Other Resources The Political Joe » Polls Rss Feed  
Moderators: k9car363, alicefoeller Reply
 
 
of 14
 
 
2016-07-26 12:17 PM

User image

Champion
10154
500050001002525
Alabama
Subject: Polls

Read an article that said Trump got a 6 point bounce from the convention last week.  But I am skeptical.  If you go from 3% down to 3% up in a poll where the margin of error is +/- 3% how can you say with any confidence that your numbers mean anything?

 

I have no idea how they calculate the margin of error.  I guess there are some people who just flat out lie....they say they are 'likely voters' and they say they are registered to vote.....when in truth they've never voted in their life.  Then I guess you have some people who know who they are voting for but refuse to answer.  And then there are people like me who just hang-up when politicians call during dinner time.  I also have a hard time getting my head around 1000 people being representative of 50 million. 

Obviously not an exact science when one polls show one candidate up and the other shows them down and both are credible polls.

I know some people put stock in the poll of polls or average between multiple polls....not sure that you can average bad data and end up with better data.....unless it's like Fermi's piano tuner estimate.

 

http://www.grc.nasa.gov/WWW/K-12/Numbers/Math/Mathematical_Thinking/fermis_piano_tuner.htm



Edited by Rogillio 2016-07-26 12:31 PM


2016-07-27 9:32 AM
in reply to: Rogillio

User image

Champion
6993
50001000500100100100100252525
Chicago, Illinois
Subject: RE: Polls
Originally posted by Rogillio

Read an article that said Trump got a 6 point bounce from the convention last week.  But I am skeptical.  If you go from 3% down to 3% up in a poll where the margin of error is +/- 3% how can you say with any confidence that your numbers mean anything?

 

I have no idea how they calculate the margin of error.  I guess there are some people who just flat out lie....they say they are 'likely voters' and they say they are registered to vote.....when in truth they've never voted in their life.  Then I guess you have some people who know who they are voting for but refuse to answer.  And then there are people like me who just hang-up when politicians call during dinner time.  I also have a hard time getting my head around 1000 people being representative of 50 million. 

Obviously not an exact science when one polls show one candidate up and the other shows them down and both are credible polls.

I know some people put stock in the poll of polls or average between multiple polls....not sure that you can average bad data and end up with better data.....unless it's like Fermi's piano tuner estimate.

 

http://www.grc.nasa.gov/WWW/K-12/Numbers/Math/Mathematical_Thinking/fermis_piano_tuner.htm




well people get a bounce. 2 weeks ago I would voted Clinton. Today I been thinking I would vote Trump. Tomorrow not sure probably go back to Clinton. changes of no change even with 3% margin of error is slim but you are right that it might not be 6% swing but 5%. Been awhile since I studied these things but I do suspect that when 3% margin of error is very slim.
2016-07-27 10:19 AM
in reply to: chirunner134

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

LA Times/USC poll have Trump up by 7. 

Guess we'll see next week how much Hillary bounced.  Maybe Bill bounced her?  :-)

2016-07-28 10:30 AM
in reply to: Rogillio

User image

Champion
6993
50001000500100100100100252525
Chicago, Illinois
Subject: RE: Polls
I just took a poll. IT was probably a bad one because apart of a congressmen trying to raise money but after the poll I thought I should have lied about my answers to make it more dramatic. Then I thought well problems with polls is that people might try to push an untrue narrative. I am glad I told the truth. With do you support trump I can see why people might be scared to say they do. There is a lot of if you support him your a racist, sexist, and misogynist going around so of course I can see why a person might lie to a stranger. People are so quick to judge sometimes.
2016-07-28 10:46 AM
in reply to: chirunner134

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

Originally posted by chirunner134 I just took a poll. IT was probably a bad one because apart of a congressmen trying to raise money but after the poll I thought I should have lied about my answers to make it more dramatic. Then I thought well problems with polls is that people might try to push an untrue narrative. I am glad I told the truth. With do you support trump I can see why people might be scared to say they do. There is a lot of if you support him your a racist, sexist, and misogynist going around so of course I can see why a person might lie to a stranger. People are so quick to judge sometimes.

