Re: [Gretl-devel] no asymmetry in the new "band" plotting option?
On Wed, 3 Aug 2016, Sven Schreiber wrote:
Am 03.08.2016 um 01:54 schrieb Allin Cottrell:
> On Wed, 3 Aug 2016, Sven Schreiber wrote:
>>
>> But (confidence) bands are not symmetric in general. So I would vote
>> for the possibility to specify arbitrary center/low/high triplets
>> instead. Or am I missing something with this new feature?
>
> No, I don't think you're missing anything. We're taking the symmetric
> case as paradigmatic, and hoping that it's easier for experts to
> "reverse-engineer" this, as you say, than for ordinary users to bother
> with expanding the symmetric case to lower and upper.
OK, so when this gets documented (or is it already, apart from mailing list
posts?) and before the syntax is extended there should be a sentence about
this reverse engineering. Perhaps it could go in the "cheat sheet" section.
The documentation is up to date. Putting something in the cheat sheet
is probably a good idea. For now, here's a little example of forcing
an asymmetric band into the format accepted by gretl:
<hansl>
nulldata 128
setobs 1 1 --special-time-series
series x = 0.1*time + filter(0.1*normal(), 1, {1.8, -0.9})
# an asymmetric band around x
series top = x + uniform()
series bot = x - 2 * uniform()
# convert to mid-point and deviation
series mid = (top + bot)/2
series dev = top - mid
gnuplot x --band=mid,dev --time-series --with-lines \
--output=display
</hansl>
Allin