Hello- I am analyzing archival data from a University. Multiple Linear Regression: Dependent variable (continuous) time to completion measured in in years Six independent variables (college engagement factors measured at continuous and nominal variables). Each case represents a unique student record. I am looking for rational to disregard the Durbin-Watson Statistic (.084) as all the other 7 assumptions of a multiple linear regression are met.
In addition to your expertise, I'm looking for a peer reviewed article or text book to cite to justify my decision to disregard this DW test.
R (.261)
R Square (.068)
Durbin Watson (.084)
The autocorrelation of errors should not be ignored since in their presence the OLS estimators remain unbiased and consistent but no longer efficient (they no longer have minimal variance). Failure to comply with this assumption also invalidates the estimates of standard errors and the t and F tests.
- Answering your question, you can use a generalized linear model. This does not require compliance with that assumption.
- The other option may be to solve the problem, using Cochrane-Orcutt method.
Cochrane, D.; Orcutt, G. H. Application of least squares regression to relationships containing auto-correlated error terms. Journal of the American Statistical Association, 1949; 44(245): 32-61. 10.2307/2280349
Regards !
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression