On Sun, Feb 22, 2009 at 12:51 AM, Jared ''Danger'' Earle <[email protected]> wrote:
> On 21 Feb 2009, at 23:45, LuKreme wrote:
>> I only have two results, yes and no (or 'positive' and 'negative'),
>> I'm not seeing how graphing it is going to be useful.  I know 19/78,
>> the order of those 19 can't possibly matter...
>
>
> You have 78 results, the last of which is 19.
>
> Figured out the graph yet?
>
> Oh, and read this: http://en.wikipedia.org/wiki/Standard_deviation

just as I was thinking of heading off to bed after watching "Le
scaphandre et le papillon" <http://www.imdb.com/title/tt0401383/>...

...BZZZT!...

>From Lewis' info, its impossible to say anything at all about the
probability beyond his initial report that he got 19 out of 78 -  and
standard deviation does not enter into the equation as it stands.

>From the info given, there are two outcomes "positive" and "negative".

He reports 78 tests from which he got "19" positive results and he
states that he does not know the mechanism that controls the outcome.
Since he does not know the mechanism, he does not know whether the
order of his test is relevant (recursion ? - its software after all),
so any "curve-fitting" would be highly speculative, since we would not
be able to assess the appropriateness of the parameters.

OK. One could plot your suggested graph for y (number of tests
completed) v. x (sum of positive results for y) and we might get
something that approximates a staircase that approximates a straight
line with smallish residuals, but will this really answer Lewis'
question any better than his current best guess of 19/78 ? My answer
is, based on the available information, no.

If we want to get std. dev. involved, then at least two more of us
would need to repeat the test and report out results. We might then
have e.g. Lewis: 19/78, Jared 26/78 and Mark 18/78 and we could start
to do some stats.

We could start thrashing around in the dark and look at Lewis results
using non-parametric methods and maybe even start looking at some meta
results (e.g. how often does a postive result follow two consecutive
negative results (software again)), but, as the data stands, the
sensible thing to do is to repeat the experiment before trying to
establish anything.

Instead of Wikipedia, in this case I would cite Rosencrantz and
Guildenstern are Dead (:

mark.
_______________________________________________
OSX-Nutters mailing list | [email protected]
http://lists.tit-wank.com/mailman/listinfo/osx-nutters
List hosted at http://cat5.org/

Reply via email to