Assignment 11, GEOS 585A, Spring 2009
1
ASSIGNMENT 11. MULTIPLE LINEAR REGRESSION
1. Run geosa11.m, selecting part I (assignment 11) from the first menu. This selection
specifies multiple linear regression without cross-validation.
Menus appear allowing you 1) select time series to be used as predictand and predictors ,
2) log-transform the predictand, 3) prewhiten the predictors, 4) include lagged predictors
(lags up to t-2 or t+2 relative to the year of the predictand, and 5) specify a calibration
period. Make choices that seem reasonable from your knowledge of the data and results
of previous analyses. For this exercise, make sure your pool of potential predictors
includes at least 6 variables. This can be achieved by various combinations of number of
predictor series and lags. For example, two predictor series and lags -1 through +1 gives
six potential predictors. Or six predictor time series and no lags also six potential
predictors.
You will then begin an interactive modeling process that consists of stepwise entry of
predictors one-by-one, with examination of model statistics and residuals analysis at each
step. Note that at each step you must press “being modeling or add another predictor” ,
followed by “review results.” For this exercise, run the model out to 4 steps. Then press
“Quit.” You should have 4 predictors are in your final model.
You will have four Figure windows. Answer the following questions and turn in your
answers with printouts of the figure windows.
2. (Caption to Fig 1) What is the explanatory power of the relationship? Refer to the R2,
overall F of the regression, the p-value for F in explaining your answer. Does the F-
value support a statistically significant relationship? Why is adjusted R2 smaller than R2?
3. (Caption to Fig. 2) For which predictors are the regression coefficients significantly
different than zero? What is multicolinearity? Name one undesirable effect of
multicolinearity. Is multicolinearity a problem with your set of potential predictors?
(Provide evidence in the form of results provided by running function vif.m on the matrix
of your predictors – this matrix is Xc in the workspace after running geosa11).
