Request Expert

Expert (rhymes with "hats off") has designed state of the art reliability and accelerated life testing software that is used throughout IBM. This software is incorporated within IBM's AGSS system, an interactive statistical graphics system that has been marketed by IBM since 1992. He has developed and implemented unique capabilities for the design of locally optimal accelerated life tests, and has presented and published papers on these topics, based on his research. He received an IBM Outstanding Technical Innovation award for his work in this area. He has provided extensive education and training in these areas, including the training of reliability engineers throughout IBM worldwide. Additionally, Expert designed and taught a course on Reliability and Quality Control at MIT, and he has conducted numerous public 5-day courses on Reliability and Accelerated Life Testing. He has also provided on-site training and consultation on Reliability and Accelerated Life Testing at major corporations. He has won Underwriters Laboratory approval for a client's product by designing an accelerated life test for the client and presenting the results to Underwriters Laboratory. This averted a costly 12-month delay in product approval.


Expert has designed and developed most of the statistical content of a major statistical graphics system (IBM's ASGG System). After retiring from IBM he developed unique software for statistical control of continuous manufacturing processes (where successive data points tend to be highly autocorrelated) as well as software for statistical design and control of discrete manufacturing processes. Methodology includes the design and construction of both Shewhart and Cusum control charts to achieve specified levels of alpha and beta risks, correction of control chart limits to account for autocorrelation, simultaneous plotting and analysis of multiple positional data, Capability Analysis, Pareto Analysis and a multiple factor Analysis of Means (ANOM). The continuous SPC software system (RollSPC) is applicable to products manufactured in a continuous roll, such as paper, plastics, metals and coatings, as well as to chemical and other batch processes. Expert produces and markets the RollSPC System, and has employed it in his seminars and in consultation with clients on these subjects.

Expert has expertise and experience in the design and analysis of full factorial, mixed factorial, and fractional factorial experiments; nested designs; central composite designs; and response surface analysis. He has designed software for these capabilities as part of IBMs "A Graphical Statistical System" (AGSS), and has provided client training and consultation in these areas. He has also published papers on the application of designed experiments to problems in computer science.


Having written his PhD dissertation on "Exact Distributions of Wilk's likelihood Ratio Statistic," Expert has considerable expertise in multivariate statistical analysis. He has published original tables of the distribution of Wilks' Likelihood Ratio Test in Biometrika Tables for Statisticians, Volume 2 and in Biometrika. He also published papers in the Journal of the American Statistical Association on the use of Expected Significance Levels (a concept that he originated) to compare the sensitivities of several competing multivariate tests. He taught a graduate level course in Multivariate Analysis in the Statistics Department at Yale University. He is currently working on applications of multivariate analysis to multivariate control charts with autocorrelated observations. He is very knowledgeable on topics such as multivariate likelihood ratio tests, T-Square and Chi-Square tests, principal component analysis, multiple discriminant analysis and canonical correlation analysis.

Expert is well-versed in probability theory as well as probability distributions commonly used in statistical applications, such as the Normal, Lognormal, Weibull, Extreme Value, Beta, Gamma, Poisson, Binomial, and Hypergeometric distributions. He has designed and implemented density, CDF, inverse CDF, and hazard functions for a large number of probability distributions, as well as functions for probability and hazard plot scaling, confidence intervals and tests of hypotheses. These are all included as past of IBM's AGSS System. He has consulted with clients on many industrial problems involving probabilistic formulation and solution of problems requiring probabilistic reasoning.


Expert has designed extensive least squares subsystems for IBM's AGSS system including: constrained non-linear regression and curve fitting, robust regression methods, regression diagnostics, residual analyses, four different types of regression confidence and prediction intervals, stepwise subset regression, and all possible regressions. This system has been widely used throughout IBM, and for graduate regression courses at MIT. He has co-authored a paper on "All Possible Regressions," dealing with computationally efficient methods for computing all possible subset regressions for data sets containing an arbitrary number of independent variables. Expert has extensive experience in fitting and testing both linear and nonlinear regression models.


In addition to his extensive education in statistics, Expert has taught courses in statistics at several universities and has also taught courses in applied statistics, reliability, statistical process control and design of experiments to engineers throughout IBM. He designed most of the statistical modules in a major statistical graphics system (IBM's AGSS System), and has consulted with clients in a number of areas of applied statistics. A great deal of his consulting work has involved statistical data analysis for a wide variety of applications. He has carried out large scale simulation studies in areas of character recogniton, computer disk accessing methods, operating system scheduling algorithms and algorithms for efficient management of free storage in computer operating systems. He has also employed Monte Carlo methods to compare the sensitivities of competing multivariate test statistics.

Expert began his studies of statistical quality control under Dr. W. Edwards Deming at New Yotrk University. After joining IBM, he hired Dr. Deming as a consultant and collaborated with him in assessing IBM's quality and reliability problems in its installed customer base. Expert has designed and developed all of the Quality Control modules contained in IBM's AGSS System, a major statistical graphics system used throughout IBM, and marketed by IBM to customers worldwide. Since retiring from IBM Research, he has designed, developed and marketed a unique statistical process control system using the AGSS System as a programming platform. A description of the system's capabilities appears above under the heading "Statistical Computer Software". Unique capabilities include facilities for: designing Shewhart and Cusum charts to satisfy user specified alpha and beta risks, and/or Average Run Lengths (ARL's); plotting of Shewhart and Cusum control charts and calculation of observed significance levels for several different runs tests; multi factor Analysis of Means (ANOM), an extremely useful analysis tool for uncovering sources of assignable causes; and automatic correction of control chart limits to account for variability due to autocorrelation. Expert has taught a course in Reliability and Quality Control at M.I.T., has trained IBM engineers in Statistical Quality Control, and has also trained and consulted with clients on quality control problems.

Expert studied the design and analysis of statistical surveys under Dr. W. Edwards Deming, author of two major books on the subject, at New York University. He hired Dr. Deming as a consultant to IBM and worked with him on the design of a sampling plan to monitor the performance of IBM equipment at customer sites. Expert taught a course on sampling at the Northeastern University Graduate School of Business. At IBM he designed and analyzed large scale market research and employee opinion surveys.

Expert has studied Time Series Analysis under Dr. T. W. Anderson, author of a major book on the subject, at Columbia University. He is knowledgable in the Box-Jenkins Time series methodology, and has applied time series methodology to the design of algorithms for handling autocorrelated data in construction of statistical quality control charts. He has consulted with clients on problems involving time series analysis.

Designed and analyzed accelerated life tests of carbon monoxide sensors for a manufacturer of such sensors, and used the analyses to quide the client through approval by Underwriter's Testing Lab. This saved the client 9 months of production that would have been suspended awaiting the standard 12 month testing certification. Carried out a series of Step Stress Accelerated Life Test analyses for a major gas and electric power supplier to assist them in policy decisions for replacement of underground electrical power cables. The software used to carry out the analyses was developed by Expert, and is the only known software for fitting and evaluating accelerated step stress models. Applied Fraction Defective Failure Models and probability theory to enable a major defense electronics company to identify the source of component failure within a complex guided missile assembly. The software, developed by Expert, is the only known software for fitting and evaluating fraction defective models.Assisted a semi-conductor test equipment manufacturer in determination of Gage Repeatabilty and Reproducibility for their test equipment, as applied to their customer's wafer fabrication. Worked with a major manufacturer of catalytic converters to evaluate test results for comparing two different catalytic converter designs and for determining sample size requirements for future product evaluations. These tests are very expensive to carry out, so that minimization of sampling requirements is crucial.