Abstracts
- Keynote sessions
- Breakout Sessions
Thursday, September 17, 8:45 a.m.
The JMP® 9 Learnflow
John Sall
Co-founder and Executive Vice President, SAS
JMP 9 is still a year away, but some of the themes are emerging now. As we listen to practitioners and work with story cases, we see plenty of amelioration opportunities to make JMP 9 a more robust tool. Real data is often messy data in many different ways, with long names and levels, missing values, outliers, multimodal distributions, mixtures, and large models that are prone to overfitting. Large problems challenge us to improve performance and take advantage of multicore processing. Large tables also allow us the luxury of cross-validation subsets. Specialty problems challenge us to provide more robust scripting, better reference tools, and even tools for building scripts more automatically. We will be listening as you describe the complexities of your real worlds, so that we can build better analytic tools to work there more gracefully.
Thursday, September 17, 10:15 a.m.
Data Mining – Fool’s Gold or the Mother Lode?
Richard D. De Veaux, PhD
Professor of Mathematics and Statistics, Williams College
See how exploratory data analysis helps uncover patterns. Through case studies, renowned applied statistician Richard De Veaux explores this analysis technique for knowledge discovery.
Data mining is the exploration and analysis of large data sets, by automatic or semiautomatic means, for the purpose of discovering meaningful patterns. You can then use these patterns for decision making via a process known as knowledge discovery. Exploratory data analysis and inferential statistics also address these problems. What's different about data mining? How is data mining similar? This presentation will answer these questions by looking at a series of case studies using the tools of a typical data miner. We will demonstrate how to:
- Identify appropriate problems for data mining.
- Explore and prepare data for mining.
- Use a variety of techniques, including decision trees and neural networks, to build accurate predictive models.
- Evaluate the quality of data mining models.
Friday, September 18, 8:45 a.m.
Synthesizing Science and Innovation
Mike Cramer
Director of Operations Research for Worldwide Restaurant Innovation, McDonald ’s
Given today’s business challenges and an unforeseeable future, McDonald’s is relying more and more on innovative solutions to grow its market share. The opportunity costs and energy required to develop and validate these solutions are driving us to think differently in regard to our innovative processes. We are adopting a new term, “Innovience,” to describe the synthesis of innovation and science. The benefits of Innovience are accelerated failure at early stages of discovery/design and accelerated validation of solutions prior to launch. This discussion will cover the dimensions of Innovience and the critical need to develop the right infrastructure to execute.
Friday, September 18, 10:15 a.m.
Exploring Predictive Analytics and Data Visualization with JMP®
Stephen Few
Principal, Perceptual Edge
What-if scenarios that predict what might happen given different business conditions and decisions are most enlightening when we understand the relationships between the variables that influence potential results. What is defined as good visual analytics? How can predictive analytics be used to understand the past, monitor the present and predict future outcomes? We will seek to provide a better understanding of how the JMP Prediction Profiler is used to build predictive business models and interact with data and graphs to observe how changes in one variable influence changes in the others. Not a statistician? This presentation will help you learn how many analytical tasks can be handled through the use of relatively simple visualizations.
Yield, Statistics and JMP®
François Bergeret, PhD, General Manager
Ippon Innovation (France)
Alexandre Couvrat, Senior Engineer
SOITEC
In a large-volume semiconductor manufacturing plant, hundreds of process parameters have to be controlled with regards to product quality parameters, including yield. Advanced statistical methods are necessary in order to quickly discover any process issue and to find the root cause of complex yield issues related to some process tools or stages. From the user point of view, a statistical tool is needed with two characteristics: automated, to quickly highlight a very limited set of process parameters responsible for yield losses; and interactive enough to be able to tune the analysis using JMP discovery capabilities. From a statistical point of view, two innovative methods have been developed for this application and will be presented in this paper: 1) smoothing splines at each process stage, allowing ranking of all the stages with a statistical criterion. This new application is able to detect a faulty process stage with unique information – process dates at each process stage; and 2) likelihood ratio test in order to be able to deal with binary response variables, like the presence of a given pattern on the product. This application is able to analyze any yield issue when the only available response variable is binary. JMP software was selected for three reasons: 1) Its advanced statistical tools are adapted to semiconductor complexity; 2) It is very user friendly; and 3) It includes a powerful programming language, JSL, to develop automatic analysis for thousands of input and output parameters. A live demo of JMP with all these features will be presented.
