2018 Midterm Elections

[carpedm9587]

nope. even the pollsters were being interviewed after the election and saying they had no idea what went wrong.

Here is 538's

Ironically, 538 was one of the sites that was predicting Trump had a better chance than a lot of people suggested (if you listen to their pre-election podcast). The polling error that keeps being focused on was in a very small handful of states (4, IIRC), and has been since attributed to a few possible causes:

1) Last minute changes (e.g., the Comey letter)
2) The "shy" Republican phenomenon (as it is being called)
3) Failure to account for representational skew (especially related to women and education)

These were the only states that had votes that fell outside the polling margin of error. The rest of the states and the national polls were well within the polling margin of error. Pretty much all of the rest of the "bad polling" people keep referring to was actually misuse of the polling results, as has been previously discussed. Your link is a perfect example. The "probability of a win" is not the results of polling - polling suggests how the vote is expected to distribute. Based on that, and the values for various states, a "probability of a win" is made. A 71/29 distribution is not "wrong" if the 29 occurs, anymore than the odds of rolling a six on a die (about 16.6%) is "wrong" if a six is actually rolled.

[Sparko]

Ironically, 538 was one of the sites that was predicting Trump had a better chance than a lot of people suggested (if you listen to their pre-election podcast). The polling error that keeps being focused on was in a very small handful of states (4, IIRC), and has been since attributed to a few possible causes:

1) Last minute changes (e.g., the Comey letter)
2) The "shy" Republican phenomenon (as it is being called)
3) Failure to account for representational skew (especially related to women and education)

These were the only states that had votes that fell outside the polling margin of error. The rest of the states and the national polls were well within the polling margin of error. Pretty much all of the rest of the "bad polling" people keep referring to was actually misuse of the polling results, as has been previously discussed. Your link is a perfect example. The "probability of a win" is not the results of polling - polling suggests how the vote is expected to distribute. Based on that, and the values for various states, a "probability of a win" is made. A 71/29 distribution is not "wrong" if the 29 occurs, anymore than the odds of rolling a six on a die (about 16.6%) is "wrong" if a six is actually rolled.

The end result is that they all got it wrong. You can't go by what the polls tell you. I am just going to wait and see. You should do the same instead of getting all worked up about this "blue wave" - it will either happen or it won't. And it will probably be decided by some event that happens a week before the election as both parties start airing the other side's dirty laundry to try to effect the outcome.

[carpedm9587]

The end result is that they all got it wrong. You can't go by what the polls tell you. I am just going to wait and see. You should do the same instead of getting all worked up about this "blue wave" - it will either happen or it won't. And it will probably be decided by some event that happens a week before the election as both parties start airing the other side's dirty laundry to try to effect the outcome.

And what I am saying is that they didn't "get it wrong," except for the talking heads and polling in four states, which were slightly outside their margins of error. The rest were well within the margins. As for being worked up, playing with stats and projections is hardly "worked up." I like playing with numbers. And these are giving me a sense of hope for the fall, that we might actually see a bit of balance restored at the federal and state levels.

Naturally - much can happen between now and then.

[Sparko]

And what I am saying is that they didn't "get it wrong," except for the talking heads and polling in four states, which were slightly outside their margins of error. The rest were well within the margins. As for being worked up, playing with stats and projections is hardly "worked up." I like playing with numbers. And these are giving me a sense of hope for the fall, that we might actually see a bit of balance restored at the federal and state levels.

Naturally - much can happen between now and then.

No. You can look at the actual polls just like I posted previously. No talking heads involved. Just like the polls you are proposing now about the 2018 election. You can go back and look at the polls and charts and see that they were predicting a win by Clinton. They were wrong. Some more wrong than others. You can't try to rewrite history now Carp. You don't remember the big upset and all of the news people being shocked at how wrong the polls were?

"within margins" is just an excuse for being wrong.

[carpedm9587]

No. You can look at the actual polls just like I posted previously. No talking heads involved. Just like the polls you are proposing now about the 2018 election. You can go back and look at the polls and charts and see that they were predicting a win by Clinton. They were wrong. Some more wrong than others. You can't try to rewrite history now Carp. You don't remember the big upset and all of the news people being shocked at how wrong the polls were?

"within margins" is just an excuse for being wrong.

Sparko, I think you perhaps do not understand polling or their relationship to statements of probability. A poll (in this case) provides a voting distribution projection. So, for example, it says "our poll suggests X will get 44.3% of the vote and Y will get 46.2% of the vote. All polls have a margin of error because they are based on a sample. Usually, the smaller the sample, the larger the margin (law of large numbers at work). Polls with 1500 respondents are considered ideal. You will typically find these with a margin of error of 2-3 points.

