« Saudi Arabia is our most worthless "ally." | Main | Diazepam? »

08 August 2016


Feed You can follow this conversation by subscribing to the comment feed for this post.


When numbers are jumping like they have been (and when numbers are so completely varied across multiple polls), that means the world is too messy (or uncertain, or has too much variance, or whatever) inherently and we literally don't and can't know. The pollsters are being dishonest, knowingly or unknowingly, when they say we can reliably predict X or Y with actual numbers when we see things like that.


Well, I brought up the Republicans because their vote, I think, will be much more decisive.

Union voters, I think, are rather less important now: there aren't too many of them left and they don't vote, and many of those who do have already gone over to the Republican camp. Now, Sanders did show that there are enough of them who are still Democratic enough to vote for a right kind of Democrat and they made up a significant minority of the Sanders voters--and I think they are the voters who are most resistant to the Clinton appeal. Still, I estimate that if they swing over to Trump in any large numbers, Trump could survive losing a decent (but not many) traditional Republican voters. These odd groups of "swing voters" are the most fascinating aspects of this campaign.

Edward Amame


I'm absolutely certain there are websites out there where Tyler's political sensiblities are pretty conventional. Parading them around mainstream websites in a comic Mike Hammer-esque voice may seem refreshing to you, but to me it sounds like the same old thing in brand new drag.


His conclusions probably are, but that would be same thing as saying that Aristarchus understood modern physics because he believed in heliocentrism, and it's not as if "certain websites" are a reflection of the real world--and if you think they are, then you have bigger problems than Tyler.

Quite frankly, I don't really care what his opinions are: they are his and he is entitled to them. I also know that, like Carlin and Rock, he is quite earnest and honest when he makes his points which are, whether I like it or not, an important component of the public opinion in this country today and understanding their reasoning is important. In this sense, I appreciate that Tyler is direct and forthright.

If, as you conservatives and liberals seem eager to do, all the people whom you disagree with are to be treated with ignorant contempt (literally--in that you don't care to know why they disagree with you other than that they do), what chance do you have to understand what makes them tick? Without understanding what makes them tick, how would you expect to win whatever you hope to win? By faith? Because "God/Righteousness/Reason" is on your side, that you know because you have faith?



It seems this has proven (((very effective))) in triggering people who have been anti-gentile, anti-western and anti-American. Somone on the alt-right should be given cudos for driving the left mad with some brackets.


I've just gotten back from a trip, and all I can say is that a lot of you are abusing each other over a topic of which you have very little understanding. Let me try to explain the whole party identity vs. voting thing in as simple and non-partisan a way as I can.

Party identity is attitudinal data. It’s about how you feel about something at a particular point in time. By itself, attitudinal data is noisy, fluid, and does not correlate strongly with outcome. In market research, it’s useful but should never be used by itself if you can avoid it. In polling, where you usually have more objective data, you should simply ignore it, which most pollsters do. The idea that you should correct relatively objective data with attitudinal data, which many here are suggesting, is just plain wrong.

To see why, let’s imagine you have a town of 15,000 inhabitants with a Walmart and a Target. One day, a company does an independent survey asking whether people consider themselves Walmart shoppers or Target shoppers. Not surprisingly, 70% identify with Target, and 30% with Walmart. This is your attitudinal data.

A few days later, a different company does a second survey. It first asks people how much money they spent at each store in the last week. Then it asks whether they are Walmart shoppers or Target shoppers. The answer to the first question reveals that 80% of the money was spent in Walmart and only 20% in Target. In addition, 60% of all shoppers in that survey say they consider themselves Walmart shoppers and only 40% Target shoppers.

What many of you are saying is that such an outcome “proves” that the second company deliberately over-sampled Walmart shoppers to produce a favorable outcome to Walmart. Obviously, this is nonsense. Rather, when asked in the abstract, people had one identity, and when reminded of their actual behavior, they had a different one.

In addition, attitudinal data is not suggestive of outcome. If we did a third, purely attitudinal survey that revealed the identity split was now 75/35 in favor of Target, we should not expect any meaningful difference in customer behavior or store revenue.

Pollsters have long recognized that answers to questions about party identity, counterintuitive as it may seem, do not correlate enough with voting behavior to be useful in predicting outcomes. That’s why they don’t account for it and certainly don’t adjust for it.


I've run into the same disdain for using party id for stratifying polling data among commercial pollsters and I'm a bit confounded by this. Even in the Target vs. Wal-Mart example you mention, those who say that they like Target more than Wal-Mart are more likely to shop at Target, ceteris paribus. While insisting that the sample must balance Target and Wal Mart supporters is silly, as there are many other factors, insisting that it must be excluded as a factor seems equally silly: companies do spend billions trying to build up their name brands, after all, in recognition of this. In electoral politics, the correlation between party id and vote choice has always been considerable and has become much greater last few decades. Accounting for party id, if it can be done accurately, does improve the predictive power of polling analysis greatly and is, furthermore, consistent with our understanding of American party politics today. (as can be seen in this paper http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf) Now, there are multiple problems with accounting for parties in polls, but they can be dealt with relatively easily. Rather than trying to balance out the sample, party id information can be used in the post stratification process along with other demographic data, as per Gelman et al. While the conditional probabilities of vote choice by each partisan-demographic group is uncertain, we do have data for that, literally. Of course, the composition of the voters who turn out on the election day, including the partisan makeup, is unknown, you can take your best shot at it, again from the data that you have. All in all, why shouldn't take party id into account, even if not as the decisive factor, knowing all that we know about relationship between voter behavior and party affiliation?

Eric newhill

I'm with you. I am not buying what Shepard says - I am not a poll/survey expert by any means, but I do have a vast amount demographic and statistical analysis experience. Shepard's views are counterintuitive. I am open to being educated. It would require some serious scientific/scholarly explanation.

Agreed that sample doesn't have to equate to the population. If not, results must be weighted - and they are not in these polls.

I, like you, can accept that party identification is not a perfect predictor of voting behavior. However, it has to be pretty strong. Again, avoiding all of the shop talk gobbedlyguk and working off basic common sense, are there actually Democrats that will vote for Trump? Are there Republicans that will vote for Clinton? In both cases, the answer is probably, "Yes, a small few".

Otherwise,Ds will vote Clinton and Rs will vote Trump and independents are up for grabs. The % of Rs, Ds and Is in the poll will determine which candidate will lead the poll if the poll results are not weighted to reflect the population proportions.

Bottom line - If you have a sample that is not representative of the population, you will have a poll result that is not representative of the population.

I'm probably risking a scolding from Richard, but saying that party affiliation is "attitudinal" seems, to me, like a fancy way of saying nothing - or a misapplication of theory. Political party affiliation is not nearly as ephemeral as which junk store one shops at. As Kao says, brand loyalty is important and nowhere is brand loyalty as strong as in politics (my perception anyhow)

Eric newhill

But at least shepard's explanation - assuming he is familiar with polling methodology - answers some of the questions as to why experts would be cavalier about sample proportions.



I'd forgotten about the Gelman paper. It was very provocative and a lot of fun when it was released, but my understanding was that no one has been able to (or maybe tried to) use its findings successfully in a live setting.

For what it's worth (and I'm sure you know) the paper has a very restrictive data set--45 days of a single election. I remember reading about it and thinking, this is really interesting, and I was hoping to see its findings confirmed elsewhere. And then I never heard of it again.

Do you know of any research that follows up on Gelman? I don't mind being proved wrong.


New Rasmussen poll has Clinton +3 in a four way race. The reuters poll referenced above is not an outlier nor is the one below.


different clue


How does Scott Adams know what anyone else "can't know"? Does he know about trade policy and its effects his own self? If he doesn't, how is he in a position to assess whether someone ( anyone) else's stated positions and understanding of trade policy is information-based and logic-achieved or not?

Before NAFTA was passed, numbers of people predicted it would lead to job losses in America. Some people even predicted it would lead to a rise in illegal immigration from Mexico into America. And those predictions turned out to come true as predicted. If the people making those predictions did not understand about the effects of the proposed NAFTA trade policy they were discussing, how did they make predictions which ended up coming true just as predicted? How would Scott Adams explain it?

( And by the way, if Trump ends up losing, what happens to Adams's self-burnished reputation as a master persuasionologist and master political predictor? He'd better hope Trump wins, or I might not be tempted to buy his book).

Eric newhill

shepard and Kao,
I'm probably obsessing about this and should have let it go several comments back, but I find this discussion very disturbing. Here is a topic I know a lot about - not polling or surveying per se, but research and statistical analysis and there are only two people who have any idea what end is up; Tyler and myself. Kao so tepidly responds that what he says loses force.

I come to SST for the high quality of the posts and comments. A lot the material is outside my real of knowledge or experience. So I stay quite and learn. Now I am worried. If a topic that I understand can be so terribly misrepresented, what is happening on those topics where I am trusting others to be knowledgeable.

Party affiliation is highly correlated with what candidate you vote for, In the general population, If there are 26% Rs and 26% Ds and the rest identify as Independents, then your poll sample MUST reflect those population proportions (or you MUST weight the results of your sample) BECAUSE party affiliation IS so highly correlated with candidate choice on election day. Party affiliation could probably even be said to be causational in a normal election cycle. I can't believe that a discussion full of grown men that are supposed to be intelligent and with some who are claiming research credentials can't grasp this basic concept.

No? Then what if I told you there is a poll with a sampling consisting of 80 Rs and 10% Ds and 10% Ind and that Trump is killing Clinton in it 71% to to 11% (balance goes to Johnson and Stein). You wouldn't point to sampling as a criticism?

