[Gretl-users] Re: Treatment of nan by the quantile function
On Mon, 15 Mar 2021, Sven Schreiber wrote:
Am 14.03.2021 um 03:50 schrieb Alecos Papadopoulos:
> How does the quantile function treats/handles nan values in a series
> (and/or matrix column if the treatment is different)?
> ...
> It appears strange to have a proper number value for the 0.8-quantile
> and for the 0.9-quantile so that the inequality condition can be
> checked, but not for the 0.85-quantile.
Hm, something seems weird with quantile and missings:
<hansl>
m = seq(1,5)'
eval quantile(m, 0.5) # 3, correct
m[4] = NA
print m
eval quantile(m, 0.5) # 1, weird
</hansl>
Hmm, apparently the back-end for quantile() with a matrix argument
was coded on the assumption that any NAs would be screened out in
advance, but that wasn't actually done anywhere. That's now fixed in
git (NAs are skipped). Snapshots will follow tomorrow.
Allin