Reviewing “Event History and Survival Analysis” in Absentia

I'm pretty bummed today. With all that's happening in both my personal and work lives, I finally had to acknowledge I'll be unable to accept friend and colleague Paul Allison's invitation to participate in his upcoming statistical seminar, “Event History and Survival Analysis”, July 14-18, in Philadelphia.

I've known Paul since graduate school many years ago. He was a star then who's gone on to a distinguished teaching and research career, currently as professor at the University of Pennsylvania. Me, on the other hand ….?

Besides his academic duties at Penn, Paul owns a company, Statistical Horizons, that offers training in “the latest statistical methods in an efficient, affordable and enjoyable manner.” He teaches many of the 2-5 day courses himself, outsourcing the rest to statistical peers.

There are a number of Statistical Horizons curricula pertinent for business analytics/data science. Linear Regression and Logistic Regression would be considered foundational, while Survival Analysis is now central to analytics, business examples including time to failure, churn and customer lifetime value. Longitudinal Data Analysis, “consisting of measurements of predictor and response variables at two or more points in time for many individuals”, is also at the core of BA. I'd love to sit in on Paul's Missing Data course that details “new methods for handling missing data ...(that) have only become practical in the last few years with the introduction of widely available and user friendly software.”

Relevant curricula for data scientists taught by other instructors include Applied Bayesian Data Analysis, Causal Inference with Directed Graphs, Data Mining, Machine Learning: A Hands On Introduction, Multilevel Modeling, Propensity Score Analysis, Social Network Analysis, Statistics with R. I'm sure many of these are excellent, though I have no personal experience with the instructors.

Paul gets consistently high marks from his students – both Penn and seminar. The one day I sat in on a Chicago version of his LDA course two years back, I was quite impressed – but not surprised – at the quality of instruction. For Paul, it was simply a continuation of teaching excellence honed as a TA in grad school and conducting demanding “brown bag” seminars with the department. Some stats instructors are excellent presenters in the math of statistical analysis; others are outstanding computationalists; fewer still can present the nuts and bolts of difficult topics in a manner comprehensible to non-experts. Paul's accomplished at all three.

Besides the fact Paul's a friend and an outstanding academic, I've had keen “student” interest in his comprehensive survival analysis curricula since I first read the promotional material several years ago. And though I've implemented “survivals” many times in R, I'm self-taught, often wishing I had the opportunity to take instruction from an expert. Plus, I've not worked with the survival procs in either SAS or Stata.

To ease my disappointment of not being able to make the trip to Philadelphia, Paul sent me the notes for both the SAS and R dialects of EHSA. And, of course, they're top-notch – all 184 pages for SAS! He's not kidding with claims the seminar covers a semester of material. I was, however, able to learn a great deal by simply spending a few hours reviewing.

What struck me as I began my read was the clarity of conceptualization – the definition/dimensions of events and censoring, in addition to explanations of how event analysis differs from conventional prediction techniques. The many real-world illustrations solidified the discussions.

The math of parametric hazard and non-parametric survival functions that follows is digestible, providing the basis for methods such as the Life Table and Kaplan-Meier. There's no shortage of quality data sets for analysis, and it seems every possible permutation of SAS procs LIFETEST, LIFEREG, PHREG, LOGISTIC, with code, is examined in detail.

I must admit to becoming fatigued at the halfway point of the notes, but nonetheless welcomed the exhaustive set of illustrations. In the end, I feel the clear exposition of material will allow me to readily find the illustration that best matches my specific point-in-time needs. Ultimately, that's the barometer of a quality stats course for me.

If you're looking for an excellent survival analysis seminar in Philadelphia the middle of July, this is it. As of now, there's at least one available spot.