Sorting NaNs

[Richard Damon]

On 6/10/18 3:28 PM, Anders Munch wrote:
> Richard Damon wrote:
>
>> The two behaviors that I have heard suggested are:
>>
>> 1) If any of the inputs are a NaN, the median should be a NaN.
>> (Propagating the NaN as indicator of a numeric error)
>>
>> 2) Remove the NaNs from the input set and process what is left. If
>> nothing, then return a NaN (treating NaN as a 'No Data' placeholder).
>
> 3) Raise an exception.
>
> I can't believe anyone even suggested 2).  "In the face of ambiguity,
> refuse the temptation to guess."
>
> regards, Anders
>
Many people doing statistics (mis-)use 'NaN' as a flag for 'missing data
for this record'. If you want the statistic of a sample as an estimate
of the statistic of the population, then omitting 'missing data' can be
a reasonable option. I will agree that I see issues with this attitude,
but it isn't uncommon.

-- 
Richard Damon