numeric evidence for independence of errors
tests for serial correlations between errors in regression models
varies from 0 to 4, 2 means the residuals are uncorrelated
<2 negative correlation >2 positive correlation
size depends on number of predictors in the model and the number of observations
Durbin & Watson (1951) has table of values
values less than 1 or greater than 3 are definitely problematic
Was this helpful?
0 / 0