I started out being interested in operations research. My senior thesis at Davidson College, which was optional and earned me honors upon graduation, was in the field of queueing theory. I studied independently about M/M/1 and M/M/∞ queueing systems, and I conducted two field studies in the Davidson community to try to define a new queueing model. I saw limitations in the models I studied, but at the same time, as I dug in my heels on the mathematics, I realized there was a reason that model simplifications had to be made.
I went straight into the statistics Ph.D. program when I enrolled at Indiana University. Along the way, I realized that my studies were taking more down the path of fundamental and theoretical statistics, whereas I had always wanted the hands-on, real-world application experience. After completing my M.A. degree, I moved on to teaching, and then eventually put my statistical training to use in job settings.
The most poignant application was developing the Sports Potential model. This was a scoring system which, based on a person's body measurements and physical test results, provided valuable feedback about the person's strengths and weaknesses relative to professional athletes across 50+ sports, and also recommended which of those sports were good matches for the individual. I worked with two of the top bio-statisticians at Stanford University, using R and S to fine-tune the data model. We called it sports vectors, which was a take-off of a vector analysis methodology currently in use by those professors. In a sense, the mathematics of it could be thought of as a least squared means - for the 88 data points which were a person's test results, we isolated 30 of them that produced the most meaningful predictive results, and then treated it as a 30-dimensional vector space and utilized appropriate mathematics to minimize the cross products. There was more to it, though, since I had to weight various aspects of the model so as to reflect the reality of the human body, to take into account the statistical range of body measurements and abilities as dictated by our test data. The end result was a fantastic model that tested 93%+ accurate for both men and women, and the sport ranking results presented to our clients were always met with approval.
I currently use some statistical analysis when processing crawled and eBay API data on QBike.com. I tie in raw data with sales data I capture from Google Merchant and Google Adwords, and I attempt to maximize my campaigns based on the anslysis. I also watch for traffic patterns on QBike using statistical filters in order to alert me to any Google search algorithm changes.