 

I totally agree with that this.  I think people tend to be non-confrontational and people-pleasers and will tell the pollsters what they think they want to hear.  This is why I don't put much stock in polls.  It's hard to do a scientific poll without any bias.  How one answers has a lot to do with how the question is posed....and who is asking the question.  If a man asking a woman?  Is a woman asking a woman?  Is a woman asking a man?  If a woman is asking a woman if who she supports a woman might be more inclined to say Hillary.  Likewise, a man asking a man he might be more inclined to say is he voting for the man.

Maybe pollsters take all this into consideration and that is how they derive their margins of error? 

 

 

2016-07-28 1:35 PM
in reply to: Rogillio

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

I've often been a sceptic of the polls over the years but again and again they've proved me wrong.  There's a lot of science that goes into it and the big name ones are usually pretty close.  There's always outliers that are pretty obvious with push polls or over sampling with no corrections, but those are the minority.

if you take a step back it makes sense that trump would get a big bump because people don't know a lot about him other than what the media tries to portray.  He did a good job of stating his case at the convention and independents and hesitant republicans likely responded.  

They will all eb and flow over the next few months with wider swings and then smaller swings.  There's also big unknowns that can have big opinion swings such as the stock market crash back in 2008. McCain was consistently ahead of Obama until the market went boom and everybody wanted change.



2016-07-28 3:02 PM
in reply to: tuwood

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

Originally posted by tuwood

I've often been a sceptic of the polls over the years but again and again they've proved me wrong.  There's a lot of science that goes into it and the big name ones are usually pretty close.  There's always outliers that are pretty obvious with push polls or over sampling with no corrections, but those are the minority.

if you take a step back it makes sense that trump would get a big bump because people don't know a lot about him other than what the media tries to portray.  He did a good job of stating his case at the convention and independents and hesitant republicans likely responded.  

They will all eb and flow over the next few months with wider swings and then smaller swings.  There's also big unknowns that can have big opinion swings such as the stock market crash back in 2008. McCain was consistently ahead of Obama until the market went boom and everybody wanted change.

 

I remember how shocked many pollsters were in the 2012 election.

Good article on polling for that election.

 

http://fivethirtyeight.blogs.nytimes.com/2012/11/10/which-polls-fared-best-and-worst-in-the-2012-presidential-race/?_r=0

 

Pretty significant republican bias that year.

2016-07-30 6:09 PM
in reply to: Rogillio

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
Clinton gets 10 point bump...


"A new poll taken after the Democratic convention revealed that Hillary Clinton had erased any gains that Trump made after the Republican convention, with a 10 point convention bounce and a 15 point lead."
2016-07-30 6:29 PM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls
Originally posted by ejshowersClinton gets 10 point bump..."A new poll taken after the Democratic convention revealed that Hillary Clinton had erased any gains that Trump made after the Republican convention, with a 10 point convention bounce and a 15 point lead."
And if you believe that one I have a big bridge in San Francisco to sell you.
2016-07-30 6:38 PM
in reply to: tuwood

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
Originally posted by tuwood

Originally posted by ejshowersClinton gets 10 point bump..."A new poll taken after the Democratic convention revealed that Hillary Clinton had erased any gains that Trump made after the Republican convention, with a 10 point convention bounce and a 15 point lead."
And if you believe that one I have a big bridge in San Francisco to sell you.


Only one poll, but the data is the data.
2016-07-30 7:41 PM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls
Originally posted by ejshowers
Originally posted by tuwood
Originally posted by ejshowersClinton gets 10 point bump..."A new poll taken after the Democratic convention revealed that Hillary Clinton had erased any gains that Trump made after the Republican convention, with a 10 point convention bounce and a 15 point lead."
And if you believe that one I have a big bridge in San Francisco to sell you.
Only one poll, but the data is the data.
I'll be curious to watch the averages over the next week or two to see where the votes draw from. For example trumps bump came almost all from undecided and virtually none from Clinton. Curious to see if Clinton draws from undecided as well or from trump.


