Re: Scoring Probabilistic Predictions.
- From: Hatunen <hatunen@xxxxxxx>
- Date: Fri, 29 Feb 2008 13:52:13 -0700
On Fri, 29 Feb 2008 10:16:18 -0800 (PST), Petra
<petrasrcf@xxxxxxxxxxx> wrote:
On Feb 28, 3:04 pm, rog...@xxxxxxxxxxxxxxx wrote:
On Feb 28, 3:59 pm, Petra <petras...@xxxxxxxxxxx> wrote:
On Feb 28, 1:44 pm, rog...@xxxxxxxxxxxxxxx wrote:
On Feb 28, 11:16 am, t...@xxxxxxxxxxxxx (Thomas A. Russ) wrote:
OK, a new methodology question here.
What would be the proper way to score probabilistic predictions?
What I mean is a prediction of the form, "There is a p% chance of event
E occurring during time period T."
How does one score such a prediction, and how does one determine
statistical significance?
Off the top of my head, it would seem that one would need many more
events and predictions in order to establish statistical significance
when one has such probabilistic predictions.
Presumably the null hypothesis is the background chance of the
particular event occurring. The test hypothesis is the difference
between the background probability and the predicted probability.
So, if to take a concrete example, let us suppose that for some event E,
the background probability is 30% and a prediction says there is a 65%
chance of E occurring during some given time period. The test would be
to see if the additional 35% probability is statistically significant.
Does anyone have any particular ideas on what that computation would
look like?
Hmmm. A bit of Googling found this abstract, but I don't have access to
the full text.
<http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&ha...>
Also: <http://tinyurl.com/yw9su3>
--
Thomas A. Russ, USC/Information Sciences Institute
Dr Russ;
Personally, I ignore such probability statements because most of them
are just wild ass guesses. The definition of a prediction includes
probability but most predictors give confidence estimates instead.
Roger- Hide quoted text -
- Show quoted text -
Roger,
I guess that's why you're not a scientist, because study upon study
shows clearly which faults have high probabilities. Why don't you
talk to David Schwartz about the Rodgers Creek Fault or perhaps Jim
Leutgart (sp?) about the Hayward Fault. Then you might be better
informed. For Gosh stakes these people know what they're talking
about. They can't predict exactly when, but believe me the words
"soon" and "2008 is the Year" for the Hayward Fault are right on the
mark.
The probably is that you're out of circulation. You don't attend
meetings of any kind or keep in touch so you've lost touch with what's
current.
Petra
Petra;
I was responding to Dr Russ, not you and we were not talking about
scientists.
Of course I know that different faults have different probabilities.
What we don't know is how close they are to failure.
Roger- Hide quoted text -
- Show quoted text -
Roger,
You said " Of course I know that different faults have different
probabilities.
What we don't know is how close they are to failure"
The answer is not "we" to begin with because you're not a scientist
and secondly "you" don't know when a fault is going to fail because
you don't have any knowledge of such things nor do you have any
equipment that works in that regard either.
Unfortunately, you, Petra, are not a scientist either. So how are
we to take your claims?
Despite your endless attempts to minimize what I can do you fail to
mention I'm right 80% of the time which only leaves 20% when I'm not
right.
That would be astonishing. Too bad you won't publish anything to
support that claim.
That's better than anyone else has done to date
Damn right, because nobody has been even close to right. Why is
that, I wonder.
and if you
think you can do better, I suggest you get out of your chair and try
going for a six mile hike everyday, eat a lot of carrots and celery
and work out when you're not walking. It improves brain cell capacity
and affords one new ideas which equates to solving problems better.
Ah. Petra's sesret of success, at last! Go for a six mile hike
and eat lots of carrots and celery and work out. Then you, too,
will be able to predict earthquakes will happen in well-known
fault zones, not to mention snow for Minneapolis next winter,
tornados in Nebraska and hurricans in the Atlantic. Now if you
could just eat someething that will allow you to predict what day
these events will occur more than a day or so ahead... Maybe
eating turnips would do it.
Petra, if you really want to do the nation some good, predict,
within a month, the next 8+ earthquake on the New Madrid fault.
Otherwise, Memphis and St Louis are going to be in a world of
hurt one of these days.
You need 92.7 Energy Cardio Radio with Craig and Fernando every day!
You can get it on your computer, go for it!
Hey. I'm not going to get any sort of crap like that on *my*
computer. I'm very careful to set my cup down some distance from
the computer.
--
************* DAVE HATUNEN (hatunen@xxxxxxx) *************
* Tucson Arizona, out where the cacti grow *
* My typos & mispellings are intentional copyright traps *
.
- References:
- Scoring Probabilistic Predictions.
- From: Thomas A. Russ
- Re: Scoring Probabilistic Predictions.
- From: rogerh
- Re: Scoring Probabilistic Predictions.
- From: Petra
- Re: Scoring Probabilistic Predictions.
- From: rogerh
- Re: Scoring Probabilistic Predictions.
- From: Petra
- Scoring Probabilistic Predictions.
- Prev by Date: Re: Supercomputer Unleashes Virtual 9.0 Megaquake in Pacific Northwest
- Next by Date: Re: Seeking Common Ground in Earthquake Prediction Delivery
- Previous by thread: Re: Scoring Probabilistic Predictions.
- Next by thread: Re: Scoring Probabilistic Predictions.
- Index(es):
Relevant Pages
|
