Daily Presidential Tracking Poll

[The Nature Boy 7,659]

 

TRUMP MATCHES OBAMA END OF FIRST YEAR
46% APPROVE, 53% DISAPPROVE

 

 

http://www.rasmussenreports.com/public_content/politics/trump_administration/prez_track_dec28

[bcking 3,157]

I'd love to polls to be more transparent in their methodology about how they weighted samples. They state that they do, but I'd love to see the numbers. I'm not saying Rasmussen is bad, they may be the very best. It would go a long way to help us compare different polls that end up with significantly different results. The transparency would also help to keep the groups doing the polls honest because they have to account for everything and their work can be checked independently. 

 

Even if it was in a "supplemental data" section that most people wouldn't click on. I'd still love to see the raw numbers, and then the weighting that they used. While generally weighting is a good idea, it can skew results if certain groups are weighted "too much".

[The Nature Boy 7,659]

31 minutes ago, bcking said:

I'd love to polls to be more transparent in their methodology about how they weighted samples. They state that they do, but I'd love to see the numbers. I'm not saying Rasmussen is bad, they may be the very best. It would go a long way to help us compare different polls that end up with significantly different results. The transparency would also help to keep the groups doing the polls honest because they have to account for everything and their work can be checked independently. 

 

Even if it was in a "supplemental data" section that most people wouldn't click on. I'd still love to see the raw numbers, and then the weighting that they used. While generally weighting is a good idea, it can skew results if certain groups are weighted "too much".

Kinda what I was hoping also

[bcking 3,157]

From an article about comparing approval ratings:

"Rasmussen Reports in our daily Presidential Tracking Poll gives respondents four options – Strongly Approve, Somewhat Approve, Somewhat Disapprove and Strongly Disapprove - as opposed to just two - Approve/Disapprove. We are also the one major national pollster who asks this question only of likely U.S. voters, those who tell us they are likely to vote in the next election."

 

In their methodology though they state:

"For political surveys, census bureau data provides a starting point and a series of screening questions are used to determine likely voters. The questions involve voting history, interest in the current campaign, and likely voting intentions."

 

Those two somewhat contradict each other. In the first they state they go off those who tell them they are likely to vote (unusual, as far as I'm aware, for polls). In the second they state a more typical explanation where they use a "series of screening questions" and then have some sort of algorithm to determine likelihood to vote (I believe that is more typical).

 

If they really only ask people if they are likely to vote and then go off that, they are (as far as I'm aware) unique in using that methodology and it may explain why they are different from others.

 

 

[Amica Nostra 4,803]

55 minutes ago, bcking said:

I'd love to polls to be more transparent in their methodology about how they weighted samples. They state that they do, but I'd love to see the numbers. I'm not saying Rasmussen is bad, they may be the very best. It would go a long way to help us compare different polls that end up with significantly different results. The transparency would also help to keep the groups doing the polls honest because they have to account for everything and their work can be checked independently. 

 

Even if it was in a "supplemental data" section that most people wouldn't click on. I'd still love to see the raw numbers, and then the weighting that they used. While generally weighting is a good idea, it can skew results if certain groups are weighted "too much".

Rassmusen has a consistent 5 point skew toward Republican candidates. Nate Silver explains how they offset the bias in  polls to  create a projection

 

https://fivethirtyeight.com/features/how-the-fivethirtyeight-senate-forecast-model-works/

 

SEP. 17, 2014 AT 7:30 AM

How The FiveThirtyEight Senate Forecast Model Works

By Nate Silver

Filed under 2014 Midterms

Get the data on GitHub

•  
•  
•  

 

UPDATE (Sept. 21, 2016; 9 a.m.): The article below describes the methodology for our 2014 Senate forecasts. In all the important ways, our model for predicting the 2016 Senate elections works the same way and abides by the same principles.

---

The FiveThirtyEight Senate forecast model launched earlier this month. Right now, it shows Republicans with about a 53 percent chance of picking up the Senate next year. We owe you a lot more detail about how that forecast is calculated and how it might change between now and Nov. 4 — and how our model differs from some of the others out there.

This article, which outlines the model’s methodology, is going to be on the detailed side. I’ve tried to keep the descriptions in plain language as often as possible (the footnotes get somewhat more technical). But it’s meant to be a reasonably comprehensive reference guide rather than breezy bedtime reading.