Predicting Health Care Outcomes by Physical and Mental Health Population Data Modeling
Alberto M. Colombi, MD, MPH, Corporate Medical Director
PPG Industries Inc.
We probed the chronic disease continuum – from risk to outcomes – in 29 working populations. The interactions of obesity, depression, worker’s compensation, productivity loss and musculo-skeletal comorbidity were profiled in prediction models in order to support health investment decisions. Worksite-based measures of mental health (percent screened for depression and percent reporting a neutral stress and satisfaction offset score) interact significantly with worksite obesity prevalence (body mass index 30 and over) in predicting the average worksite medical plus wage disability compensation in case of injury. Additionally, annual depression episode rate and severity/risk-adjusted payment per episodes show significant differences in the presence of musculo-skeletal disorders comorbidity. Finally, certain mental health services utilization outcomes are significantly predicted by a host of factors, including productivity loss, showing that productivity can be enhanced by increased mental health services use, when needed. Screening and modeling techniques from JMP allowed using historical and population data to prioritize health investments.
The Basics of Experimental Design for Multivariate Analysis
Steve Figard, PhD, Senior Scientist
Abbott Laboratories
This paper is designed for beginner to intermediate practitioners of a form of analysis known as design of experiments (DOE). Specific objectives include: 1) defining some of the terminology; 2) introducing major thought processes, philosophy, strategies and rules of thumb; 3) keeping objectives in the context of JMP as an example of how this type of analysis is implemented in the software; 4) presenting a relevant example of the use of DOE in assay development; and 5) letting the software worry about “the how” of DOE, and instead focusing on “the when and why” of DOE.
Classification of Breast Cancer Cells Using JMP®
Marie Gaudard, PhD; Phil Ramsey, PhD
Partners
North Haven Group
Mia Stephens, MS, JMP Academic Ambassador;
Leo Wright, JMP Product Manager of Quality and Six Sigma Solutions;
Ian Cox, PhD, JMP Marketing Manager – Europe
SAS
This paper illustrates the use of JMP software’s powerful visualization and modeling capabilities in the context of a classification effort. We will utilize the Wisconsin Breast Cancer Diagnostic Data Set, a set of data used to classify breast lumps as malignant or benign based on the values of 30 potential predictors, obtained by measuring the nuclei of fluid removed using a fine needle aspirate. We begin by illustrating visualization techniques that help build an understanding of the data set. After partitioning our data into a training set, a validation set and a test set, we fit four models to the training data. These include a logistic model, a partition model and two neural net models. We then compare the performance of these four models on the validation data set to choose one. The test set is used to assess the performance of this final model.
Standard Least Squares Model for Identifying Problem Tools in a Manufacturing Environment
Ed Hutchins, Sustaining Engineering Manager
Cree
We describe an analytical method to identify problem tools in a manufacturing environment. Using JMP, we visualized a yield problem, collected a relevant data set and created a standard least squares model to identify the tools contributing most to the problem. Once identified, further analysis in JMP gave insight into the underlying root cause that allowed us to fix the problem and prevent future occurrences. On numerous occasions, we have found this to be an effective and efficient method to identify problem tools.
Designed Experiments that Changed the World
Bradley Jones, PhD, Director of JMP Research and Development
SAS
Folklore has it that great breakthroughs come as a result of a genius thinking out of the box and coming up with something creative – think Edison and Einstein. While this myth of genius makes a good story, it does not provide a reproducible way for organizations to make quantum jumps in innovation or productivity. This talk tells the story of several applications of designed experiments that “changed the world.” Its moral is that designed experiments provide a systematic approach that can pave the way from concept to customer.
Life Data Analysis with JMP® 8
Edward Kram, PE, President
Blue Arc Energy Solutions Inc.
Life data analysis is an important tool for reliability engineers. JMP 8 recently upgraded its lifetime data analysis capabilities to provide useful new features for reliability engineering applications. This paper discusses the use of JMP 8 with field data challenges, such as sparse failures, heavily censored data and failure data truncation. The use of JMP software’s other tools (e.g., Graph Builder, tabulations, etc.) are also discussed as useful visualization tools in reliability analysis.
