Re: [S] Problem with predict()

John Maindonald (
Tue, 26 May 1998 09:22:14 +1000 (EST)

Bill Venables wrote
> In my view the correct procedure is always to provide a SAFE
> technique as the default and if there is a quick risky alternative
> you provide that as an alternative for the desparate or the
> foolhardy. You should not provide a risky technique as the
> default, especially one that fails without any noise or smoke.

I strongly agree. Once one has an algorithm that works, one can
work over it to make it efficient. If one starts with an algorithm
that does not work, not even in principle, the fudging and patch that
goes on inevitably produce an inefficient mess.

> Unequivocally you should go against the global S mistake and do
> the sensible, safe, correct thing.

John Maindonald email :
Statistical Consulting Unit, phone : (6249)3998
c/o CMA, SMS, fax : (6249)5549
Australian National University
Canberra ACT 0200
This message was distributed by To unsubscribe
send e-mail to with the BODY of the
message: unsubscribe s-news