Which factors affect the Durbin-Watson statistic value?

Which factors affect the Durbin-Watson statistic value?

• A. The t-statistics of coefficients
• B. The skewness of the residuals
• C. Number of predictors in the model and number of observations
• D. The sample mean

Answer: (C)

The Durbin-Watson statistic measures whether residuals from an ordinary least squares regression are serially correlated from one observation to the next. It’s calculated using the residuals and is centered around 2 when there is no first-order autocorrelation; values tend to drop below 2 with positive autocorrelation and rise above 2 with negative autocorrelation. What drives the actual value you get, and the interpretation near the critical cutoffs, are the sample size (how many observations you have) and the number of predictors in the model (how many estimated parameters you’ve included). The exact critical values used to judge whether there is autocorrelation depend on both n and k, so these two quantities shape the range and interpretation of the statistic, not the individual coefficient t-stats, the skewness of residuals, or the residuals’ mean in the traditional sense. The t-statistics reflect significance of coefficients, skewness describes distribution shape, and the mean of residuals is typically zero with an intercept in OLS, none of which determine the DW value itself.

The Durbin-Watson statistic measures whether residuals from an ordinary least squares regression are serially correlated from one observation to the next. It’s calculated using the residuals and is centered around 2 when there is no first-order autocorrelation; values tend to drop below 2 with positive autocorrelation and rise above 2 with negative autocorrelation.

What drives the actual value you get, and the interpretation near the critical cutoffs, are the sample size (how many observations you have) and the number of predictors in the model (how many estimated parameters you’ve included). The exact critical values used to judge whether there is autocorrelation depend on both n and k, so these two quantities shape the range and interpretation of the statistic, not the individual coefficient t-stats, the skewness of residuals, or the residuals’ mean in the traditional sense. The t-statistics reflect significance of coefficients, skewness describes distribution shape, and the mean of residuals is typically zero with an intercept in OLS, none of which determine the DW value itself.