Company: Applied Research Associates, S. E. Div.
Company background: Specialization: Statistical Computing; Multivariate Regression Analysis; Software Development; Statistical Graphics; Non-Linear Regression; Design of Experiments (DOX); Geostatistics; Categorical Response Modeling.
Contact: Marshall B. Hardy
Title: Staff Scientist and Statistician
Country: USA
Award Category: Innovative Problem Solver
Tell us about the business problem you were trying to solve?
I had a long career in industry and developed many formulas. One for reducing alpha in multiple comparison and stepwise procedures is an approximate False Discovery Rate, AFDR = min(mean(alpha)) for m dependent tests =alpha*(m+1)/(2m*[ln(m)+.57721]+1), AFDR in wikipedia.org/wiki/False_discovery_rate
How did you use SAS to solve that business problem? What products did you use and how did you use them?
I've used this for decades in SAS multiple comparison and stepwise procedures requiring the user to input an alpha level. AFDR = min(mean(alpha)) for m dependent tests = alpha*(m+1)/(2m*[ln(m)+.57721]+1). https://en.wikipedia.org/wiki/False_discovery_rate#Benjamini%E2%80%93Yekutieli_procedure
What were the results or outcomes?
Better, more reliable models from SAS multiple comparison and stepwise procedures that take into account many dependent tests used to produce the models, by reducing alpha.
Why is this approach innovative?
SAS multiple comparison and stepwise procedures requiring the user to input an alpha level don't adjust the alpha level for many dependent tests used and should.
