Past Work


The list below gives examples of work that I either led or supported.  It is not exhaustive; it lists mostly those studies that went beyond standard applications.  I do not pretend that every outcome was perfect … but nearly all were effective, providing a useful business or engineering result.  Several examples involve adapting or combining methods.  The success of each effort should be attributed to the team with which I worked for any particular study.  And – yes – there were one or two studies that didn’t pay off as anticipated, but these were very few in number (I can remember only two); these involved associated challenges related to trying to coax stable behavior from a manufacturing process with ill-suited equipment, and unauthorized changes that were applied to a process (equipment, settings, environment) without permission in the middle of a study.

Power Analysis and Sample Size Determination
  • Several examples of statistical power analysis and calculating appropriate sample sizes.
  • Includes methods based on theory and the underlying statistical method and distribution.
  • Includes methods based on resampling and simulation.  (See simulation section below for examples.)


  • Worked with data team of a billion-dollar multinational corporation tasked with developing a model for projecting transportation costs under certain conditions.  Used regression model diagnostics to improve the initial model, noticeably reducing model bias and error.  Resulting model incorporated polynomial terms and a variance stabilizing transform.  Suggested methods for monitoring future data for shifts in the model.

Latin Square ANOVA combined with Principal Components Analysis (PCA)

  • Made minor adjustments to site-specific OFAT-based studies across three sites that allowed a reduction in overall testing, while incorporating effects associated with three manufacturing sites, and three suppliers each for two raw material components.
  • Found that the most significant effect of inputs on outcomes from a mixing process came from variation attributed to three facilities in the U.S. and EU

Combined DOE with Reliability (Survival) Analysis

  • Resulted in impressive improvement in urinary incontinence bag seal reliability. (Chinese supplier)

Combined DOE with Reliability Analysis

  • Optimized and validated manufacturing process for sealing ostomy pouches, resulting in reduction in process time by 50%

Combined DOE with Logistic Regression

  • Resulted in impressive improvement and stability in manufacturing process outcome for sterile seals. (India factory)

Survival Analysis for qualification of ALT equipment (ALT test ovens)

  • Found manufacturing discrepancy in test equipment that was overlooked by the engineers; fixed the issue and demonstrated equivalence in outcomes after improvement. Result was consistent test results across four facilities globally.

Statistical Process Control

  • Introduced SPC on regression residuals from manufacturing process for urinary catheter tipping.  Result – showed engineers that a process problem came from someplace other than the one they thought.
  • Introduced advanced Quality Engineering and Six Sigma training in the mathematical statistics behind SPC.   Implementation included calculating ARL and statistical power, with practical application during the start-up of a new manufacturing facility.
  • Introduced several methods of Multivariate SPC for more appropriate analysis and monitoring of multi-response processes.

Combined Logistic Regression with Plackett Burman DOE

  • Multistage process with over 100 process variables.
    • Quickly reduced to less than two dozen and screened the remaining using PB.
    • Along with Logistic Regression, introduce the use of Odds Ratios to quantify the improvement, and the use of ROC curves to optimize the process.
    • Results included several rounds of improvement in cycle time (reduced by 1/4 to 1/3 each time) and process consistency/stability for catheter coating process.

Measurement System Analysis

  • Used fundamental statistical principles to extend the standard GR&R model to new situations.
  • Introduced to the company the mathematical statistics behind the GR&R model, and how they are used.
  • In conjunction with DePaul stats professor, developed a multivariate approach to GR&R for colorimetry measurement systems (based on CIE-Lab color scales.  Employed colorimetry multivariate GR&R on several occasions.
  • Introduced methods and procedures for Measurement Uncertainty (Measurement Uncertainty Budgets using the PUMA method) to evaluate the appropriateness of test systems and equipment.

Used Fleiss’ Kappa method of evaluating subjective rating methods

  • Used to assess the effectiveness of subjective rating systems. Where a system was shown as effective, was used as evidence for regulatory review, and also as a method to qualify new inspectors/evaluators.
  • Demonstrated the pitfalls of using ‘expert’ assessors to subjectively evaluate film sound/noise (found very poor intra- and inter-rater agreement). Combined with the frequency-domain assessment of sound.
  • Introduced the use of Fleiss’ Kappa to assess the effectiveness of QA inspection of product; extended the method to find the maximum line speed that still maintained acceptably-low risk of non-detection of nonconforming product.

Simulation – various examples

  • Used simulation in R to demonstrate the reasons for preferring R or S charts.
  • Used simulation combining time series (ARIMA) and SPC (Cusm chart) to calculate the minimum shift in sales that would be needed in order to observe a change over a one year period.
  • Used simulation to enhance Logistic Regression in the presence of quasi-complete separation – eventually replaced by using Firth’s penalized likelihood.
  • Used simulation to model statistical power of a study based on the Cumulative (Ordered) Logit model.

Spectral Analysis

  • Used frequency-domain assessment of sound to build a spectral ‘envelope’ that could be used to evaluate or screen incoming or proposed new materials for suitable noise characteristics. (Similar to using IR or GC scans to verify incoming materials.)
  • Guided design teams toward specialized lab ( in order to tailor the operating sound of a commercial product.