Re: Determining Confidence Interval
From: Paige Miller (paige.miller_at_kodak.com)
Date: 09/29/04
- Next message: yuliana: "does anyone know how to get durbin-watson table for n=240"
- Previous message: Dr. Knud Werner: "Re: Mann-Kendall test statistic"
- In reply to: Roger Sherman: "Re: Determining Confidence Interval"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 29 Sep 2004 08:40:32 -0400
Roger Sherman wrote:
> Thanks for indulging my ignorance. Put as succinctly as I can and
> allowing for my limited knowledge of statistics, given a "specific
> result" (number of events or instances out of a total number of
> occurrences), how large a sample of total occurrences will be required
> to obtain a high level of confidence in the "specific result."
>
> More to the point, in the play of cards, "specific events" take place.
> For the sake of this exercise, I have determined arbitrarily that out
> of the total number of occurrences (not defined at this point), the
> "specific events" take place 20 % of the time. I similarly could have
> expressed the situation in this way: Out of 100 total occurrences,
> the "specific events" occurred 20 times. Can I have a high level of
> confidence that this result reflects a consistent trend or would I
> need more than 100 total occurrences (200, 300 ….1000) in order to
> have confidence that the "specific events" take place 20% percent of
> the time ??
>
> Finally, I would appreciate any input you could provide regarding the
> methodology that could be used to sort out the issues raised above.
Let's back up. Confidence intervals are appropriate when you have
some data, and you estimate something (often a mean or a percent)
from the data.
So, when you say ... a certain event in a game of cards occurred 20
times out of 100 ... that certainly could be the thing you want a
confidence interval for. However, the questions you are asking still
are way off the mark for what statistics can tell you.
A confidence interval would tell you that although you observed 20
out of 100, repeated sampling would provide results in the range x
to y 95% of the time (or some other percent). An hypothesis test
would tell you if the data you actually observed was consistent (or
not) with a true event rate of z times out of 100.
Your question: "Can I have a high level of confidence that this
result reflects a consistent trend or would I need more than 100
total occurrences (200, 300 … 1000) in order to have confidence that
the 'specific events' take place 20% percent of the time ??" This is
meaningless from a statistical point of view. We don't say we have
confidence in an event, or in a sequence of events; sometimes we say
that we have a certain level of confidence in an hypothesis test, or
that we have a confidence interval about our observation. If your
hypothesis is that the true probability of this event is 0.2 (20 out
of 100), then you can have as high a level of confidence as you
desire, but this is backwards ... we don't observe a value and then
make that value our hypothesis, we state a hypothesis first and then
see how well the data conforms to that hypothesis.
Basically, any introductory statistical text will cover hypothesis
testing, confidence intervals and sample size in much more detail
than I can cover it here.
-- Paige Miller Eastman Kodak Company paige dot miller at kodak dot com http://www.kodak.com "It's nothing until I call it!" -- Bill Klem, NL Umpire "When you get the choice to sit it out or dance, I hope you dance" -- Lee Ann Womack
- Next message: yuliana: "does anyone know how to get durbin-watson table for n=240"
- Previous message: Dr. Knud Werner: "Re: Mann-Kendall test statistic"
- In reply to: Roger Sherman: "Re: Determining Confidence Interval"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