2016-08-01 7:29 AM
in reply to: 0

User image

Elite
4547
2000200050025
Subject: RE: Polls
CBS News poll gave Trump a 2-point bump from the RNC convention, and Clinton a 4-point bump from the DNC convention, to give Clinton an overall 7-point lead on August 1st. (reported by the Washington Times)

The Morning Consult poll had Clinton gain big with men, and has her now leading Trump in the male vote. (it also had her pick up 4 points with independent voters)
Hillary's favorability went from 37 to 43%. Trump's fell from 42 to 39%.





Edited by ChineseDemocracy 2016-08-01 7:31 AM
2016-08-01 9:48 AM
in reply to: ChineseDemocracy

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

Yet Trump lead the LA Times daily tracking poll in all but one day last month.
Comparatively (in this poll anyways) he got a much larger bump than Hillary.

http://www.latimes.com/politics/

 

2016-08-01 10:41 AM
in reply to: 0

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

Originally posted by tuwood

Yet Trump lead the LA Times daily tracking poll in all but one day last month.
Comparatively (in this poll anyways) he got a much larger bump than Hillary.

http://www.latimes.com/politics/

 

 

 

I saw that too and am left scratching my head.  How can two credible polls come up with dramatically different result that are not even within the polls margins of error?!



Edited by Rogillio 2016-08-01 10:42 AM
2016-08-01 10:41 AM
in reply to: 0

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

dp



Edited by Rogillio 2016-08-01 10:42 AM
2016-08-01 9:19 PM
in reply to: 0

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
This is what I am paying attention to - the blend of all three models:

http://projects.fivethirtyeight.com/2016-election-forecast/#now


Edited by ejshowers 2016-08-01 9:22 PM


2016-08-02 7:50 AM
in reply to: ejshowers

User image

Champion
10154
500050001002525
Alabama
Subject: RE: Polls

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

 

I like this!  Just bookmarked it.  Lots of good data....some bad data.  :-)  The very top "chance of winning" is meaningless.  The chance of winning is mostly of function of what happens between now and Nov....how many more times Trump steps on his d......how many more Wikileaks emails come out damning Clinton.....how they do in the debates....how well their ground games work....more probably most important, who turns out to vote.

 

 

2016-08-02 8:46 AM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

2016-08-02 9:30 AM
in reply to: 0

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.




You didn't look very hard. The link to the model explanation is near the bottom of the page:

http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh...

For Florida they say:

"We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."



Edited by ejshowers 2016-08-02 9:32 AM
2016-08-02 9:45 AM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

You didn't look very hard. The link to the model explanation is near the bottom of the page: http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh... For Florida they say: "We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."

You missed the bolded part.  So like I said, we have no idea whatsoever how they're coming up with their data.

Lets look at their statement and compare it to actual polls from RCP in Florida.   There have been 5 polls that RCP recognizes as legit conducted in Florida since 6/27.  in those it was Tie, Trump +3, Clinton +7, Trump +5, Trump +.3.  Now you apply the logic mentioned above and it is obvious they're adjusting "based on other factors" to give Florida to Clinton.  

Another datapoint that's a little interesting is that in poll after poll Trump is crushing Hillary in independents.  Meaning the only reason Hillary is ahead is because of lower support within the Republican party.  That may be true today, but it's a little hard to hang your hat on all the undecided Republicans who don't support Trump now voting for Hillary to give her the win. 
Public Policy Polling (Hillary is up big overall, but Trump leads independents 46%-41%)

All I can say is that polling is really weird this time around.  I genuinely try to stay out of the conspiracy camp of the media and pollsters trying to throw public opinion one way or the other via polls, but it's really hard to.  lol

 

2016-08-02 12:48 PM
in reply to: tuwood

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

You didn't look very hard. The link to the model explanation is near the bottom of the page: http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh... For Florida they say: "We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."

