Regarding'Durbin Watson'

Home > Statistics Every Writer Should Know > The Stats Board > Discusssion

Regarding'Durbin Watson'
Message posted by Huiying Wang on July 31, 2000 at 12:00 AM (ET)

Would someone help me by providing information of ' Durbin Watson' in terms of the concept, applicability and application.

Thanks and regards.

READERS RESPOND:
(In chronological order. Most recent at the bottom.)

Re: Regarding'Durbin Watson'
Message posted by nancy diehl on July 31, 2000 at 12:00 AM (ET)

A problem somtimes occurs in a multiple regression model when the data involves
time series data. You will find this in models built to examine sales forecasting
or in econometric studies. When one or more key variables have been omitted from the
multiple regression regression model for time series data, the residuals may not behave
in a random fashion, but actually display dependency. When this happens the independent
variables are said to be "autocorrelated" or serially correlated. Serial correlation affects
the precision of the estimation of the coefficients (or betas) in the multiple regression
model. When serial correlation is present the true variances of the betas are
underestimated which causes the t-values to be larger than they should be, hence some of the
beta parameters may become statistically significant when in fact they aren't.
The Dubin-Watson test is a common test used to detect the presence of
serial correlation. For text reference, see Neter & Wasserman book: Applied Linear Regression Analysis.

Lastly, autocorrelation can be detected through some diagnostic plots.
A simple plot of residuals plotted against time is helpful. If unusually large number of residuals
with the same sign tend to cluster together, then there is an indication of positive correlation.
Rapid changes in sign signify a negative correlation..

Your $5 contribution helps cover part the $500 annual cost of keeping this site online.