|
Abstract:
|
Many of the design or analysis problems facing a collaborative statistician can be handled time-tested techniques. There are plenty of good reasons for this practice: these methods tend to be applicable in most situations, their operating characteristics are well-studied and peer reviewers are likely to be familiar with them. Nevertheless, there are some occasions where using a new method will be necessary or beneficial. I will draw on my own experience how to identify these occasions, pinpoint, understand and apply an appropriate method, and making it stick. Some of the examples I will use to drive home essential points are recursive partitioning, competing risks and multiple imputations. I will discuss the advantages and disadvantages of using short courses, peer discussions and review articles as resources in the context of these examples as well as the increasingly crucial role of software. Finally I will try to convince the audience that learning new skills should be a life-long enterprise.
|
