Sunday, July 31, 2016

Why USC/Dornsife/LATimes Poll Is The "Gold Standard" Of Presidential Polls




UPDATE POST ELECTION
Megyn Kelly ! explains why USC/LATimes got it right
(Video from 5:57 to 7;50)
*************************************************************
Post election update #2. 

The man behind the USC/LATimes poll Arie Kapteyn explains his methodology. What is especially impressive is his statement that he personally wanted Clinton to win. This example of total non-bias is instructive and a lesson to other polling firms and pundits  who failed so badly.

Via LATimes

As election returns rolled in Tuesday night, the creator of the USC/Los Angeles Times Daybreak tracking poll was in Washington for a speaking engagement. He watched the results on television in a hotel bar, surrounded by about 20 drunken Danes who were in the capital to study the election.


“It was an odd experience,” Arie Kapteyn said Wednesday morning.




***************************************************************************
ORIGINAL POST

With IPSOS/Reuters Polling utterly exposed and exploded as under reporting Trump's results by up to five points and then tweaking Hillary's upwards the polling industry, which had massive failures with "Brexit" and Donald Trump's primary campaign, is in major trouble.


Examples such as "RABA Research giving this "result" a 15 point Clinton lead

2016 General Election - Clinton 46%, Trump 31% (RABA Research 7/29)

while USC/LATimes poll on the same day give Trump a four point lead makes a mockery of any pretensions pollsters in general may lay dubious claim to.

It is my opinion that the USC/LATimes poll is the only reliable and best reference from all the polling companies mainly because of their voter pool and stability of their reporting since the poll commenced
on July 10th especially as this was a turbulent time with the Wikileaks sensations and both conventions.





Whereas traditional polling firms canvass a different group each time they poll USC has a pool of about 3000 pre-selected from which pool they choose one-seventh of the members and ask them a series of three set, unchanging questions.

The question that arises, as it does with all polls, is "what is the composition of those polled as regarding previous support i.e. how many
Dem's/Republicans/Independents? 

It is the balance of these groups that can distort polls as the Reuters and RABA examples clearly show and the Mason Dixon poll of Pennsylvania which had a much lager number of Democrat's in their sample set which gave a 9 point Clinton lead, which is more than president Obama won the state by in 2012.

What these polling outfits hope to gain by losing their credibility in pursuit of their agenda is a mystery.

The USC proportion of respondents is now shown but given the closeness of their polling which is in line with the "gold standard" the aggregate of all polls it seems apparent that they have their balance correct and that any major movement to either candidate from a static pool of voters reflects changed minds and a significant event.

At this point it can be fairly stated that the 'gold standard" for polling is the Real Clear Politics aggregate of all  (responsible) polls and the gold standard for a single poll is USC/LATimes.

I would also commend The PPD Poll as of value as it uses as similar mechanism and its result, although slighty more favorable to Trump, mirrors USC's e.g.
7/30/16. Their mechanism and latest update is at the link above.
USC Trump 46.1 Clinton 41.9 Clinton +1.3 points post DNC PPD Trump 46.8 Clinton 42.2 Clinton +0.7 points post DNC

NB; "PPD "The result of the People’s Pundit Daily U.S. Presidential Election Daily Tracking Poll released on Saturday July 30 marks the first time our internal polling data, which helped propel the PPD Election Projection Model to #1 in 2014, has been released for public consumption."

********************************************************************
Here is the USC/LATimes poll advice on how they conduct and report their findings


LA Times Tracking Poll response:

“I have no magic words. Here is what the Director of our center, Arie Kapteyn, who devised this polling approach, has to say about it:
We are being as transparent as possible by posting all information about our approach (including data, and sampling and weighting procedures) in excruciating detail on the web. We have set our procedures at the beginning of the data collection and now we just wait and see what happens. The outcomes are definitely striking and indeed different from most other polls. Four years ago, when we did the RAND Presidential poll ( most of us were at RAND at the time) it looked as if we had a bias in the direction of Obama, since we were always more to the left in our forecasts than most other polls. When all the votes were counted by the end of 2012 we were actually underestimating the final count for Obama (by a small amount) and we were closer to the final tally than any other major poll. That of course does not imply at all that we are right again this time. We have a new sample (but essentially the same method) and outcomes may just vary because samples are never a perfect reflection of the population. We are looking at the outcomes every morning with as much interest as anyone else. So we are not claiming to be right and everyone else is wrong; we have simply done the best we could and this is the result. Only in November will we know if we were right or not.
Not sure that is helpful if you are engaging in debate about which poll is correct. We have a long way to go until November.”
Frank

