Hypothesis Testing for Observational Data

Learning to test for difference in observational data

This one and a half-day Hypothesis Testing for Observational Data workshop will enable students to answer multiple key questions. Has there been a change in process or product? Is this process or product different from this one? Is there a difference in customer satisfaction from before to after the project completion? Are there differences in process cycle times or yields among the operations around the world? Ultimately, hypothesis testing are statistical methods to answer the fundamental question, “Does the data support a hypothesis that says there is a difference?”

Hypothesis testing applies to Design of Experiments and “observational data.”

The Design of Experiments refers to a plan designed to systematically manipulate factors (Xs) to understand if they have an effect on some dependent variables (Ys) and model that effect. Observational Data refers to information derived from observations of processes or products where there is no systematic manipulation of causal factors (Xs).

Medical research is an excellent example of this latter kind of observational data since it may be unethical to withhold treatment that could improve one’s medical condition or to apply treatment where there is a risk that it could be harmful.

Course Topics

  • Statistical significance
  • Reference distributions including the Normal, Chi-square, Student t, and F-distributions, how they work and how to apply them
  • How to formulate and test the null and alternative hypotheses
  • Know which tests to use for which comparison, and the nature of the comparison and the type of data available
  • Confirm statistically significant differences in processes or products for continuous variables like time, temperature, pressure and discrete variables like defect rates or percent compliance
  • Test the probability that claims of process performance are valid
  • Determine the minimum sample size you will need to confirm statistically significant differences
  • How to assess the probability that you will conclude correctly about differences in products or processes given that real change does or doesn’t exist — “statistical power” as depicted on a “power curve”

Day 1 — Introduction to Hypothesis Testing to Determine Difference

  • Review of basic statistical concepts – distributions, parameters and statistics, central limit theorem
  • An introduction to the key reference distributions – Z, t, F, Chi-square
  • Review of the central limit theorem and the concept of the standard error of the mean
  • Workshop to generate reference distributions and determine change
  • The null and alternative hypotheses and alpha levels
  • Testing for differences in means for continuous variable data – t-tests, paired t-tests, ANOVA

Day 2 — Hypothesis Testing for Discrete Data – Proportions & Rates - 4 Hours

  • Review of discrete data vs. continuous data
  • Hypothesis testing and interpretation of analysis of proportions
  • Hypothesis testing and interpretation of analysis of rates
  • How to determine the minimum sample size and the power curve
  • Guidelines and summary of when to use a particular method

Course Objectives

Upon completion of this one and a half-day course, participants will have learned the methodologies used in testing for differences in observational data.

What Career Paths Benefit the Most From Six Sigma Hypothesis Testing for Observational Data Workshop?

  • People working with manufacturing or non-manufacturing processes
  • Manufacturing, Process and Quality Engineers
  • R&D scientists and engineers
  • Product and process development and design engineers
  • Marketing and business analysts
  • Candidates for Six Sigma Black Belt certification

Additional Course Notes

Customized versions of this course are available for organizations on- or off-site for organizations desiring to target specific groups or objectives, e.g., overview for management including how to lead successful implementation of advanced analytical methods, quality and process engineers with emphasis on the analysis and interpretation, sampling method design and follow-up strategies and how to communicate the story that data tells us; marketing personnel and how to design studies to answer questions key to market interpretation.

Prerequisites: “Intro to Statistical Methods to Manage & Improve Processes” or equivalent knowledge about basic statistical methods is required. Equivalent knowledge includes awareness and understanding of basic concepts such as distributions, sample statistics, and the central limit theorem. The courses “Introduction to Statistically Designed Experiments” and “DOE-Intermediate Design of Experiments” provide excellent backgrounds for this course but aren’t required.

Credit & Follow-Up

Participants who complete this course will earn credit toward certification toward Six Sigma Black Belt certification.

Credits: 1.15 CEUs