PLEASE READ!!!

***** ALL users will have to request a password reset BEFORE you will be able to log into the forum. See the thread in the forum issues section for further instructions. If you have issues with this, email us at admin@urbanohio.com. *****

Author Topic: Statistics: What do they even mean?  (Read 1425 times)

0 Members and 1 Guest are viewing this topic.

Offline mu2010

  • Key Tower 947'
  • ****
  • Posts: 946
Re: Statistics: What do they even mean?
« Reply #90 on: January 11, 2018, 04:27:31 PM »
^The hard scientists definitely have a point, there's just no way to do anything about it. This is an age old problem that goes far beyond Nate Silver. The best way to choose a health care law is a controlled test of several health care laws, but we can't do that so we're stuck shouting at each other over whether Obamacare caused premium increases or whether they would have happened anyway.

The second part of your post here is deeply metaphysical. If we could turn back the clock to that morning would it all go the same? If I run a random number generator on my computer and then turn back the clock 10 seconds would I get the same result? I love that stuff and try to read up when I can but the answer is definitely far above my pay grade. I do know that quantum physics supports some element of true randomness in nature. If I flip a coin and get heads I would say the probability was still only 50% heads, and that's where I am coming from. If I win the lottery, it wasn't 100% probability that I was going to just because I did.

On the very last part of your post I'd say the new data coming in was certainly wiping away some bad data and analysis, but definitely also wiping away uncertainty. When they plug in the result of a battleground state that is going to remove uncertainty for sure.
« Last Edit: January 11, 2018, 04:31:11 PM by mu2010 »

Offline X

  • Global Moderator
  • Jeddah Tower 3,281'
  • *****
  • Posts: 9660
    • Western Reserve Meadery
Re: Statistics: What do they even mean?
« Reply #91 on: January 11, 2018, 08:01:47 PM »
The new data wasn't wiping away uncertainty, it was wiping away bad data and/or bad analysis.

The new data was certainly wiping away uncertainty, as it closed out unknowns with each vote counted.

Offline 327

  • Jeddah Tower 3,281'
  • *****
  • Posts: 6641
Re: Statistics: What do they even mean?
« Reply #92 on: January 12, 2018, 10:39:15 AM »
Votes aren't random.  People have reasons for voting the way they do, or for not voting at all. 

Vegas professionals do a better job handicapping football games because they have better information and better models than the average gambler.  They have more data points and they understand how to weigh this data point vs that one. 

If we had enough data, and knew how to apply it, we could accurately predict a coin flip.  They way it faces initially, where you're flipping from, the force and angle of the flip, air currents and aerodynamic properties of that particular coin, etc.  Crunch enough numbers and it's not 50/50 at all.

Uncertainty exists only to the extent that understanding hasn't wiped it away.  Not everyone was surprised when those battleground numbers came in.  It was never Trump who had a slim chance, it was Clinton.  Some people knew that or at least suspected it.  How?  That's worth finding out.  Did 538 suffer from a garbage in garbage out problem?  Is it one they can fix?  Or is the problem that their model needs to look deeper into issues they didn't consider important?

Offline mu2010

  • Key Tower 947'
  • ****
  • Posts: 946
Re: Statistics: What do they even mean?
« Reply #93 on: January 12, 2018, 11:53:23 AM »
Votes aren't random.  People have reasons for voting the way they do, or for not voting at all. 

Certainly true for most voters, and that's why in an election such as LBJ's 1964 landslide, we'd probably get the same result no matter how many times it was run. But in an election which was decided by 70,000 votes spread across three states (0.035% of 2 million registered voters), perhaps some randomness would determine who makes it to the polls and who doesn't. Once again, I'm getting out of stats and into metaphysics, but this is where I'm coming from.