When someone says, "X has a Y% probability of winning," they are using the polling data to project a likelihood of an outcome. Although that likelihood is related to the distance between the voting projections, it is not equivalent to them. So if the poll projects "44.3% of the vote goes to X with margin of error of +/- 3 points," then the poll was accurate if X gets between 41.3% and 47.3%. If X actually gets 39% or 49% of the vote, then the poll missed the mark - the results are outside the margin of error. With only 4 state-level exceptions that I know of, the polls nailed the 2016 election.

Projections of "probability of win" are not polls - they are projections based on polls. They cannot be wrong unless they give a candidate a 0% chance and the candidate actually wins.

[Sparko]

Sparko, I think you perhaps do not understand polling or their relationship to statements of probability. A poll (in this case) provides a voting distribution projection. So, for example, it says "our poll suggests X will get 44.3% of the vote and Y will get 46.2% of the vote. All polls have a margin of error because they are based on a sample. Usually, the smaller the sample, the larger the margin (law of large numbers at work). Polls with 1500 respondents are considered ideal. You will typically find these with a margin of error of 2-3 points.

When someone says, "X has a Y% probability of winning," they are using the polling data to project a likelihood of an outcome. Although that likelihood is related to the distance between the voting projections, it is not equivalent to them. So if the poll projects "44.3% of the vote goes to X with margin of error of +/- 3 points," then the poll was accurate if X gets between 41.3% and 47.3%. If X actually gets 39% or 49% of the vote, then the poll missed the mark - the results are outside the margin of error. With only 4 state-level exceptions that I know of, the polls nailed the 2016 election.

Projections of "probability of win" are not polls - they are projections based on polls. They cannot be wrong unless they give a candidate a 0% chance and the candidate actually wins.

End result? They were still wrong. No matter how much you try to excuse them, they were wrong. ALL of them. They predicted Clinton would win and all of the liberal media went along with it. And they were wrong. It doesn't matter if they were wrong by 1% or 50% - wrong is still wrong. You would think that out of dozens of polls statistically some would have Trump winning if it were that close of an error margin, but nope. They all had Clinton winning.

Close only counts in horseshoes and hand grenades.

[carpedm9587]

End result? They were still wrong. No matter how much you try to excuse them, they were wrong. ALL of them. They predicted Clinton would win and all of the liberal media went along with it. And they were wrong. It doesn't matter if they were wrong by 1% or 50% - wrong is still wrong. You would think that out of dozens of polls statistically some would have Trump winning if it were that close of an error margin, but nope. They all had Clinton winning.

Close only counts in horseshoes and hand grenades.

Four states (that I know of) had polls that were wrong, Sparko. The other 46 and the national scale polls were all accurate (within their margin of error). That's the point I was trying to make.

As for the projected winners, as I said before, if a projection is that X has a 75% chance of winning and Y has a 25% chance of winning, that's the same as saying "you have a 75% chance of tossing two coins and getting at least one head, and a 25% chance of getting both tails." If you toss the coins and get two tails, that doesn't mean the probabilities were wrong. It means the 25% chance is the one that happened to hit. If you toss the coins repeatedly, you'll expect (over time) to see a 3/1 distribution based on the probabilities. Trump probability of winning was anywhere from 5% (NYT) to about 30% (RCP/538), and he won. That doesn't make the projections wrong. It just means the 5-30% is what hit, not the 70-95%.

[Sparko]

Four states (that I know of) had polls that were wrong, Sparko. The other 46 and the national scale polls were all accurate (within their margin of error). That's the point I was trying to make.

As for the projected winners, as I said before, if a projection is that X has a 75% chance of winning and Y has a 25% chance of winning, that's the same as saying "you have a 75% chance of tossing two coins and getting at least one head, and a 25% chance of getting both tails." If you toss the coins and get two tails, that doesn't mean the probabilities were wrong. It means the 25% chance is the one that happened to hit. If you toss the coins repeatedly, you'll expect (over time) to see a 3/1 distribution based on the probabilities. Trump probability of winning was anywhere from 5% (NYT) to about 30% (RCP/538), and he won. That doesn't make the projections wrong. It just means the 5-30% is what hit, not the 70-95%.

again, "accurate within the margin of error" is still wrong when they predicted Clinton as the winner. And when nearly ALL of them predict the same winner then it is more than a statistical variance. If it were a random margin of error you would have seen an even number of polls predicting Trump would win. If 100 people flip a coin and all report heads, you might want to check those results. While statistically possible, it is not statistically likely that they nearly all got the same results (I think a couple actually did predict Trump)


Edited to add:

Here is a list of all of the polls:
https://www.realclearpolitics.com/ep...tein-5952.html

Out of 186 polls, 13 predicted Trump would win.