Lure Your Black Belt Students into Learning Six Sigma Statistical Tools by Using Real-Life Examples
Don Lifke, MS, Process Engineer
Sandia National Labs
Teaching statistical tools to reluctant black belt students can be a challenge. One successful technique is to use examples that they can relate to, and to show that these tools can be applied to their personal lives as well. Two such examples will be presented. The first example will show how home prices can be predicted using the Fit Model feature in JMP. The example shows how a realtor could inadvertently list a home at much lower than its fair market value due to inaccurate evaluation by only using the typical $/square foot calculation. More factors need to be considered, such as age and lot size. The second example will demonstrate how design of experiments can be used to improve the golf game; specifically, the stance setup for hitting a drive optimized using the Custom Design feature in JMP (note: This was presented at the Better Golf Through Technology Conference at MIT in February 2008). Both examples have proven to be useful in capturing the students’ attention and motivating them to apply these tools to their work, and perhaps even their personal lives.
JMP® and SAS® Complement Each Other to Produce a Virtual Laboratory
Brian McFarlane, Senior Consultant; Wayne J. Levin, Founder
Predictum Inc.
This paper describes a system that combines the scripting and design-of-experiments functionality of JMP with the powerful modeling capabilities of SAS to create a virtual lab. This new approach replaces the previous practice of one-factor-at-a-time physical experiments with a rigorous system of custom designed experiments that estimate responses using models built on years of empirical observations. The complementary use of JMP and SAS produces a productive system that has the potential for significant results in reduced time to market for new products.
A Prospective Study of Cognitive Fluency and Originality in Children Exposed in Utero to Carbamazepine, Lamotrigine, Valproate Monotherapy
Kelly M. McVearry, PhD; John VanMeter, PhD; William D. Gaillard, MD; Kimford J. Meador, MD
Georgetown University Medical Center
Anti-epileptic drugs (AEDs) can produce behavioral teratogenesis in animals at dosages lower than required to produce anatomical teratogenesis, but cognitive effects of fetal AED exposure in humans are uncertain and predominantly limited to assays of psychometric intelligence. This paper examines the effects of fetal exposure to commonly used AEDs on psychometric creativity. It is a substudy of a prospective, multicenter, parallel-group study of neurodevelopment that enrolled pregnant women with epilepsy on AED monotherapy (carbamazepine, lamotrigine and valproate exposure) from 1999 to 2004. Children were tested with the Torrance Thinking Creatively in Action and Movement (TCAM), a standardized behavioral measure of cognitive fluency and originality. Assessment was blinded to drug exposure. Forty-two subjects met inclusion criteria (lamotrigine = 17; carbamazepine = 16; valproate = nine). Mean age was 4.2 years (SD = 0.5). Analyses included one-way ANOVAs, Student’s t-test and bivariate regression models. TCAM fluency was reduced in the valproate group (mean = 76.3; SD = 7.53) vs. lamotrigine (mean = 93.76; SD = 13.5; p < 0.0015) and carbamazepine (mean = 95.5; SD = 18.1; p < 0.003). TCAM originality was reduced in the valproate group (mean = 84.2; SD = 3.23) vs. lamotrigine (mean = 103.1; SD = 14.8; p < 0.002) and carbamazepine (mean = 99.4; SD = 17.1; p < 0.01). These effects were not explained by factors other than in utero AED exposure. Children exposed to valproate in utero had impaired cognitive fluency and originality compared to two other AEDs.
Analysis of Military Occupational Skills Requirements for the Army National Guard
Christopher K. Mitchell, MS, Research Analyst
Army National Guard Bureau - Training Division
Each fiscal year, the Distributed Learning branch performs a Military Occupational Skills (MOS) analysis to identify those most needed for transition (reclassification) to meet Army National Guard (ARNG) readiness and deployment needs. Before 2004, the methodology for selection had been to request input from the field, compare the Army Training Requirements and Resources System (ATRRS) database with the Standard Installation/Division Personnel System (SIDPERS) database and survey the ARNG Chief of Training for future needs. The selection criteria and the analytical methods were weak and lacking probative value because they primarily relied on two sources of information that rarely agree – the ATRRS and SIDPERS databases. Using JMP we visually demonstrate the “how and why”value of JMP software’s principal component analysis (PCA) by: 1) effectively and efficiently determining which courses warrant recommendation; 2) determining courses that are underutilized, under- or over-funded, understaffed or in too few locations; 3) identifying skills that, due to density or length, should be considered for inclusion or exclusion from funding or staffing; and 4) providing more finite strength management input and training requirement expectations.
