Regression Analysis - Time Effect - Solving The Problem

The problem is often referred to as first order autocorrelation.  First order autocorrelation is reported for: a) positive, b) negative, and c) both positive and negative autocorrelation.  This is reported beginning on row 20 of the ANOVA tab.

If the Durbin-Watson coefficient indicates the presence of first order autocorrelation (Do Not Reject), try one of the following data differencing techniques. First try Durbin adjustment.

WinORS automatically computes the Durbin adjusted data.  Follow the menu tree: /SC Durbing Adjusted.  The data from the last OLS solution is the only data in the file.  Simply open the file and solve using the Application menu selection.

How does it work?  The method uses the estimate of the autocorrelation rho (r) from the regression Analysis of Variance Table to adjust the dependent and independent variables:

                        Y_it  =  Y_it - rY_{it -1}

                        X_it  =  X_it - rX_{it -1}

A more traditional data adjustment is to ignore rho and simply use first differences.  This is the same as setting rho equal to 1.  The adjusted data for this method is also available through the most recently opened file list.  The file name is: FIRSTDIF.OR?.