[The Nature Boy 7,659]

28 minutes ago, Il Mango Dulce said:

Rassmusen has a consistent 5 point skew toward Republican candidates. Nate Silver explains how they offset the bias in  polls to  create a projection

 

https://fivethirtyeight.com/features/how-the-fivethirtyeight-senate-forecast-model-works/

 

SEP. 17, 2014 AT 7:30 AM

How The FiveThirtyEight Senate Forecast Model Works

By Nate Silver

Filed under 2014 Midterms

Get the data on GitHub

•  
•  
•  

 

UPDATE (Sept. 21, 2016; 9 a.m.): The article below describes the methodology for our 2014 Senate forecasts. In all the important ways, our model for predicting the 2016 Senate elections works the same way and abides by the same principles.

---

The FiveThirtyEight Senate forecast model launched earlier this month. Right now, it shows Republicans with about a 53 percent chance of picking up the Senate next year. We owe you a lot more detail about how that forecast is calculated and how it might change between now and Nov. 4 — and how our model differs from some of the others out there.

This article, which outlines the model’s methodology, is going to be on the detailed side. I’ve tried to keep the descriptions in plain language as often as possible (the footnotes get somewhat more technical). But it’s meant to be a reasonably comprehensive reference guide rather than breezy bedtime reading.

isnt 538 famous for blowing the presidential predilections in a big way ?

[bcking 3,157]

I like their "Meta-analysis" of polls...

 

https://projects.fivethirtyeight.com/trump-approval-ratings/?ex_cid=rrpromo

 

EDIT: Though I wish they would have a column that shows the rating scale each individual poll used. Since their combined percentages don't total 100% (39.6 Approve + 55.9% Disapprove = 95.5), there must be at least a few polls that allow for a "neutral" response. I also assume (they probably have it in their methods somewhere) that in polls that allow for multiple approval/disapproval options ("somewhat" vs "strongly") they are lumping them both together. Perhaps they are weighting them in some way though? (I assume combining them because the data they report from the Rasmussen poll is 46/53 which is the same as what Rasmussen reports if you combine the groups).

 

I'd also love it if they allowed you to set a specific time period to look at polls. In their methodology they describe how they use a formula for older polls to be gradually weighted less and less (to make more recent polls more important), but I'd love for it to be able to just show December polls, or all polls in the last week etc...

Edited December 28, 2017 by bcking

[Amica Nostra 4,803]

20 minutes ago, Nature Boy Flair said:

isnt 538 famous for blowing the presidential predilections in a big way ?

They predicted a 35% chance for a Trump victory. That sounds about right given the closeness and late breaking news that favored TRUMP

[Amica Nostra 4,803]

18 minutes ago, bcking said:

I like their "Meta-analysis" of polls...

 

https://projects.fivethirtyeight.com/trump-approval-ratings/?ex_cid=rrpromo

 

EDIT: Though I wish they would have a column that shows the rating scale each individual poll used. Since their combined percentages don't total 100% (39.6 Approve + 55.9% Disapprove = 95.5), there must be at least a few polls that allow for a "neutral" response. I also assume (they probably have it in their methods somewhere) that in polls that allow for multiple approval/disapproval options ("somewhat" vs "strongly") they are lumping them both together. Perhaps they are weighting them in some way though? (I assume combining them because the data they report from the Rasmussen poll is 46/53 which is the same as what Rasmussen reports if you combine the groups).

 

I'd also love it if they allowed you to set a specific time period to look at polls. In their methodology they describe how they use a formula for older polls to be gradually weighted less and less (to make more recent polls more important), but I'd love for it to be able to just show December polls, or all polls in the last week etc...

Keep poking around the site ,I think  that is out there.  They also have weekly roundatbles where they wonk out on stats and predictions.

[bcking 3,157]

8 minutes ago, Il Mango Dulce said:

They predicted a 35% chance for a Trump victory. That sounds about right given the closeness and late breaking news that favored TRUMP

Sometimes it is hard to understand that when something predicts a "35% chance" of an event happening, it doesn't mean the prediction was "wrong" when the event happens. Events with far less than a 35% chance of occurring happen all the time. It can be genuinely confusing (I mean no disrespect to anyone when I say that). It's similar to when we talk with families whose child was born with something that has a less than 1% chance of occurring, or when we counsel and given survival rates that are low or high. It can be hard to hear those "chances", experience a single event, and understand how it fits. If the "chance of survival" was 80% and your child dies, that would naturally make you uncomfortable and feel like somehow the data provided was wrong when unfortunately it was still likely right.

