Re: Pearson-r in SQL

On Thu, 23 Dec 2004 16:45:54 -0800, "Mikito Harakiri" <mikharakiri_at_iahu.com> wrote:

>"Matthias Klaey" <mpky_at_hotmail.com> wrote in message
>news:f6oms0heulhqar17hlfnbgqhdpgaj0s6bs_at_4ax.com...
>> On Thu, 23 Dec 2004 15:00:51 -0800, Gene Wirchenko
>> <genew_at_mail.ocis.net> wrote:
>> > I would toss it. It is not valid data.
>>
>> What if the cost of getting one measurement was, say 37000 Dollars?
>> Even if you could get only the x result and not the desired (x, y)
>> pair, due to some experimental mishap?
>
>What the cost has to do with mathematical definitions?
>
>The values
>
>(3kg, NULL)
>(5kg, NULL)
>(7kg, NULL)
>(NULL, 1m)
>(NULL, 2m)
>(NULL, 3m)
>(NULL, 4m)
>
>are orthogonal. There is no corellation is this data set whatsoever.

The cost has nothing to do with the mathematical definition, but only with practical considerations. Your example is "pathological* in the sense that there is no posssibility to compute a correlation on a part of the data set that would allow a reasonable estimation of the missing values.

Greetings
Matthias Kläy

-- 
www.kcc.ch