StatReasonWhat this answers
The Durbin-Watson statistic answers "do this model's residuals show a pattern where one residual predicts the next, rather than looking like independent noise?" This matters specifically for data with a genuine order, such as a time series, where a regression model's standard error calculations assume independent residuals; correlated residuals can make reported significance levels unreliable.
How it is calculated
The statistic is the sum of squared differences between each residual and the one immediately before it, divided by the sum of squared residuals overall. Its value always falls between 0 and 4. A value near 2 suggests no strong first-order autocorrelation; a value well below 2 suggests positive autocorrelation (residuals tend to repeat their neighbor's sign); a value well above 2 suggests negative autocorrelation (residuals tend to alternate sign).
Worked example
For the ordered residuals 1, 0, and negative 1: the squared differences between consecutive residuals sum to 2, and the sum of squared residuals is also 2, giving a Durbin-Watson statistic of exactly 1, in the range this calculator flags as suggestive of positive autocorrelation.
Assumption audit
What this result does not mean
A Durbin-Watson value near 2 does not prove the residuals are truly independent, only that this specific diagnostic found no strong evidence of first-order autocorrelation. It cannot detect other patterns of dependence, such as correlation between a residual and a value two or more steps earlier.
Limitations
This calculator reports the raw statistic without an exact critical-value table, since those tables depend on the number of predictors in the original model, which is not an input here. Use the rough reading as a starting signal, and consult a formal Durbin-Watson table referencing your model's specific predictor count for a rigorous test.