a desktop statistical analysis program using exploratory graphics to promote statistical discovery. JMP was released for Windows in 1995 and has been available for 32-bit Linux since 2003.

SAS's version 6.1 release of JMP later in 2006 harnesses the vast computational power of 64-bit Linux, which is not only exciting news for JMP and Linux, but is also a milestone in statistical computing.

To understand the importance of a 64-bit version of JMP, let us contemplate the purpose and history of statistical analysis.

Ultimately, the purpose of statistics is to make sense out of too much information. For example, the only possible way to digest the results of the United States census data every ten years, with its dozens of measurements on 275-million people, is by reducing it to statistical conclusions, such as the average household income by county and median age by city or neighborhood. Nobody could possibly look at the raw census data and draw a meaningful conclusion beyond “the United States has a large, diverse population”.

The problem is that there are hundreds and thousands of statistical measures—in fact, SAS has already spent 30 years extending and refining its analytical capabilities and doesn't see any end in sight. Learning what techniques to use for which real-world situations can take years, and developing the insights to proceed effectively from raw data to knowledge can take a lifetime. This is what led John Sall to develop JMP in the late 1980s. Inspired by the way the Macintosh made desktop computing accessible to a whole new audience by introducing a graphical user interface, Sall realized he could make statistics accessible to a wider audience by making the analysis process visual.

Comprehending the meaning buried in pages of statistical test results—p-values, standard deviations, error terms, degrees of freedom and on and on—is a mind-boggling task even for experts, but Sall knew that just about anyone could look at a well-drawn graph and understand things about his or her data. JMP always leads every analysis with graphs, so that researchers needn't waste time poring over statistics when those graphs make it intuitively obvious whether they are on the right analysis path or not. JMP also groups related analyses together and presents them in the order a researcher would need them in the course of a sound data exploration process. Researchers do not have to wrack their brains to remember which procedure might be helpful next. Instead, JMP provides the tools that are appropriate at each stage. Further, all of JMP's graphs and data tables are dynamically linked, so that users can point and click to select points in a graph or bars in a histogram and instantly see where those points are represented in all other open graphs and data tables.

Setting aside for a moment what it takes to understand statistics, consider what it takes to calculate statistics. For a researcher to compute a standard deviation on thousands of observations using only a pencil and paper could take weeks or months.

When he created SAS in the early 1970s, Jim Goodnight's idea was to store all that data in a file and then write procedures that could be used and reused to compute statistics on any file. It's an idea that seems ludicrously simple today, but it was revolutionary at the time. The agricultural scientists using SAS could perform calculations over and over again on new data without having to pay for computer scientists to write and rewrite programs. Instead of taking weeks, these computations took hours. Fast-forward 30 years, and modern statistical software can do these calculations on hundreds of thousands of rows, instantaneously.

When it took months to compute simple descriptive stats, researchers often didn't get much further before they'd burned through their grant money. Now that the basics take seconds, researchers can dig much deeper, and thus the science and practice of statistics have evolved along with computing power.

For the last decade, desktop computing has been built on operating systems such as Windows, Linux and Mac that rely on 32-bit memory addressing. Accordingly, desktop applications have operated within the computational limits implied by this architecture. In practical terms, this meant statistical programs like JMP that load the entire dataset into RAM before performing any computations were limited to about a million rows of data. They couldn't handle the large-scale problems confronting researchers today. Geneticists are probing 3-billion base pairs of DNA. Semiconductor manufacturers are squeezing millions of transistors onto ever-tinier chips. Pharmaceutical companies comb through thousands of potentially therapeutic properties on countless known and theoretical compounds.

Figure 1. JMP Main Window Showing a Distribution for Genetics Data

Figure 2. Hierarchical Cluster Diagram or “Heatmap” Showing Gene Expression, Protein and Metabolite Production Patterns

Dr Richard C. Potter, Director of Research and Development for JMP Product Engineering, was responsible for porting JMP from Macintosh to Windows and later from Windows to Linux, in collaboration with Paul Nelson, lead Linux System Developer. Potter says:

Figure 3. Bivariate Fit with Nonparametric Density Contours Representing Density Levels

Figure 4. Bivariate Fit Volcano Plot for Comparing Gene Expression