Is Donald Trump Done for?

So a prediction cannot be a prediction unless it turns out to be correct? An odd use of language.

It's not a prediction unless it can be tested. Maybe it is possible to look back and determine if there actually was a 70% chance of rain yesterday, but that is well beyond what is being discussed here.
 
It's not a prediction unless it can be tested. Maybe it is possible to look back and determine if there actually was a 70% chance of rain yesterday, but that is well beyond what is being discussed here.

It's difficult for a single event, but over a large number of predictions you can see in a pretty straight forward way if the probabilities hold up or not.
 
It's difficult for a single event, but over a large number of predictions you can see in a pretty straight forward way if the probabilities hold up or not.


In other words, your entire diatribe about how the predictions of a single event, which were couched as probabilities, being 'proven wrong' by the less probable outcome coming to pass, is now disavowed?
 
Colloquially people would likely say you 'got it right', strictly speaking, how many alternative realities did you check?

I am checking all the alternative realities that I have access to, each and every time, proving that I am far more diligent in the effort to upgrade my results than those people who freely admit that they do not check even one alternative reality.
 
In other words, your entire diatribe about how the predictions of a single event, which were couched as probabilities, being 'proven wrong' by the less probable outcome coming to pass, is now disavowed?

Someone hasn't been paying attention
 
I'm betting lots of people haven't been paying attention. This entire exchange has been dismally dull.

Agreed. It's just funny that you think I was saying something I specifically said wasn't true from my very first post and have restated several times since.
 
My bad...missed that when my eyes rolled back in my head at yet another "debate" over minutia of statistics, college freshman style.
 
Yeah, a claim about a future event or outcome, right, I said event. Which includes probabilistic predictions. "Tomorrow will be Friday" I suppose is a kind of generic prediction, silly of course and outside of probability theory.

I mean, I gave you the wiki and more importantly we have the original site in question, which happens to be one of the most popular forecast sites around, 538, calling his probabilistic forecasts predictions... like, what more do you want?

I would like you to stop using stupid methods for defining things. Like Wikipedia, or website headers, neither of which stands in as a reasonable way to define a word.

It's difficult for a single event, but over a large number of predictions you can see in a pretty straight forward way if the probabilities hold up or not.

Which is exactly what we have been saying all along. Well, actually, it's completely impossible to tell from a single event whether a 70% or even 90% probability derived from a model was a solid probability to assign to the outcome. But anyways.

For something where you can generate a large number of trials - say, rolling dice or flipping a coin - then the probability does, in fact, predict the distribution of outcomes.

What is the probability saying about a single outcome? Can you answer that question? What is it describing about the future event, to use your definition?
 
I predict a 51% chance the sun will come up tomorrow. It does, was I right?

Before I answer that, let's slaughter several billion harmless pixels scattered across twenty pages figuring out common ground on what we mean by "sun coming up."
 
This is England, so we'll have to settle it by a slight lightening of the drab grey layer of stratus.
 
This is England, so we'll have to settle it by a slight lightening of the drab grey layer of stratus.

It is so tempting to just go along with that, but it would be a total betrayal of what this thread has come to represent, so perhaps we need to agree on calibration of an official grey scale meter.
 
Probability of a single outcome is telling you, under these conditions, what the chance of X event or Y event happening. It's pretty straightforward.

Which is exactly what we have been saying all along. Well, actually, it's completely impossible to tell from a single event whether a 70% or even 90% probability derived from a model was a solid probability to assign to the outcome. But anyways.

For something where you can generate a large number of trials - say, rolling dice or flipping a coin - then the probability does, in fact, predict the distribution of outcomes.

What is the probability saying about a single outcome? Can you answer that question? What is it describing about the future event, to use your definition?

Yeah I mean, I think it's pretty straightforward. In the case of "There is a 70% chance Clinton wins the election.", what is the event? The event is the election. What is the description of the outcome? There is a 70% chance the outcome of the event will be Clinton won, 30% chance that the outcome will be that Trump wins.

How do we verify the prediction?

There are methods like the Brier score:
https://en.wikipedia.org/wiki/Brier_score

academic paper if you're too cool for wiki:
https://web.archive.org/web/2017102...b.noaa.gov/rescue/mwr/078/mwr-078-01-0001.pdf

I would like you to stop using stupid methods for defining things. Like Wikipedia, or website headers, neither of which stands in as a reasonable way to define a word.

When you are talking about whether or not 538's forecasts are predictions, his website is relevant.

It's not my definition, as I keep telling you. When I get home I'll quote you my textbook, if that would help for some odd reason. It's more or less the same thing that Wikipedia says, which is in fact a great source source of info, and in this case has a good definition. Previously when I defined the word for you, you got really upset about it, so I figured you were capable of looking it up or defining it yourself. It's not even clear that you are contesting the previous definitions I have given?
 
It's difficult for a single event, but over a large number of predictions you can see in a pretty straight forward way if the probabilities hold up or not.

You're certainly then testing the model. It's what tetlock does in superforecasters.

But then the prediction is not being tested until x amount of time after the statement. In 3 years, we know whether my statement about a 70% chance of rain was correct, or reasonably correct actually
 
It is so tempting to just go along with that, but it would be a total betrayal of what this thread has come to represent, so perhaps we need to agree on calibration of an official grey scale meter.
Pft, easy. Colorimetry is a thing you know. Or we could measure the amount of light energy incident upon a clearly defined area orthogonal to the solar azimuth. 120Wm-2 looks like a sensible threshold to use? We could use a lens to focus the light on a target of some sort and measure how fast it heats? Or we could use a pyranometer (but they look expensive to be honest).
 
Moderator Action: This is still not the thread for tedious (and entirely tangential) maths discussions. Start a new thread on probability if you must (or not) but stop spamming this thread. Thank you.
 
I'm betting lots of people haven't been paying attention. This entire exchange has been dismally dull.
Yeah, I skipped about 4 pages. In NM a 100% chance of rain means...nothing and there is no need to even think about an umbrella. I haven't seen mine in years.
 
Top Bottom