Advanced Statistical Methodologies

Overview

The combination of Lean and Six Sigma has matured over the years. They have always provided empirical methodologies for:

• Problem solving: defining problems
• Structuring problems
• Assessing their importance
• Developing teamwork and abilities
• Addressing uncertainty
• Experimentally solving problems and ultimately controlling solutions

As Lean and Six Sigma have matured so too has the level and complexity of the problems they are called upon to address. Low hanging fruit has been picked. Tough technical problems and Big Data problems are starting to dominate the more mature implementations of Lean Six Sigma. Predictive modeling, Big-Data data mining (thousands of variables!), high throughput experimental design, experimental methods in novel arenas such as website optimization, differential advertising purchasing, infrastructure flow and tuning, simulation modeling and acceptance sampling programs all demand new skills beyond the original technical skills of Six Sigma.

We have always treated the IMPROVE phase of Lean Six Sigma as including not just Experimental Design but also the tools of data mining and simulation studies. Historically we taught such tools as Crystal Ball, CART (classification and regression trees), MARS (multivariate adaptive regression splines) and their more recent extensions using modern ensemble methods (random forests, etc.).

It is not the case that we always finds problems that need to be addressed with these high-level technologically advanced algorithms but rather that Circle 6 has the capability to assess the problems needs and to respond appropriately: sometimes all the client needs to do is fix the observed problem, sometimes they need to design and implement an experiment to find out what to do, sometimes it’s more complex than that and they need to data mine large databases or they need to intelligently simulate very complex processes to try to get a handle on uncertainties. We are prepared to deal with each of these contingencies.

Regression Trees

Regression and classification trees are data mining’s most popular modern modeling techniques that build easy to understand/implement predictive models of high dimensional processes. They are used extensively in marketing, in resource utilization modeling and process performance characterization.

Advanced Simulation Studies

Modern simulation and resampling techniques: We typically use the Crystal Ball environment to show how to effectively use your own data to answer process performance questions by empirically exploring thought experiments (what-ifs). Applications have ranged from bid evaluations in Super Fund clean up sites to modeling various evolutionary paths of island colonization.

Acceptance Sampling Program Design

We teach you how to understand, model and control the risks inherent in complex supply chains. There are well developed but modestly complex dynamic programs to proactively protect the organization from accepting or releasing “unacceptable” product.

Multidimensional Discrete Event Modeling

Multidimensional discrete modeling is a general class of computer modeling activities that subsumes trees and classic general linear models. They are focused on processes that have multiple potential causes and some discrete set of outcomes of interest. Some examples we work on are: purchasing product as function of how approached and who you are; revisiting hospital in some short interval having just left hospital; being injured during training activities in sports or service training; successful breeding as function of setting, sire and dam characteristics.