A coin flip is just used as a textbook example of something that's true 50/50 even if it's not. (Though, the idea is to predict before the way it faces initially, where you're flipping from, and the force and angle of the flip are decided. And if you flip a coin 100 times and you aren't doing anything funny, you will get nearly 50/50 results regardless of those above factors.) But you can substitute a random number generator set to spit out zero or one if you prefer.
« Last Edit: January 12, 2018, 11:57:42 AM by mu2010 »

Offline X

  • Global Moderator
  • Jeddah Tower 3,281'
  • *****
  • Posts: 9660
    • Western Reserve Meadery
Re: Statistics: What do they even mean?
« Reply #94 on: January 12, 2018, 01:17:21 PM »
^^If you know everything you don't need probabilities.  You never know everything, so probabilities are useful.  Not because they tell you what WILL happen, but because they tell you the LIKELIHOOD that something or things will happen, based on what you know now.  I suggest that if you are still having problems with this, you should read a statistics textbook or wikipedia it or something, because clearly no amount of explanation is going to get through to you.

Offline 327

  • Jeddah Tower 3,281'
  • *****
  • Posts: 6641
Re: Statistics: What do they even mean?
« Reply #95 on: January 12, 2018, 06:48:21 PM »
A coin flip is just used as a textbook example of something that's true 50/50 even if it's not. (Though, the idea is to predict before the way it faces initially, where you're flipping from, and the force and angle of the flip are decided. And if you flip a coin 100 times and you aren't doing anything funny, you will get nearly 50/50 results regardless of those above factors.) But you can substitute a random number generator set to spit out zero or one if you prefer.

No you can't.  Consider the numbered balls they use for lottery drawings.  They have tolerance limits on lots of factors, for instance the weight of the paint-- too much and it makes a 1 more likely to get chosen than an 8.  Or less likely, depending on the design of the system.  True randomness is impossible to simulate.  Even with computers.  All we can do is approximate it within acceptable standards. 

The result of a coin flip is guaranteed by a bunch of factors that are difficult to measure, making it seem random.  But it isn't.  Neither is an election.  I alluded earlier to garbage in garbage out-- there is little benefit in aggregating the data from ineffective polls.  That's what I suspect happened to 538 this time around.  Can we at least agree that something happened to 538 this time around?

Weather patterns used to seem random too, until we learned enough to kinda sorta predict them.  And that science will continue to develop-- as long as we're willing to reject or modify predictive models that don't work as well as others.  But that means we have to be able to test the models and evaluate them.  No method can be deemed infallible because "probability."  That approach defeats the very purpose of the science.

I suggest that if you are still having problems with this, you should read a statistics textbook or wikipedia it or something, because clearly no amount of explanation is going to get through to you.

No problems here.  My major included a six-course progression of stats, right after the calculus.  Why are you so intent on undercutting me personally?  It's not valid as an argument and it's unbecoming.

Offline StapHanger

  • Jeddah Tower 3,281'
  • *****
  • Posts: 8485
Re: Statistics: What do they even mean?
« Reply #96 on: January 12, 2018, 09:43:03 PM »
^Seems like the concept of "randomness" being discussed is a bit of a distraction. Isn't the issue ex ante uncertainty, and the fact that polling is expensive, so at the state level it's limited, making it pretty significantly unreliable in ways that are impossible to know for sure in real time? Trump may win in every universe with identical circumstances, but even in our own universe, other candidates with his exact polling probably wouldn't. I think that's the more intuitive interpretation of 29% probability.

These kinds of discussions are always so swamped with hindsight. Now that we know the outcome, so many of us will exaggerate the certainty we remember feeling that Trump would win, and talk about all the stupid other people who were overconfident about HRC's chances. And so many of the people who did call it right did so out of pure luck, wishful thinking, or dour paranoia. There will also be people who made the right call from genuine insight, but it's *impossible* for us to distinguish those people from the wischasters and lucky people based on a single prediction. Without a track record or detailed method, it's a fools game to try to try to pick out the real geniuses. This is exactly the same pattern you see in the financial world with so much dumb money plowing into stock pickers who end up getting middling returns after making their name.

As others have said, I don't think we can ever know if 538 was "wrong" or "right" with it's odds in this particular case. It's a dead end argument. The best we can probably do is look at 538's overall track record and make some basic Bayesian deductions about whether Trump's victory means its 29% odds were probably more accurate then the systems that gave Trump much lower odds.