About the Survey✝ 

The USC Dornsife/LA Times Presidential Election "Daybreak" Poll is part of the ongoing Understanding America Study: (UAS) at the University of Southern California’s (USC) Dornsife Center for Economic and Social Research, in partnership with the Jesse M. Unruh Institute of Politics and the Los Angeles Times. Every day, we invite one-seventh of the members of the UAS election panel to answer three predictive questions: What is the percent chance that… (1) you will vote in the presidential election? (2) you will vote for Clinton, Trump, or someone else? and (3) Clinton, Trump or someone else will win? As their answers come in, we update the charts daily (just after midnight) with an average of all of the prior week’s responses. To find out more about what lies behind the vote, each week we also ask respondents one or two extra questions about their preferences and values. The team responsible for the USC Dornsife/LA Times Presidential Election Poll four years ago developed the successful RAND Continuous Presidential Election Poll, which was based on the same methodology.


Survey Methods 




The USC Dornsife/LA Times Presidential Election "Daybreak" Poll is part of the Understanding America Study (UAS) at the University of Southern California's Dornsife Center for Economic and Social Research. It is being conducted in partnership with the Jesse M. Unruh Institute of Politics and the Los Angeles Times.
The team responsible for the Daybreak Poll four years ago developed the successful RAND Continuous Presidential Election Poll, which was based on the same methodology.
The Daybreak Poll is based on an internet probability panel survey. Daybreak Poll members are participants in the ongoing UAS internet probability panel of about 4500 U.S. residents who were randomly selected from among all households in the United States. Members of recruited households that did not have internet access were provided with tablets and internet service. The UAS panel is still growing. We project it will reach about 6000 members in the coming months.
More than 3200 UAS panel members so far (July 2016) have agreed to participate in answering questions about the election, and we expect that number will increase over time. Each day, 1/7th of those who have agreed to participate (more than 400 per day) are invited to answer three predictive questions: What is the percent chance that... (1) you will vote in the presidential election? (2) if you were to vote, you will vote for Clinton, Trump, or someone else (percentages add to 100) and (3) Clinton, Trump or someone else will win (percentages add to 100). The order of the candidates in the questions is randomized so that about half of the respondents see Clinton as the first choice and half of the respondents see Trump as the first choice.
Each night, Daybreak Poll results are weighted to match demographic characteristics (such as race and gender) from the U.S. Census Current Population Survey, and aligned to the 2012 presidential election outcome using how respondents tell us how they voted in that election. Then the latest results, averages of all of the prior week’s responses, are posted online at https://election.usc.edu and on the LATimes.com Politics site shortly after midnight.
In particular, to obtain the values shown in the election forecast chart, we weight each respondent's likelihood of voting for a candidate with their likelihood of voting in the presidential election. Next we calculate the mean of that number for all respondents during the last 7 days, taking into account respondent level weights based on demographics and past voting behavior. This is the estimated fraction of the population that will vote for the candidate. The graph shows the estimated fraction of the votes that a candidate will get, which is computed by dividing the estimated fraction of the population that will vote for the candidate by the estimated fraction of the population that will vote for any candidate. The latter is analogously obtained as the weighted mean of the respondents' likelihood of voting in the presidential election.
To find out more about what lies behind the vote, each week we also ask respondents one or two extra questions about their preferences and values. Links to documents detailing question text, sample sizes, response rates and other information for these separate surveys are provided in the detailed information section below, linked to stories or press releases where the results were disseminated.
The Daybreak Poll began on July 4, 2016, and will run through the November election.
More information about UAS panel methodology, the panel management and survey software we developed, or our publicly available datasets are available in the links here or at the UAS site (https://uasdata.usc.edu). For other questions, or to inquire about how you can conduct surveys with the UAS panel, contact us.

Summary of links to more detailed information about the UAS Panel and the Daybreak Poll 

UAS Panel Sample and Recruitment
UAS Panel Weighting (pdf) 
Details of weighting the Daybreak Poll
Screen shot of the 3 weekly vote questions (note order of candidates is randomized)


UNDERSTANDING AMERICA STUDY

THE USC DORNSIFE / LA TIMES PRESIDENTIAL ELECTION "DAYBREAK" POLL


The 2016 USC Dornsife / LA Times Presidential Election Poll represents a pioneering approach to tracking changes in Americans' opinions throughout a campaign for the White House. Around 3000 respondents in our representative panel are asked questions on a regular basis on what they care about most in the election, and on their attitudes toward their preferred candidates. The "Daybreak poll" is updated just after midnight every day of the week. 

No comments: