Re: Re: Rounding??

[Guenter Milde <G Milde physik tu-dresden de>]

On Tue, 12 Jun 2001 18:02:36 -0400 wrote James Ball <sybex009 madnetter com>:

> On Jun 12, 2001 -  5:38 PM Morten Welinder wrote:
> > An accountant will want "0.365" rounded to "0.36".  A mathematician
> > will want "0.37". 

It's the other way round. (At least as we learn it in school in Germany):
The accountant always rounds up and the mathematican (and scientist using
mathematics) will use the "Geradzahlregel" - rounding to the even number so
rounding errors expectation value goes to zero. (cf. the mail from : Kjetil
Torgrim Homme) (Of course it could also be "round to odd" - that is just a
convention.

Furthermore, the rule says: If you happen to know that the trailing 5 is a
result of rounding up: round down (and vice versa).

Therefore some table books have the convention to show whether a value like
3.65 is gained by rounding down or rounding up by putting a dot above or a
stroke below the 5. This way it is possible to take the previous rounding
into account when doing further rounding. The essence of this is, if we do
the rounding just for displaying, we should take the precise number as base
for computations/further rounding (and warn the user that something
unexpected might happen becouse of this (not only in NEWS but also in Help)

> Couldn't you have a preferences sort of dialog that has an option like
> "Round up/Round down)

I'd prefer something like:

Rounding behaviour:
  [ ] Round up
  [ ] Round down
  [x] Round even
  [e-3] epsilon
  
With "epsilon" - to set the scope
of the "1.5 - epsilon -> 1.5 -> 2" rounding procedure. 
(e.g. if epsilon := 0.0001: 1.49999 -> 1.5 -> 2 versus 1.499 -> 1)


Guenter

--
G Milde physik tu-dresden de