web analytics

July 27, 2016

Open Thread: Wednesday, July 27

  7:00 am

This is the open thread for Wednesday, July 27.

“Have you ever seen Elaine dance? It’s more like a full-body dry-heave set to music.”

This is a good place to post anything that would be off topic in other threads (articles of interest, polls, etc).


Fasten your seat belts . . .

  4:16 pm

To sing along, just follow the bouncing ball.  Early Karaoke

A bounce can come from a pot hole or a speed bump or a really large cobble stone.  I would argue that there was a change following the Republican convention, it was just not as large as some had hoped or others had feared.

My weighted average model shows a 3.5 point move.  Here are the before and after pictures using the six polls nearest in time.  (Sorry for the spread sheets’ blurriness. It’s not your  eyeglasses; it’s something I need to work on.)


Republican Bump Before


Republican Bump After

Is that swing from Clinton to Trump enough to constitute a “bounce”?  I leave that to the spin doctors, but here are the rules for the model:

Data in green is from FiveThirtyEight.com.  The minimum grade from there is C  (i.e. in the top 89% of 370+ listed pollsters.)    Bias is their mean-reverted bias, positive for Republican and negative  for Democrat.  Date is the final one for data gathering, not the release date.  Size is the number of respondents included in the poll, with the minimum size being 750.   Other includes undecided and 3rd parties.  Difference is always Clinton – Trump.  Unbiased is the Difference with house effect Bias removed.  Weight is Unbias times Size, so the weighting is by sample size.  On the bottom line the last figure colored in the usual way is the weighted reverted polling difference.

This model has several short comings.  It pushes “likely” voters back into the category of just registered voters.  It gives more credit to internet polls which tend have larger sample sizes.  It relies on a 3rd party to supply house bias and pollster screening data.  If you don’t like the weighting, just follow the large bold black averages in the bottom line. I hope you can see clearly what the model is doing and take it for what it’s worth.  I do this daily for the ten most recent polls.  Would you be interested in seeing these as not so fuzzy bouncing balls on a regular basis?  (Yes, I’m sampling your opinion.)


Deming’s Disciple




  3:56 pm

It’s the 3rd day here in Philadelphia and the disunity at the DNC isn’t subsiding. The Bernie Sanders supporters are still mad as hell and are not conforming to the Hillary coronation. Just as in Cleveland last week, the parties are not united. However, there is a stark contrast however.

Last week in Cleveland, the RNC and the Trump supporters showed no signs or willingness to mend the fences with the Cruz and Kasich supporters. They did everything to strip away Cruz delegates from amending the rules for future contests. The Trump delegates who announced their state tallies neither congratulated nor complemented any other candidates not named Trump. There were no speeches or videos highlighting the 55% of Republicans who worked hard for other candidates. Granted, Cruz and Kasich didn’t exactly embrace Trump but the way the Trump and party leaders treated them, it’s not surprising that they didn’t. The feeling that I got in Cleveland from the party and Trump supporters were that they hate the non-trump wing of the party and the booing of Cruz proved that.

The Democrats on the other hand were completely the opposite. The party leaders and Hillary supporters heaped non stop praise on Bernie and his supporters. The party is showing that it really wants to unite and get the Bernie Sanders supporters on board. They were sending out that olive branch all day yesterday and Mon. Yet the majority of Bernie supporters have no interest in it at all. Speaking to many of them on the floor after Bernie’s speech on Mon night and I got all different replies. One woman from CA said she doesn’t like Hillary but she’s open to wait and see if she’ll vote for her. A elderly guy from IL said he’s going to support her begrudgingly. The rest however were Bernie or bust. On Tuesday after Hillary got nominated I chatted with a few of the Bernie walkouts outside. They now view Bernie as a sellout and are all voting for Jill Stein or staying home.

RNC – did everything to make it into a Trump convention and ignore the opposition. The main opponent didn’t outright endorse the nominee and got booed. The nevertrump camp appreciated what Cruz did but the Trump people are livid at him.

DNC – did everything to make peace and unity in the party, even getting rid of DWS last minute and heaping praise on Bernie all week. The main opponent embraced the nominee and got booed by his supporters. The Hillary supporters appreciated what Bernie did but the Bernie crowd are livid at him.

It was basically a lose lose situation for both parties but that’s what you get when you nominate unpopular candidates.