[carpedm9587]

again, "accurate within the margin of error" is still wrong when they predicted Clinton as the winner. And when nearly ALL of them predict the same winner then it is more than a statistical variance. If it were a random margin of error you would have seen an even number of polls predicting Trump would win. If 100 people flip a coin and all report heads, you might want to check those results. While statistically possible, it is not statistically likely that they nearly all got the same results (I think a couple actually did predict Trump)

I highlighted where you are making the error. You are conflating two things: the polls (which have a margin of error and were broadly correct, except in four states) and the probability of a win (which is based on the polls but not equivalent to them, and cannot be incorrect unless they give someone a 0% chance of winning and they win). Beyond that, simply repeating myself is not going to get us anywhere.

[Sparko]

I highlighted where you are making the error. You are conflating two things: the polls (which have a margin of error and were broadly correct, except in four states) and the probability of a win (which is based on the polls but not equivalent to them, and cannot be incorrect unless they give someone a 0% chance of winning and they win). Beyond that, simply repeating myself is not going to get us anywhere.

The polls predict a winner based on their research. If you check the list of polls I posted in my last post, you will see 186 of them with only 13 predicting Trump as the winner. When they predicted Clinton as the winner they were wrong. That is not "broadly correct" - it is wrong. You can repeat yourself all you want, but they were still wrong.

giphy.gif

[carpedm9587]

The polls predict a winner based on their research. If you check the list of polls I posted in my last post, you will see 186 of them with only 13 predicting Trump as the winner. When they predicted Clinton as the winner they were wrong. That is not "broadly correct" - it is wrong. You can repeat yourself all you want, but they were still wrong.

[ATTACH=CONFIG]27531[/ATTACH]

I think you could benefit from a course in probability and statistics. You don't seem to understand that the 29.1% for Trump is a 29.1% probability of a win.

[Sparko]

I think you could benefit from a course in probability and statistics. You don't seem to understand that the 29.1% for Trump is a 29.1% probability of a win.

I am mostly pushing your buttons at this point. But the polls were wrong. If I giving odds on a horse to win 2/1 and he loses then I was wrong. Even though I said there was a 50% chance that he could lose!

[carpedm9587]

I am mostly pushing your buttons at this point. But the polls were wrong. If I giving odds on a horse to win 2/1 and he loses then I was wrong. Even though I said there was a 50% chance that he could lose!

If a horse has 2/1 odds, it means the payout for the horse is twice what was bet. That ratio is related to the belief in the likelihood of a win, but it is NOT the likelihood of a win. You keep conflating terms, suggesting you don't actually understand them. The only errors that were made in 2016, AFAICT, is the four states whose polls were outside the margin of error. The other error was the talking heads who, like you, didn't understand the numbers and overhyped the "Clinton win."

As for pushing buttons...only my wife can do that...

[Sparko]

If a horse has 2/1 odds, it means the payout for the horse is twice what was bet. That ratio is related to the belief in the likelihood of a win, but it is NOT the likelihood of a win. You keep conflating terms, suggesting you don't actually understand them. The only errors that were made in 2016, AFAICT, is the four states whose polls were outside the margin of error. The other error was the talking heads who, like you, didn't understand the numbers and overhyped the "Clinton win."

As for pushing buttons...only my wife can do that...

Well horse odds are determined by betting, so it means that like in a poll, more people bet on the horse to win than lose, so the track adjusts the payout to accommodate it. You can think of it as a poll of the betters. If a lot of people bet on the horse to win, it's payout will be less. Crowdsourcing in a way. The track is betting the people know what they are doing as a group. Most times they are right, but sometimes they are wrong. If the odds are 100/1 that would indicate that the track (based on betting) thinks that horse will lose. If it wins the track was wrong and it has to pay out.

[carpedm9587]

Well horse odds are determined by betting, so it means that like in a poll, more people bet on the horse to win than lose, so the track adjusts the payout to accommodate it. You can think of it as a poll of the betters. If a lot of people bet on the horse to win, it's payout will be less. Crowdsourcing in a way. The track is betting the people know what they are doing as a group. Most times they are right, but sometimes they are wrong. If the odds are 100/1 that would indicate that the track (based on betting) thinks that horse will lose. If it wins the track was wrong and it has to pay out.

Yes - the adjustment is based on the betting - not on the capabilities of the horse, nor the probability of winning. It has more to do with the house controlling their win/loss ratio. Ergo, the link to the election process is essentially nonexistent.

Look, Sparko, I know it's a major meme that the "polls were wrong." A lot of people buy it and it's an easy meme to put out there because most people do not understand polling and probability concepts. It just won't convince someone who actually does.

Last word to you. Opportunity to practice my carpescape!