Tuesday March 19th 2pm - 3:30pm Weniger 287
The final is closed book.
I will provide statistical tables if you need them (so you should know how to use them).
You should bring a calculator (although arithmetic errors are generally forgiven).
All material on the midterm study guide is also examinable.
You should be able to:
State the assumptions required for making inferences in regression.
Rank the assumptions in rough order of importance.
Describe the consequences of violating a particular assumption.
Describe/sketch residual plots that should be examined to diagnose problems with the regression assumptions.
Sketch a residual plot that illustrates a violation of a particular assumption (non-linearity, non-constant variance or non-normality).
Given a residual plot, describe evidence you see for violations of the regression assumptions.
Suggest a remedy for a particular violation.
Describe three ways a point may be considered “unusual”.
Name three case influence statistics and describe conceptually how they measure “unusualness”.
Identify from a scatterplot (in the simple linear regression case) if a point is likely to be high leverage, influential and/or an outlier.
Describe a limitation of case influence statistics.
Describe what is meant by multicollinearity.
Describe how multicollinearity might be detected.
Discuss the consequences of multicollinearity.
Describe the assumption that generalized least squares is designed to relax.
Derive the generalized least squares estimates (for known \(\Sigma\)).
Give an example of data where using weighted least squares is desirable.
Conduct a lack-of-fit test.
Interpret the result of a lack-of-fit test.
Describe the goal of robust regression techniques.
Describe why we might transform the response and/or the explanatory variables.
Choose a transform of the response based on a Box-Cox plot.
Interpret a parameter estimate based on a regression with a log transformed response.
State the additional assumption required to make inferences about medians in a regression using a log transformed response.
Give a reason why variable selection might be recommended.
Give a reason why variable selection might be avoided.
Describe the process of model selection by forward selection/backward elimination.
Give an advantage and a disadvantage of stepwise methods.
Perform one step of forward selection/backward elimination.
Name and describe four model selection criteria.
Discuss the similarities and differences between model selection criteria.
Discuss why it is dangerous to use the same data to fit a predictive model and to evaluate the model’s predictive ability.
Describe two regularized regression methods.
Describe why might we prefer biased estimates (or predictions).
Describe/sketch an example of a predictive model that would have low/high variance and low/high bias.
Discuss the difference in goals between explanation and prediction.
Describe the difference between linear and logistic regression models.
Describe the difference between linear and non-linear regression models.