Durbin-Watson test

Hi, All: the concept and statistic of the test is simple. However, the decision rule is complicated, do we need to memorize it for exam? thanks

Yes, I would not take any chances.

Remember the formula 2(1-r) and the decision rule is easy. High (positive) correlation ® makes the equation negative. Low (negative) correlation ® makes the equation positive. Then just remember the ranges. 0, dl, du, 4-du, 4-dl, 4 and the respective meanings: positive, can’t tell, fail to reject, can’t tell, negative

The Durban Watson statistic tests for serial correlation in the regression errors. However for the LOS in that reading (Reading 12, Vol1) they ask “discuss… and the effects of… and serial correlation on statistical inference.” I have been studying according to the LOS, and in the above case I ignored the computation of the Durban-Watson stat. Can anyone give an argument in favour of learning this material (and similar in other instances). Also, kudos to ‘the show ny’ for the concise explanation.

you may not see a computation question but i have seen before (including in CFAI questions) where you have to compare dw stat to dw crit so you need to know the ranges and meanings.

I really don’t think we have to remember the entire table. Our focus should be more on d1 and du which prove posive and negative serial correlation. the others don’t seem to be consequetial to LOS.

I agree… the concept is captured well by dl and du, and the questions I have seen only go that far. Most important thing is not to confuse “r” the correlation between the variables, which can be positive or negative with the error correlations, which can also be positive or negative, as someone did above.