A measure of first-order autocorrelation, that is, of autocorrelation in time series where independent variables are not lagged. (Autocorrelation is sometimes referred to as serial correlation).
The Durbin-Watson statistic is defined as:
where T is the total number of periods and e is defined as in R Square.
The Durbin-Watson statistic lies in the range 0-4. A value of 2 or nearly 2 indicates that there is no first-order autocorrelation. An acceptable range is 1.50 - 2.50.
Where successive error differences are small, Durbin-Watson is low (less than 1.50); this indicates the presence of positive autocorrelation. Positive autocorrelation is very common. Where successive error differences are large, Durbin-Watson is high (more than 2.50); this indicates the presence of negative autocorrelation. Negative autocorrelation is not particularly common.
In time series with lagged variables, the Durbin-Watson statistic is unreliable as it tends toward a value of 2.0.
Autocorrelation occurs when the error terms of a regression model are not independent; that is, when the values of historical periods in the forecast model are influencing the values of current periods. Time series with a strong seasonal or cyclical pattern are often highly correlated. High autocorrelation means that MLR using the ordinary least squares method is not a suitable forecasting technique for this data.