Test for autocorrelation by using the Durbin-Watson statistic

Use the Durbin-Watson statistic to test for the presence of autocorrelation in the errors of a regression model. Autocorrelation means that the errors of adjacent observations are correlated. If the errors are correlated, then least-squares regression can underestimate the standard error of the coefficients. Underestimated standard errors can make your predictors seem to be significant when they are not.

For example, the errors from a regression model on daily stock price data might depend on the preceding observation because one day's stock price affects the next day's price.

The Durbin-Watson statistic (D) is conditioned on the order of the observations (rows). Minitab assumes that the observations are in a meaningful order, such as time order. The Durbin-Watson statistic determines whether or not the correlation between adjacent error terms is zero.

To get a conclusion from the test, you can compare the displayed value for the Durbin-Watson statistic with the correct lower and upper bounds in the following table from Savin and White¹. If D > D_U, no correlation exists; if D < D_L, positive correlation exists; if D is in between the two bounds, the test is inconclusive. The table provides values to test for first-order, positive autocorrelation. The significance level for the test is 0.05. The table is for models with an intercept.

You can also use this table to test for first-order, negative autocorrelation. The test statistic is (4 – D). If (4 – D) > D_U, no correlation exists; if (4 – D) < D_L, negative correlation exists; if (4 – D) is between the two bounds, the test is inconclusive. ²

To calculate the Durbin-Watson statistic, choose Stat > Regression > Regression > Fit Regression Model, click Results, and check Durbin-Watson statistic.