# what is a good value for Durbin-Watson test in regression models?

I have a D-W = 1.312 with a sample of 22 cases.

Is it a good value for running a multiple linear regression model?

A DW of 1.312 suggests you have some positive auto-correlation. A "good" value is 2. Whether a value of 1.312 is a problem depends on your number of predictors. If you have 1 or 2 predictors (excluding the intercept), then your value is above the upper bound and you can't reject the null hypothesis (i.e., the DW is "OK" at the 0.05 level of significance). If you have more than 2 predictors you may have a problem. Of course, the sample size is small so the test has low power, and there may be a benefit of bootstrapping. See here for more information.

If the software that you use reports the p-value, you can use it to interpret the test results. If the p-value is smaller than 0.05, you can conclude that there is no autocorrelation between your residuals.

• The null hypothesis of the DW statistic is $\rho=0$, i.e. no serial correlation. A p value of less than 0.05 implies rejection of the null, which means serial correlation being an issue. – Durden Aug 14 at 17:05