You missed the bolded part.  So like I said, we have no idea whatsoever how they're coming up with their data.

Lets look at their statement and compare it to actual polls from RCP in Florida.   There have been 5 polls that RCP recognizes as legit conducted in Florida since 6/27.  in those it was Tie, Trump +3, Clinton +7, Trump +5, Trump +.3.  Now you apply the logic mentioned above and it is obvious they're adjusting "based on other factors" to give Florida to Clinton.  

Another datapoint that's a little interesting is that in poll after poll Trump is crushing Hillary in independents.  Meaning the only reason Hillary is ahead is because of lower support within the Republican party.  That may be true today, but it's a little hard to hang your hat on all the undecided Republicans who don't support Trump now voting for Hillary to give her the win. 
Public Policy Polling (Hillary is up big overall, but Trump leads independents 46%-41%)

All I can say is that polling is really weird this time around.  I genuinely try to stay out of the conspiracy camp of the media and pollsters trying to throw public opinion one way or the other via polls, but it's really hard to.  lol

 




Did you even look at the user guide??? Their approach and assumptions are laid out in quite a bit of detail. They don't reveal the actual code of the model of course. That would be like Coke displaying its recipe for all to see on the internet! Also, as I posted, they say they used 26 Florida polls and WEIGHT them, so taking the latest 5 from RCP won't match except by chance.

"There are five adjustments, listed here in the order in which the model applies them. (The trend line and house effects adjustments are generally the most important ones.)
?Likely voter adjustment
?Convention bounce adjustment (in only the polls-plus model)
?Omitted third-party candidate adjustment
?Trend line adjustment
?House effects adjustment"

The guide then goes into detail for each adjustment.

538 nailed 2008 and 2012, and their approach seems very rigorous, so until I see something better or their methodology fails, I will respect their analyses.


2016-08-02 12:55 PM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

You didn't look very hard. The link to the model explanation is near the bottom of the page: http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh... For Florida they say: "We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."

You missed the bolded part.  So like I said, we have no idea whatsoever how they're coming up with their data.

Lets look at their statement and compare it to actual polls from RCP in Florida.   There have been 5 polls that RCP recognizes as legit conducted in Florida since 6/27.  in those it was Tie, Trump +3, Clinton +7, Trump +5, Trump +.3.  Now you apply the logic mentioned above and it is obvious they're adjusting "based on other factors" to give Florida to Clinton.  

Another datapoint that's a little interesting is that in poll after poll Trump is crushing Hillary in independents.  Meaning the only reason Hillary is ahead is because of lower support within the Republican party.  That may be true today, but it's a little hard to hang your hat on all the undecided Republicans who don't support Trump now voting for Hillary to give her the win. 
Public Policy Polling (Hillary is up big overall, but Trump leads independents 46%-41%)

All I can say is that polling is really weird this time around.  I genuinely try to stay out of the conspiracy camp of the media and pollsters trying to throw public opinion one way or the other via polls, but it's really hard to.  lol

 

Did you even look at the user guide??? Their approach and assumptions are laid out in quite a bit of detail. They don't reveal the actual code of the model of course. That would be like Coke displaying its recipe for all to see on the internet! Also, as I posted, they say they used 26 Florida polls and WEIGHT them, so taking the latest 5 from RCP won't match except by chance. "There are five adjustments, listed here in the order in which the model applies them. (The trend line and house effects adjustments are generally the most important ones.) ?Likely voter adjustment ?Convention bounce adjustment (in only the polls-plus model) ?Omitted third-party candidate adjustment ?Trend line adjustment ?House effects adjustment" The guide then goes into detail for each adjustment. 538 nailed 2008 and 2012, and their approach seems very rigorous, so until I see something better or their methodology fails, I will respect their analyses.