Pluralism and Liberty

  8:00 am

This is Part 2 of 3 in a series on pluralism. Part 1 can be found HERE

As I write this, I’m sitting in my office at home. On my right, on a bookshelf, sits a small bust of Vladimir Lenin – I can explain. My father, a 31-year career Army officer, returned with it as communist memorabilia sold (with a delicious irony) on the streets of Budapest. Under the bust of Mr. Lenin are books of all sorts on topics too numerous to mention: Jesus, the Iraq War, Caesar, travel, Spanish dictionaries, civil disobedience, political party theory, the U.S. Constitution and more.

In one room sit both the architect of an ideology that rejected pluralism of thought and philosophy, and whose nation rapidly deteriorated into predictable tyranny, as well as the fruits of decades of the West’s embrace of pluralism, emblematic of the wealth and prosperity it facilitates.

In my last entry, I discussed why pluralism and patriotism are synonymous, that pluralism is and must be the reason for our patriotism. I received some fantastic feedback, but one question stood out: “Isn’t liberty, not pluralism, why we celebrate the United States?” In the last entry, I asked, “Patriotic about what?

Today I ask, “Liberty to do what?”

The Constitution guarantees certain freedoms that the Founders believed in existed in natural law. That is, the Founders, shortly after the construction of a republican government (which has been since torn to tatters by the almost intentionally counterproductive injection of far too much direct democracy … more on that at another time), theorized a state of nature, attempted to discern what rights existed in that theoretical pre-government humanity, and sought to codify the results.

The result was glorious, and boils down to essentially one thing – pluralism. The right to be and believe what one wants, to associate with whom one wants, to maintain privacy in the exercise of those rights against government actors (and in a democracy, that means one’s neighbors), and to defend those rights against those who would seize them with force, if necessary.

There are two conceptions of the ends of liberty. Some believe that liberty is its own end, that it is morally correct that man should live according to his own conscience and the maximization of his own abilities. Others believe that the goal of liberty is one step further, to create a more peaceful, stable society. Regardless of one’s approach to the question of the end of liberty, pluralism is the beginning and end of the answer.

What is liberty if not the right to be something different from one’s neighbor? Indeed, there is no point at all in having liberty if the hope is for a society in which there are but a few modes of expression, dress, belief, attitude, and political outlook. The liberty you, the readers, highlighted in some of your questions, is meaningless unless it is the liberty to choose one’s own mode in a pluralistic society. Pluralism is not secondary to liberty; pluralism is liberty.

Republicans today, as exemplified in that fascistic farce in Cleveland, cling to the language of liberty, but have little to no regard for the thing itself, especially the foundational First Amendment. The modern Republican Party, through its nominee and in a Sophocles-style plot twist, have become the judicial activists, inserting caveats into the First Amendment where there are none, all to supplant the pluralism intended and anticipated by the document.

Some examples:


“Congress shall make no law respecting an establishment of religion, or prohibiting the free exercise thereof; or abridging the freedom of speech, or of the press…[.]”

Where are the caveats Donald Trump insists should be there? Mr. Trump on religion:

“But just remember this: Our Constitution is great. But it doesn’t necessarily give us the right to commit suicide, okay? Now, we have a religious, you know, everybody wants to be protected. And that’s great. And that’s the wonderful part of our Constitution. I view it differently.”

– Donald Trump, Meet the Press (07/24/2016)

Mr. Trump on freedom of the press:

“One of the things I’m going to do if I win…I’m going to open up our libel laws so when they write purposely negative and horrible and false articles, we can sue them and win lots of money. We’re going to open up those libel laws. So when The New York Times writes a hit piece which is a total disgrace or when The Washington Post, which is there for other reasons, writes a hit piece, we can sue them and win money instead of having no chance of winning because they’re totally protected.”

– Donald Trump, Fort Worth, TX (02/26/2016)

It is plain that Republicans do not value pluralism and so have chosen a nominee that seeks to undermine liberty in order stymie it. If pluralism and liberty were so severable and so distinct, Mr. Trump would be able to espouse a non-pluralistic vision for the United States without denigrating liberty.

He cannot because pluralism is liberty. Full stop. So, please, conservatives, recognize this reality. One can love liberty or despise pluralism, but he cannot do both.

Just ask the dearly departed Mr. Lenin.


Cutting on the bias

  7:15 am

Just because you are right does not mean you are unbiased. 

Internet polls (database polls) have a skeleton in their closet, namely, the 1936 Literary Digest poll that predicted President Alf Landon’s landslide 57% to 43% victory. What went wrong? There were two problems with the 1936 poll. It had a selection error and a response error.

First let me explain the selection error in creating the 1936 sample. During the depression only upper middle class people had phones or subscribed to the magazine. These were the people who were mailed a poll ballot. The rest of the country could not afford the subscription nor the phone and were not selected to participate. That is a biased sample.