 

You have to look at how well the methodology they applied to Trump's chance for victory has successfully predicted events overall (Possibly just Presidential elections or, since the number of those would be fairly small, you could include other political elections). I believe they (FiveThirtyEight) are pretty good.

Edited December 28, 2017 by bcking

[Amica Nostra 4,803]

15 minutes ago, bcking said:

Sometimes it is hard to understand that when something predicts a "35% chance" of an event happening, it doesn't mean the prediction was "wrong" when the event happens. Events with far less than a 35% chance of occurring happen all the time. It can be genuinely confusing (I mean no disrespect to anyone when I say that). It's similar to when we talk with families whose child was born with something that has a less than 1% chance of occurring, or when we counsel and given survival rates that are low or high. It can be hard to hear those "chances", experience a single event, and understand how it fits. If the "chance of survival" was 80% and your child dies, that would naturally make you uncomfortable and feel like somehow the data provided was wrong when unfortunately it was still likely right.

 

You have to look at how well the methodology they applied to Trump's chance for victory has successfully predicted events overall (Possibly just Presidential elections or, since the number of those would be fairly small, you could include other political elections). I believe they (FiveThirtyEight) are pretty good.

I look at it this way, if we replayed (God Forbid) the election 9 times would Hillary prevail in 6 of those? Probably.

 

If you look at the state by state prediction that is where 538s model shines. They were really quite close state by state.

[TBoneTX 45,273]

Statistics...

 

"During the experiment, 33% of the rats lived, 33% died, and the last one got away and remains on the run in the laboratory."

[bcking 3,157]

19 minutes ago, TBoneTX said:

Statistics...

 

"During the experiment, 33% of the rats lived, 33% died, and the last one got away and remains on the run in the laboratory."

The last one was "lost to follow-up"

 

Bloody attrition bias...

[bcking 3,157]

Looking all the way back to October Rasmussen is responsible for the majority of polls of "likely voters". They are responsible for 20 polls, out of the 24 total done in that time period among "likely voters" (as opposed to "Registered voters"). McLaughlin and Associates are responsible for 3 in that time period, and Zogby Interactives did one back in October.

 

They all give pretty much the same numbers. The Zogby poll was 44%/53%, and the McLaughlin ones have been around the same. Around 5 percentage points higher (In approval) than the entire data set combined (when "likely" and "registered" are combined). Using "likely voters" may be superior if you can get close to reality, but it does add a layer of subjectivity (You can't really be subjectively "registered to vote", though if they are going off a person's answer you can still end up with misleading data). The fact that they all lean in the same direction provides external validity to each one independently, but that may either be due to their collective accuracy, or to a collective bias in how "likely voters" is determined since they all share that methodology. 

[Dashinka 24,492]

1 hour ago, bcking said:

Looking all the way back to October Rasmussen is responsible for the majority of polls of "likely voters". They are responsible for 20 polls, out of the 24 total done in that time period among "likely voters" (as opposed to "Registered voters"). McLaughlin and Associates are responsible for 3 in that time period, and Zogby Interactives did one back in October.

 

They all give pretty much the same numbers. The Zogby poll was 44%/53%, and the McLaughlin ones have been around the same. Around 5 percentage points higher (In approval) than the entire data set combined (when "likely" and "registered" are combined). Using "likely voters" may be superior if you can get close to reality, but it does add a layer of subjectivity (You can't really be subjectively "registered to vote", though if they are going off a person's answer you can still end up with misleading data). The fact that they all lean in the same direction provides external validity to each one independently, but that may either be due to their collective accuracy, or to a collective bias in how "likely voters" is determined since they all share that methodology. 

Aren’t all polls subjective?  Even voting exit polls are subjective and reliant on folks actually answering the pollsters truthfully.  Unfortunately, these pollsters are not the FBI, CBP or USCIS officers, so there is no penalty for answering dishonestly.  Regardless, news organizations, and political parties pay a lot of money for polling.