Wonder how they faired in the Republican primary.  I seem to recall that not a single prediction site had Trump winning that one either.   
It's just polls and they're all over the place so you and I really don't have much to stand on either way.  It's continuing to be an interesting election.
Obama is certainly scared because he's out there bashing on Trump every chance he can.  If Trump was as far behind as everyone believes then Obama would sit back and smile I suspect.  His attacks are quite unprecedented for a sitting President. 

2016-08-02 2:15 PM
in reply to: tuwood

User image

Master
2802
2000500100100100
Minnetonka, Minnesota
Bronze member
Subject: RE: Polls
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

You didn't look very hard. The link to the model explanation is near the bottom of the page: http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh... For Florida they say: "We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."

You missed the bolded part.  So like I said, we have no idea whatsoever how they're coming up with their data.

Lets look at their statement and compare it to actual polls from RCP in Florida.   There have been 5 polls that RCP recognizes as legit conducted in Florida since 6/27.  in those it was Tie, Trump +3, Clinton +7, Trump +5, Trump +.3.  Now you apply the logic mentioned above and it is obvious they're adjusting "based on other factors" to give Florida to Clinton.  

Another datapoint that's a little interesting is that in poll after poll Trump is crushing Hillary in independents.  Meaning the only reason Hillary is ahead is because of lower support within the Republican party.  That may be true today, but it's a little hard to hang your hat on all the undecided Republicans who don't support Trump now voting for Hillary to give her the win. 
Public Policy Polling (Hillary is up big overall, but Trump leads independents 46%-41%)

All I can say is that polling is really weird this time around.  I genuinely try to stay out of the conspiracy camp of the media and pollsters trying to throw public opinion one way or the other via polls, but it's really hard to.  lol

 

Did you even look at the user guide??? Their approach and assumptions are laid out in quite a bit of detail. They don't reveal the actual code of the model of course. That would be like Coke displaying its recipe for all to see on the internet! Also, as I posted, they say they used 26 Florida polls and WEIGHT them, so taking the latest 5 from RCP won't match except by chance. "There are five adjustments, listed here in the order in which the model applies them. (The trend line and house effects adjustments are generally the most important ones.) ?Likely voter adjustment ?Convention bounce adjustment (in only the polls-plus model) ?Omitted third-party candidate adjustment ?Trend line adjustment ?House effects adjustment" The guide then goes into detail for each adjustment. 538 nailed 2008 and 2012, and their approach seems very rigorous, so until I see something better or their methodology fails, I will respect their analyses.

Wonder how they faired in the Republican primary.  I seem to recall that not a single prediction site had Trump winning that one either.   
It's just polls and they're all over the place so you and I really don't have much to stand on either way.  It's continuing to be an interesting election.
Obama is certainly scared because he's out there bashing on Trump every chance he can.  If Trump was as far behind as everyone believes then Obama would sit back and smile I suspect.  His attacks are quite unprecedented for a sitting President. 




Ask and yea shall receive. From May 18th, 2016:

"The FiveThirtyEight “polls-only” model has correctly predicted the winner in 52 of 57 (91 percent) primaries and caucuses so far in 2016, and our related “polls-plus” model has gone 51-for-57 (89 percent). Furthermore, the forecasts have been well-calibrated, meaning that upsets have occurred about as often as they’re supposed to but not more often."

Primaries are MUCH harder to predict given their ever changing rules, the lack of and spotty nature of the data, and how one day's voting can affect another, among other issues.

LONG analysis here. Enjoy.

http://fivethirtyeight.com/features/how-i-acted-like-a-pundit-and-s...


2016-08-02 2:23 PM
in reply to: ejshowers

User image

Pro
9391
500020002000100100100252525
Omaha, NE
Subject: RE: Polls

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers
Originally posted by tuwood

Originally posted by ejshowers This is what I am paying attention to - the blend of all three models: http://projects.fivethirtyeight.com/2016-election-forecast/#now

The problem I have with sights like this is I have no idea how they're coming up with their information.  For example they've got Florida solidly in Clinton's camp, but if you look at the RCP polls Trump has been ahead in Florida for a long time.  I think Clinton is in the lead in one of their last 4 polls.  So how in the bleep can they call Florida for Hillary with that data?  I didn't go through all their states, but it's certainly hard to trust them in any way.