The second problem was the response error. It was an era in which many were making 10 cents an hour, so spending 3 cents on a stamp was akin to $3.00 today. Out of 10 million people sent a ballot, 2.4 million responded, but 7.6 million did not elect spend the time nor the postage! The respondents were disproportionately Republicans. In other words, motivated respondents with money to spare selected themselves into the process. This “volunteering” to be polled is the essence of response error. It is a problem for live phone surveys where people screen their calls and is a major problem with internet polls.

In correcting for this, another issue is created: the adjusting of the poll responses to reflect a model of the voting population based on census data. It sounds good but this is a version of quota sampling. It differs in that it is done after the fact on the data base rather than by a live interviewer who is subject to bias and human error in conducting the interview. The problem with quotas is where does the list of quotas to be met come to an end, and what does it include? How finely can you apportion the electorate into cubbyholes and still get it right? How much adjusting for each one needs to be done? This handling of quotas to correct for bias is the secret sauce of polling companies.

Compounding all of this is the fact that a census becomes less reflective of reality the older it is, even if it is subjected to a model of population change. Likewise, unseen changes in the demographics of the electorate can also be a source of error. For example, if non-college educated whites move from 60% participation to 62% in a given election how does that distort the panel? So we have a model on a model to correct for an inherent response error. That makes three entry points for bias.

Perhaps an internet pollster did a good job of getting the right answer with his secret sauce in 2012 based on the 2010 census and past electorate behavior but that is no guarantee of future performance. We can argue endlessly about theory but it is only repeated success over time that can validate internet polling and then only for that pollster. In the meantime, my solution has been to use FiveThirtyEight’s grading system to filter out inferior pollsters and to adjust for the history of bias in those I include. Well that’s my bias.

-Demings Disciple


July 26, 2016

Poll Watch (NBC/SM): No Convention Bounce For Trump

  3:27 pm

NBC News / Survey Monkey Weekly General Election Tracking Poll:

  • Hillary Clinton: 46% (46%)
  • Donald Trump: 45% (45%)

W/ Third Party Candidates:

  • Donald Trump: 41% 
  • Hillary Clinton: 39%
  • Gary Johnson: 10%
  • Jill Stein: 5%

Rate Trump’s Convention Speech (Excellent or Good / Just Okay / Poor or Terrible):

  • Republicans: 84% / 10% / 5%
  • Independents: 30% / 29% / 40%
  • Democrats: 10% / 19% / 70%

View of GOP After Convention (More Favorable / Less Favorable / No Change):

  • Republicans: 52% / 11% / 36%
  • Independents: 9% / 42% / 47%
  • Democrats: 4% / 68% / 27%

The NBC News|SurveyMonkey Weekly Election Tracking poll was conducted online July 18 through July 24, 2016 among a national sample of 12,931 adults aged 18 and over who say they are registered to vote. Respondents for this non-probability survey were selected from the nearly three million people who take surveys on the SurveyMonkey platform each day. Results have an error estimate of plus or minus 1.2 percentage points. For full results and methodology, please click here.


Poll Watch (PPP, Landmark/Rosetta Stone): Close Races in Ohio, Georgia

  2:37 pm

Landmark / Rosetta Stone (R) Georgia General Election:

  • Donald Trump (R) 45.5% 
  • Hillary Clinton (D) 44.2%
  • Gary Johnson (L) 5.2%
  • Jill Stein (G) 2.5%

Among Whites

  • Donald Trump (R) 67.6% 
  • Hillary Clinton (D) 21.3%
  • Gary Johnson (L) 6.0%
  • Jill Stein (G) 3.5%
Among Blacks
  • Hillary Clinton (D) 87.6%
  • Donald Trump (R) 6.2%
  • Gary Johnson (L) 2.5%
  • Jill Stein (G) 0.6%
Survey of 500 likely Georgia voters was conducted July 24, 2016 for WSB-TV. The margin of error is +/- 4.4 percentage points.
  • Donald Trump (R) 42%
  • Hillary Clinton (D)  39%
  • Gary Johnson (L) 6%
  • Jill Stein (G) 2%

Two-Way Race: 

  • Donald Trump (R) 45%
  • Hillary Clinton (D) 45%

Among Men

  • Donald Trump (R) 50% 
  • Hillary Clinton (D) 40%

Among Women

  • Hillary Clinton (D) 50% 
  • Donald Trump (R) 41%

Favorable / Unfavorable {Net}

  • Donald Trump 37% / 56% {-19%}
  • Hillary Clinton 35% / 54% {-19%}

Survey of 1,334 registered Ohio voters was conducted July 22-24, 2016.  The margin of error is +/- 2.7 percentage points. Party ID: 39% Democrat; 38% Republican; 23% Independent/Other.

