On Sat, 17 May 2014, oleg_komashko@ukr.net wrote:

> (The regression is incorrect from econometric point of view, just to give an easily reproducible example)
> open /usr/local/share/gretl/data/misc/denmark.gdt
> ols LRM const LRY
> dw=$dwpval
>
> (here Durbin-Watson = 0.298798)
>> From script output:
> ? dw=$dwpval Imhof integral gave negative value
> Error executing script: halting > dw=$dwpval
> May be it worth to include into the function something like this:  if dw<dw0 (or if  Imhof <=0) or pvalue<p0  print( pvalue<e-N )for some N, say e-16?
> R ouput for the same dataset (dwtest{lmtest})
> DW = 0.2988, p-value < 2.2e-16

Please don't re-post the same question. We got your first two posts, 
and it's the weekend...

But anyway, I don't think I like the proposal very much. The $dwpval 
accessor by its nature returns a definite value: there's no way it 
can say "< 2.2e-16" or some such (and besides, I'm not sure where 
that upper bound would come from).

What we could do (and might be an improvement) is to return NA 
instead of generating an error if we get a negative Imhof integral, 
and to say in the documentation that this probably indicates that 
the true p-value is "very small", although we can't quantify it.

Allin Cottrell