Conjoint Marketing Experiment Design Considerations
Chris Nachtsheim, PhD, Professor and Chair of Operations and Management Sciences
University of Minnesota
Rob Reul, MS, Founder and Managing Director
Isometric Solutions LLC
Unlike operational design-of-experiments implementations where a very select set of trials can be easily completed, marketing choice experiments are easy to "trial," but often difficult to complete. The conjoint challenge is to maximize the amount of quality information any given respondents provide before either their responses become indistinguishable or they quit the survey. Conjoint designers have many variables with which to tailor a marketing choice experiment, including the number of levels within each attribute, the number of attributes in the study, the number of profiles a respondent chooses from and the number of choice tasks given to each respondent. The mathematical basis directing various design considerations will be explored along with accompanying client case studies.
The Turning of JMP® into a Semiconductor Analysis Software Product: The Implementation and Rollout of JMP® within Freescale Semiconductor Inc.
Jim Nelson, Manager IT, Yield Management Systems
Freescale Semiconductor Inc.
The semiconductor manufacturing environment is a very specialized area of manufacturing. Fabrication of a computer-integrated circuit is a very complex process. Chemical, photographic, mechanical, electrical and spatial factors all have to intersect in a highly choreographed process to produce a working computer chip. The process of optimizing the yields or investigation of unexpected yield losses require both specialized data visualization tools and sophisticated statistical analysis to quickly find the root cause of the low yield. Out-of-the-box JMP is not a semiconductor analysis software product. While a skilled engineer can manipulate data and various JMP platforms to approximate the required visualizations and analyses, today’s marketplace does not permit such a time-consuming process. Engineers need analysis tools that quickly provide the precise visualizations and analyses they need with the fewest number of mouse clicks. This paper details the processes, additions and modifications made to JMP to turn it into a semiconductor software analysis product.
The Best of Both Worlds: Designing Experiments in the JMP® and SAS® Environment
José G. Ramírez, PhD, Industrial Statistician
W.L. Gore and Associates Inc.
Complexity and supply chains that may expand across the globe require us to design experiments that must take into account: 1) the multistep nature of manufacturing processes; 2) the large number of factors involved; 3) the different sizes of experimental units; 4) the restrictions in randomization that occur; and 5) the constraints on the number of experiments that can be run. Designing experiments for these situations can challenge many design-of-experiments software packages. However, the integration of JMP and SAS now allows us to take advantage not only of the flexibility and power of the Custom Designer, but also of new capabilities in SAS 9.2 PROC FACTEX to design experiments for multistep processes. The gamut of manufacturing situations that can now be handled with JMP and SAS will be illustrated using the Custom Designer to design an experiment for a complex manufacturing situation involving a hard-to-vary factor, nine process factors and eight mixture factors, demonstrating how a JMP and SAS application can be used to design experiments for three-, four- and five-step processes.
Resected Pancreatic Adenosquamous Carcinoma: Clinicopathologic Review and Evaluation of Adjuvant Chemotherapy and Radiation in 38 Patients with the Use of the JMP® 8 Reliability/Survival Platform
K. Ranh Voong*, Jon Davison, Timothy M. Pawlik, Manny Uy*, Charles C. Hsu, Jordan Winter, Ralph H. Hruban, Dan Laheru, Sonali Rudra, Michael J. Swartz, Hari Nathan, Barish Edil, Richard Schulick, John L. Cameron, Christopher Wolfgang, Joseph M. Herman*
Johns Hopkins University
* Co-presenters
The purpose of this study was to examine the impact of adjuvant chemoradiation therapy (CRT) on overall survival, identify clinicopathologic features associated with prognosis and assess whether the percentage of squamous differentiation is associated with an inferior prognosis in pancreatic adenosquamous carcinoma (PASC). PASC is a rare morphologic variant of pancreatic adenocarcinoma with an especially poor prognosis. Forty-five of 3,651 patients who underwent pancreatic resection at Johns Hopkins Hospital between 1987 and 2007 were identified as having PASC with any squamous differentiation. Statistical analyses were performed using JMP 8 statistical software on the remaining 38 patients amenable to adjuvant chemo-radiation therapy (CRT) with clinical outcome data. Survival curves were estimated using Kaplan-Meier techniques based on JMP software’s survival platform.
Our series supports that survival following pancreatic resection of PASC is poor (median overall survival: 10.9 months [95 percent CI: 8.2-12.8]). However, treatment with adjuvant CRT is associated with improved survival (p = 0.0048), suggesting that patients with this rare form of pancreatic adenocarcinoma could benefit clinically from the incorporation of adjuvant CRT into their treatment regimen.