H/T to The Argo Journal.

Poll Watch (CBS/NYT): Trump Holding 1% Lead Nationwide, Johnson at 12%

  2:31 pm

CBS News / New York Times National General Election:

  • Donald Trump (R) 44% {40%}
  • Hillary Clinton (D) 43% {40%}

W/ Gov. Johnson:

  • Donald Trump (R) 40% {36%}
  • Hillary Clinton (D) 39% {36%}
  • Gary Johnson (L) 12% {12%}

Favorable / Unfavorable {Net}

  • Donald Trump 34% / 53% {-19%}
  • Hillary Clinton 31% / 56% {-25%}

National survey of 1,118 registered voters was conducted July 22-24, 2016. The margin of error is +/- 4.0 percentage points. Results from the poll conducted July 8-12, 2016 are in curly brackets.

H/T to The Argo Journal


Things that go bump in the night.

  2:18 pm

“Predictions are always difficult, especially about the future.”  An old Danish proverb attributed to Yogi Berra who was apparently much older than we thought and Danish.

Polling and astrology have a lot in common. Historically, astrologists pursued the accurate and precise location of each influential planet. They recorded the information and used it to predict the future, including the future locations of planets. This observational work actually paved the way for astronomy. Indeed, some early astronomers earned their living as astrologists to the gullible aristocracy.

Polling likewise seeks the accurate and precise location of public opinion, both at the time of the poll and into the future. But opinion is ill behaved, because people change their minds. This doesn’t stop pollsters from earning a living selling their information to marketers, who tend to be a superstitious lot.

However, unlike the locations of the stars, the position of the public’s opinion is up for debate. This debate is publicly apparent in presidential elections where the various internet sites’ models produce different results. Here are some results from poll aggregation sites as of early on July 24, 2016:

  1. Huffington Pollster: 3.2% Clinton
  2. Real Clear Politics: 1.9% Clinton
  3. Pollheadlines: 2.25% Clinton
  4. Election Projection: 2.9% Clinton
  5. 270 to Win: 3.5%C linton
  6. FiveThirtyEight (polls-only forecast): 1.7% Clinton
  7. New York Times: 2% Clinton

But how can there be different answers to the mathematical question: what does the average polling show? The differences come from how many polls they use, how (or if) they weigh them, which polls they exclude, how diligent the site is about updating its list of polls, when polls have two results (one with minor parties and the other without) which result they choose, and whether their final result is rounded.

Consider for example the number of polls used: Huffington Pollster uses 239 with proprietary and unrevealed weighting, FiveThirtyEight shows some tiny weight to even the 207th poll on its list, Pollheadlines has a proprietary scheme that does not disclose the number of polls nor the weights used by the model, Election Projection uses 20,  Real Clear Politics uses 10, The New York Times and 270 to Win each use 5 polls.

Which polls are used? FiveThirtyEight evaluates over 370 pollsters so there are a lot of polls to pick from. Everybody uses ABC/Washington Post, CBS/New York Times, CNN/ORC, Marist College, NBC/Wall Street Journal, and Rasmussen. But only four of the aggregators include Morning Consult Polls. Real Clear Politics and Pollheadlines have never included a poll from Morning Consult which does not mean much. However, more troubling is when a pollster’s results are selectively included. For example, 270 To Win included Morning Consult’s poll of August 8, 2015 but none since then. If they dropped the pollster based on some criteria, then that should be findable on their site. Does that mean their result is “wrong”? No, it could be perfectly correct but this selectivity does help explain how poll  aggregators can have different answers. Time will tell who had the correct prediction.

Update: Prior to today, Monday July 25, the polling average of poll aggregators hovered around 2.6% Clinton. Now, after only 3 new polls (whose average has been 2.7% Trump) the poll aggregation average has dropped to 0.9% Clinton. This seems to be pointing to a convention “bump” of over 5% for Trump. The next 7 polls should clarify the exact size of the bump, but the bump is real.

–Deming’s Disciple

Editor’s Note: This is Deming’s Disciple’s first post as a new contributor at Race. He is a retired mathematics/statistics teacher and will be providing us with professional poll analyses throughout the election. Welcome aboard, Deming’s Disciple!


Open Thread: Tuesday, July 26

  2:15 pm

This is the open thread for Tuesday, July 26.

“Yada, yada, yada…”

This is a good place to post anything that would be off topic in other threads (articles of interest, polls, etc).


Recent Posts

Tweets by @Racefour

Search R4'16