0

I have two related questions.

Question 1

I ran a Durbin-Watson test on my dataset and I got the following output:

It says the autocorrelation value is 0.04757039. But what does it mean?

Question 2

My data size is really big (about 62000), so the p-value becomes 0 even though there is no autocorrelation. I don't really believe the test result, so I want to know is there a way to check autocorrelation for such a big data set? (manually or graphically).

Karolis Koncevičius
  • 4,282
  • 7
  • 30
  • 47
Venna
  • 1
  • Venna, welcome to Cross Validated! If you find my answer helpful, you may accept it by clicking on the tick mark to the left. If it is lacking, you may ask for clarification by commenting. This is [how Cross Validated works](https://stats.stackexchange.com/tour). – Richard Hardy Feb 12 '20 at 14:02
  • So what do you think about my answer? – Richard Hardy May 02 '20 at 06:33

1 Answers1

1

  1. The result of durbinWatsonTest is an object of class durbinWatsonTest. I cannot find the description of the durbinWatsonTest class, but I guess Autocorrelation is the point estimate of first-order autocorrelation of the time series, since the test statistic $$ DW\approx 2(1-\hat\rho), $$ (see "Understanding the Durbin Watson test") and we have $1.904856\approx 2(1-0.04757039)$.

  2. The test works well asymptotically when its assumptions are satisfied. If you do not believe the answer, then either your sample is atypical for the data generating process (DGP) or some of the assumptions have been violated, e.g. perhaps your DGP is changing over time? Also, mind the difference between statistical and subject-matter significance. Perhaps the autocorrelation is truly present in the DGP, but it is small enough so that you can judge it negligible from the subject-matter viewpoint.

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219