Course Highlight
Two SAS Education JMP Courses Take the Mystery Out of Design of Experiments
Design of experiments is a concept that has been around for decades and can be useful in any given industry, according to SAS Education JMP Instructor and Curriculum Developer Mark Bailey. Bailey has worked for SAS for almost a decade and has been using JMP Software to solve DOE problems for even longer. STR spoke with Bailey recently about JMP and DOE, as well as two of SAS Education's most popular courses dealing with this topic -- JMP Software: Classic Design of Experiments and JMP Software: Custom Design of Experiments."Everyone should think about using DOE in any situation that demands a decision or an action."
STR: Before we talk about these two JMP courses, what is DOE, how is it used, and who does it affect?
Design of experiments, or DOE, originated in studies of agricultural methods. It still enjoys a central role in agriculture, even more so since genomics has become so important to modern developments. After WWII, industry began to adopt DOE, especially in manufacturing. Unfortunately, the very success that DOE has enjoyed in that realm has also defined it in narrow terms. "Oh, yeah, I heard about DOE. They use it in manufacturing." That kind of awareness is too bad. Everyone should think about using DOE in any situation that demands a decision or an action. Why?Let's say, you need to fix something, to improve something, to innovate, or invent something entirely new. In order to accomplish this goal, you need to understand how something works: a process, a technology, whatever. To understand something, you must study or observe it. In other words, you must collect data for it. Then you need to distill the information out of this data, and separate the information from the noise that is caused by ever-present random variation. You want to collect the best data that will support the analysis that will provide the information you want.
This is true in any case, no matter which industry is involved.
STR: Can you give an example of how you have used JMP for DOE outside the classroom?
I used JMP before coming to SAS, while working in R&D at Abbott Laboratories. I worked in teams that developed medical diagnostics. These tests were actually automated chemical assays that combined chemistry, physics, optics and engineering. It was very difficult to study these systems using the simple one-factor-at-time approach that we were taught in school.I learned design of experiments (DOE) while previously working at Eastman Kodak, so I knew how powerful it was. We introduced design of experiments to Abbott Diagnostics in 1991, with help from nearby Motorola and what they later called Six Sigma. JMP quickly became our software of choice for DOE because it was the most interactive and powerful desktop statistics software for Macintosh and, later on, Windows.
STR: SAS Education currently offers two JMP courses on DOE -- Classic Design of Experiments and Custom Design of Experiments. What is the difference between these two courses?
First of all, there is nothing wrong with the classic design approach. In fact, custom design will make a classic design, if it is the optimal design for a given problem. The problem with a classic design is that it has severe limitations when it comes to many common situations that demand an experiment. The kinds of factors you may use, the factor levels that you can include, the model for your response, how runs might be grouped into blocks (if possible), and the number of runs is already fixed. We sometimes refer to these designs as 'tabled designs' for this reason. They're done.So the Classic Design of Experiments course teaches students about the unique aspects of each kind of design so that they can recognize which one to employ in a given problem and how to change their problem to match a given design. You are not really designing an experiment, so we do not talk about design principles but instead learn how to make decisions about a tabled design.
Custom design, however, starts with no previous design. So Custom Design of Experiments is about the underlying principles of design and how to exploit them using JMP. Custom design allows any kind of factor, any number of factor levels, any kind of blocking, any linear model, and any number of runs. Many times, one or more factors are hard to change between runs, so a completely randomized design is not practical or even possible. This situation used to demand a new kind of design, which required a statistician who was an expert in design. Now, however, it is just another part of custom design.
Finally, all of the examples in Classic Design of Experiments are provided as JMP data tables. Several of them come from the design literature. The data already exists and the answers are already known. In Custom Design of Experiments, there is no data. Instead, students design their experiment and then run it - on a simulator. In this way, each student experiences the whole process and collects their sample. We are able to talk about real-world random variation in experimentation. The students are lead through examples at first, but become more independent as the course progresses. We felt that it was important to not only learn about design, but also to develop enough confidence to be able to apply it to a new problem outside of the course.
STR: How does this course fit into a student's overall training?
Before attempting either of these courses, a student should be comfortable with JMP and the analytics involved, namely ANOVA and regression. So, they should attend JMP: Statistical Data Exploration and JMP Software: ANOVA and Regression, or have the equivalent knowledge and experience.As far as what other JMP courses students can go on to take, the options are wide and varied. In recent years we've added courses on scripting and modeling cycles in manufacturing processes, based on requests by customers. We are continually updating our courses so that they represent the best teaching methods and relevant examples, as well as introduce new JMP features that interest users.
View more information on JMP courses available through SAS Education.