4

Given the Durbin Watson statistic, what do I need to calculate the p-value? Is there a formula I can use?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
  • Instead of looking for a formula, try searching for a standard Durbin-Watson significance table. – rolando2 Feb 07 '15 at 01:44
  • Welcome ChetanMV, There are a couple of related statistics you might mean [see here](https://en.wikipedia.org/wiki/Durbin%E2%80%93Watson_statistic). The first step is to be clear about which statistic and circumstances we're dealing with. – Glen_b Feb 07 '15 at 02:07
  • The function `dwtest` in the R package [lmtest](http://cran.r-project.org/package=lmtest) returns the Durbin-Watson statistic and a p-value. You may look at the source code of that function and see the part where the object `pval` is obtained. – javlacalle Feb 07 '15 at 12:23

2 Answers2

5

The answer is NO. Unfortuneately, the null distribution of the DW test statistic in regression analysis depends on the design matrix $X$. This is discussed in for instance Seber and Lee section 10.4.4. There is also referenced some approximate procedure, one based on upper and lower bound $D_L \le DW \le D_U$ which have distributions not depending on $X$ so can be tabled.

A modern alternative is to use simulations, the R package car implements this (using bootstrapping) in the function durbinWatsonTest.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
0

Critical values given n and k for .01 p https://www.real-statistics.com/statistics-tables/durbin-watson-table/

thistleknot
  • 147
  • 6