ORIE 576 - SPRING 2004

TAKEHOME PRELIM EXAM - DUE April 21
-----------------------------------

1. Find a multivariate dataset suitable for a multiple regression
   analysis. The dataset must not be taken from the course textbook.
   However, it can be from any other statistics text, from a journal,
   or from data sources on the internet. The data set must include at
   least four independent variables (predictors) not including
   polynomial and interaction terms. When you find a suitable data set,
   send me an email with a brief description of it, and a reference
   for its source. If two students choose the same data, the first one
   to email me will have priority.

2. Conduct an all subsets regression analysis to find the best models
   based on the Mallow's Cp criterion. Compare the Cp results with
   those based on adjusted Rsquared. Also compare your all-subsets
   results with the model found using the forward stepwise regression
   procedure.

3. After selecting a model (or a small set of models), use the
   diagnostic tools discussed in class (such as partial regression,
   studentized residual, leverage and Cook's distance plots) to check
   the MLR model assumptions, and to identify potential problems
   (such as influential observations and outliers). In some cases
   the diagnostics may suggest modifications to your model(s).

4. Type a short report summarizing your findings. Your report should
   be self-contained, including a description of the data, the type of
   models considered, and your results. In particular, what does your
   final model say about the relationships between the response and
   predictors? Only include relevant statistical summaries and plots
   in your report. An outline might be as follows:

   Introduction: Description of the data and models considered.
   Model selection: Summary of results of model selection procedures.
   Diagnostic checks: Summary of model diagnostics.
   Conclusions: Have you found a good model for the data? What does 
    the model tell you about the relationships among the variables?

   The target readership for your report should be your classmates;
   i.e. someone who is not familiar with your data, but has a similar
   knowledge of statistics to yourself.