Stat 505 - Fall 2013 - Homework Guidelines and Suggestions

Statistics is best learnt by doing. The homework problems are intended to help you understand the methods we discuss in class, and apply them to solve problems. I also try to set homework problems that make interesting points.

There will be four homework sets, one for each of the major sections of the course. They will be distributed in the middle of the appropriate set of lectures and due approximately one week after lectures on that topic. Each homework set will consist of four problems. Most problems may have multiple parts. You are to turn in answers to three of the four problems. The first two problems are intended to be simpler than the last two, so I presume most of you will answer the first two problems and choose one of the second two. However, the choice is yours. In general, the four problems will be:

• An environmental question (or questions) and a set of data. You are to analyze the data set and answer the question(s).
• Some theoretical question(s) about the methods.
• A more complicated question and data set.
• More complicated theoretical question(s) about the methods.

You are encouraged to work together on the problems. I encourage groups to include both statisticians and non-statisticians. However, you must each write your own answers. Copying will not be tolerated.

Some of the problems during the semester will be labelled "Major data questions". Your answer to these problems should be in the following form:

• 'Executive summary' or abstract: short statement of question, approach, and answer. Should be no more than one paragraph. No formulae or technical language here. This should be written for the busy plant manager or state regulator.
• Approach: Describe your model or approach. Briefly explain why you chose it. Describe computational approach. If you considered multiple models, do not explain all of them in detail. Just say what you considered and the reasons for your final choice.
• Answer: answer the environmental question(s) using the 'final' model. Make sure you provide the context for the answer. For example, saying 'Estimated Phi equals 0.2349887' is not a sufficient answer because it does not provide context (what is Phi? How precise is the estimate? How big a value really matters?). You must interpret the answer. Estimates of precision (or confidence intervals) are (almost) always useful.
• Critique: use model diagnostics (if appropriate and available) or other information to critique your answer. What, if any, are the weaknesses or problematic assumptions? Another way to critique your answer is to discuss what are better models/methods to use if you had more time. You should discuss (briefly) why each model or method would be more appropriate. I do not want a laundry list of all possible analyses.

I expect total length to be no more than three pages for each data problem.

Answers to the "short" data questions and theoretical questions are more straightforward. Answer the question, showing your work.