Вы находитесь на странице: 1из 1

1

Durbin Watson Test

The Durbin-Watson statistic is a simple numerical method for checking serial


dependence. Let rk be the residuals sorted into time order. Then the DurbinWatson statistic is:
Pn1
(rk rk+1 )2
DW = k=1Pn
.
2
k=1 rk
We have the following criterion:
DW
Decision

<2
Positive Serial Correlation

2
NO serial correlation

>2
Negative Serial Correlation

As a rough rule, serial correlations corresponding to DW outside the range


1.5 2.5 are large enough to have a noticeable effect on our inference techniques.
Note that DW itself is random and may be outside the range 1.5 to 2.5, even if
the errors are uncorrelated. For data sets with long runs of units from the same
treatment, the variance of DW is a bit less than 4/N.
In R, the durbin-watson test is located in library(lmtest):
> dwtest(f ormula) # z time index.

Figure 1: Durbin Watson Test

Вам также может понравиться