Summary of experience

The experience, though worrying at times, has been overwhelmingly positive. Through their course evaluations and conversations afterwards, the students tell me that they appreciated working on a complex interdependent project. The team and active learning approach was universally praised. They were delighted to have an instructor cover topics as they chose. Each learned much from their independent studies. The principal negative comment was that they had not started the project soon enough.

Most interesting was the common remark that they came to better understand statistical topics that they covered in other courses. This latter is due, I believe, to the fact that they were responsible for developing a statistical strategy. It forced them to review and synthesize previous material and to explore the boundaries of the strategy. This naturally led to wide ranging statistical discussion. Moreover, there was a dawning recognition that although details differ from area to area that perhaps the following structure was common to many areas of statistical analysis:

• Analysis session
• Graphical and numerical summaries
• Model specification and editing
• Model Search
• Diagnostics
  • Influence
  • Identifiability
  • Model checking
• Inference

Covering two or more areas could lead to interesting discussion as to how one might capitalize on this (or some other) common structure.

The group project was essential in having the students experience computational thinking. They designed, developed, and debugged software structures that dealt with routine statistical calculations (e.g. diagnostics, hypothesis tests), relatively complex statistical algorithms (e.g. leaps and bounds model search, lowess smoothing), new interactive statistical graphics (e.g. regression plots, plots of h_ii/(1-h_ii) versus t_i), interface design (e.g. every node in the strategy), object-oriented design of the classes and functions associated with the various analysis hubs, and the overall strategy of carrying out a linear regression analysis. While each student focussed on an area in the strategy, the breadth of the area and the detail required in implementation ensured that each student worked on statistical and computational issues at a variety of abstraction levels. The creativity of the students is easily seen in the results.