Independence of observations
An important assumption for the simple linear regression model is the independence of errors. In many time series models, this assumption is violated because of the correlation of errors in successive observations. This is termed to as autocorrelation.
Autocorrelation occurs if a positive error is followed by another positive error and a negative error is followed by another negative error. If autocorrelation occurs then time should be considered as an important independent variable and therefore time varies analysis should be used.
We can use Durbin Watson ‘D’ statistics to determine whether observations are independent.
D = ∑(ei – ei-1)2
∑ei2
Where:
ei is the error in time i
ei-I is the error in time i-I
The Durbin Watson statistics provides a measure of association between successive values of the error term. The computed statistics is compared against two tabulated values du and dl that depend on the desired confidence level of the test and the degrees of freedom of the data.
If the computed Durbin Watson “D” statistics is greater than Du, then we can conclude that there’s no positive correlation between error terms.
If dl ≤ D≤ du then the test is inconclusive and therefore we can neither accept nor reject the null hypothesis.