Probability modelling question

[Claire Blair]

I have a basic question, which I hope someone can answer for me.

I have sample data that consists of binary (Yes/No) data. The Yes's
correspond to a 'success' event, and the No's correspond to a 'Failure'
event.

I want to know how I can use the data to model and predict the
probability of a success event.

I think the logistic model is what I should use, but I am not sure.

Please advise.

[Nasser Abbasi]

"Claire Blair" <no.r...@here.com> wrote in message
news:GZCdnScOP_M5u93a...@bt.com...

I think what you want is to fit a probability distribution to your data?

Assuming your X's (events) are i.i.d. you could try to fit Binomial or may
be Poisson? (but with Poisson, we need n to be large and p very small to get
good approximation to the binomial).

So, For Binomial, using Maximum likelihood, p comes out to be X_bar, i.e.
the probability which maximizes the likelihood of observing your data is
this probability. (X_Bar is the mean of the sample data). For Poisson,
also using Max. likelihood, p is X_bar.

Now you can use your 'model' distribution with the above parameter to
calculate other probabilities.

Nasser

[richar...@comcast.net]

On Sun, 18 Nov 2007 11:46:40 +0000, Claire Blair <no.r...@here.com>
wrote:

Assuming that by "model" you mean you have explanatory variables that
you think help predict success or failure, then yes, a logistic model
(or a probit) is appropriate.
-Dick Startz

[deepa....@gmail.com]

On Nov 18, 10:08 pm, richardsta...@comcast.net wrote:
> On Sun, 18 Nov 2007 11:46:40 +0000, Claire Blair <no.re...@here.com>

I think u can do a run test which gives u the probability u are luking
for