You didn't look very hard. The link to the model explanation is near the bottom of the page: http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeigh... For Florida they say: "We’ve collected 26 polls in Florida so far. The model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm. The model then adjusts each poll based on other factors."

You missed the bolded part.  So like I said, we have no idea whatsoever how they're coming up with their data.

Lets look at their statement and compare it to actual polls from RCP in Florida.   There have been 5 polls that RCP recognizes as legit conducted in Florida since 6/27.  in those it was Tie, Trump +3, Clinton +7, Trump +5, Trump +.3.  Now you apply the logic mentioned above and it is obvious they're adjusting "based on other factors" to give Florida to Clinton.  

Another datapoint that's a little interesting is that in poll after poll Trump is crushing Hillary in independents.  Meaning the only reason Hillary is ahead is because of lower support within the Republican party.  That may be true today, but it's a little hard to hang your hat on all the undecided Republicans who don't support Trump now voting for Hillary to give her the win. 
Public Policy Polling (Hillary is up big overall, but Trump leads independents 46%-41%)

All I can say is that polling is really weird this time around.  I genuinely try to stay out of the conspiracy camp of the media and pollsters trying to throw public opinion one way or the other via polls, but it's really hard to.  lol

 

Did you even look at the user guide??? Their approach and assumptions are laid out in quite a bit of detail. They don't reveal the actual code of the model of course. That would be like Coke displaying its recipe for all to see on the internet! Also, as I posted, they say they used 26 Florida polls and WEIGHT them, so taking the latest 5 from RCP won't match except by chance. "There are five adjustments, listed here in the order in which the model applies them. (The trend line and house effects adjustments are generally the most important ones.) ?Likely voter adjustment ?Convention bounce adjustment (in only the polls-plus model) ?Omitted third-party candidate adjustment ?Trend line adjustment ?House effects adjustment" The guide then goes into detail for each adjustment. 538 nailed 2008 and 2012, and their approach seems very rigorous, so until I see something better or their methodology fails, I will respect their analyses.

Wonder how they faired in the Republican primary.  I seem to recall that not a single prediction site had Trump winning that one either.   
It's just polls and they're all over the place so you and I really don't have much to stand on either way.  It's continuing to be an interesting election.
Obama is certainly scared because he's out there bashing on Trump every chance he can.  If Trump was as far behind as everyone believes then Obama would sit back and smile I suspect.  His attacks are quite unprecedented for a sitting President. 

Ask and yea shall receive. From May 18th, 2016: "The FiveThirtyEight “polls-only” model has correctly predicted the winner in 52 of 57 (91 percent) primaries and caucuses so far in 2016, and our related “polls-plus” model has gone 51-for-57 (89 percent). Furthermore, the forecasts have been well-calibrated, meaning that upsets have occurred about as often as they’re supposed to but not more often." Primaries are MUCH harder to predict given their ever changing rules, the lack of and spotty nature of the data, and how one day's voting can affect another, among other issues. LONG analysis here. Enjoy. http://fivethirtyeight.com/features/how-i-acted-like-a-pundit-and-s...

I'll have to read that later, but it does look interesting to me.  Glancing at the headlines it appears they had some challenges with Trump predictions in the primaries, but are open about them and trying to identify why.   I wonder how much of the pre-Trump modeling remains in their current general election system.

2016-08-04 10:05 PM
in reply to: tuwood

Master
5557
50005002525
, California
Subject: RE: Polls

This page had a cool graph:

http://www.nytimes.com/interactive/2016/upshot/presidential-polls-forecast.html

 

Yeah yeah I know some of you may not like the NY Times liberal slant.  Don't read it - just scroll to the bottom.  The last graph shows the 10 most competitive states and different ways they can add up electoral votes to win.

New Thread
Other Resources The Political Joe » Polls Rss Feed  
 
 
of 14