Edward Amame


What makes Tyler tick?

I don't know, but I do know that anti-immigrant groups have been around since the beginnings of this country, so there is probably no "winning," at least in the short run. Hispanics, blacks, Asians, Jews, women and gays all have lobbies now to protect themselves/advance their interests. That suggests to some that white nativism has been, or will be, the response and that violence might be the result. I don't. The days of the Posse C and militia groups are probably over since the Oklahoma City bombing. Since that time, there's been an effort to mainstream nativism via the Tea Party. Candidate Trump has my heartfelt thanks for putting aside the dog whistles and speaking directly and forthrightly to/for nativists. There will be a referendum on that in November. There's talk that they'll resist majority rule if a President Clinton is elected, there are threats that there will be insurrection. If it comes to that, those threats would have to be addressed. There are economic concerns of whites with only high school education that will need to be addressed, but congress....

Meanwhile, this is how immigrants won back in 1990s California and it's instructive. There was a GOP backlash against immigration there that led to passage of a slew of ant-immigrant propositions, most of them specifically targeting Latinos -- the same Lations who had voted out 50% for Ronald Reagan as governor. Latinos organized and along with enough white voters alienated by those propositions, eventually flipped California eventually from being a GOP leaning state with a GOP governor and legislature to having Dems control every statewide elected office and being heavily pro-immigrant by 2016.



Yes you are a shill, and yes, they are in the tank for Clinton, as they would have been in the tank for any other Republican candidate.

When they screw with their methodologies so massively in order to create a "Hillary inevitable" narrative, what else can you call it but what it is?

Well, unless you're a progressive borg drone, then you invent new terms in order to pretend that your girl isn't doing horribly when she can't fill a high school gym while Trump stuffs 10k+ into an auditorium. Another question of who am I to believe? You or my lying eyes.



Mr. Habakkuk nailed it, so there's not really a whole lot for me to say except to add onto Fred's "Are you triggered?" question.

We tried the Tea Party, and a bunch of old people who liked to LARP as colonial era actors got made into a group of crazy racists.

Now you get the Alt Right Nationalists in the form of Trump, who you are trying to suborn as crazy racists.

The snap back keeps building as the Left tries to stuff anarcho-tyranny down the throats of people who just want to be left alone.
At least you'll finally be right about the "crazy racists" part when the RWDS kick in your door.


Eric Newhill,

While it's amusing to be called dangerous, I'll have to decline the title.

I'm sorry, but you've made a basic mistake in terminology, which I think drives a lot of the confusion in this discussion. Party affiliation, i.e. which party you belong to, is an important metric. It is highly correlated to how you vote and important in polling. You have to account for it, and polls do. In that, you and I are in complete agreement.

Party identification is something entirely different and the beloved metric of conspiracy theorists and those behind in the polls. It refers to how you answer a question about which party you feel you identify with in the course of a survey. It's not whether you are a Democrat or Republican, it's whether you feel like one at that moment.

Obviously, this is a problematic question. Think of this election cycle, where a lot of registered Republicans may or may not feel like Republicans based on what Trump said yesterday. Or they may feel like Republicans in a general survey, but not when reminded of Trump's leadership of the party and their intention to vote for Gary Johnson. The same person will answer this question in different ways at different times, not least depending on the context provided by the survey itself. Thus, using one survey's results to judge the accuracy of another, which you suggest, doesn't work. Nor is the metric stable enough normally to use in predicting results.

Kao brings up a really intriguing paper that calls this basic premise into question. It analyzed the last month and a half of the Obama/Romney campaign. Its conclusion was that the electorate had become so partisan that party identification had become consistent. Thus, it was something from which you could derive meaningful, predictive insight. However, to my knowledge the paper has not been peer reviewed, has not been widely cited, and has not been confirmed by other research. Stay tuned.

Kao and I are not tepid or fanciful. I think we're both practitioners in our own ways and come at this from very different angles.



You are a historical illiterate.

The "backlash" occurred before Prop 187 passed, because of REAGAN'S AMNESTY. This is more liberal "memory holing of history".

Mind you, Prop 187 was voted in and failed because Gray Davis refused to appeal when the Ninth Circus' unelected federal mandarin decided that it was "rayciss".


And you don't think the already resident population has any right to object to their standard of living and the cohesiveness of their communities being undercut by mass immigration? You think they should just, for want of a better phrase, bend over and take it? They've been bending over and taking it for a long time and now they've had enough. How can you blame them?



"What makes Tyler tick?"

So many pithy answers here. The old Conan line about "driving your enemies before you" comes to mind, or maybe "Make America Great Again".

While those have their appeals, the honest answer is that progressive policies lead to tyrannical, violent, hellhole and the destruction of the human spirit over and over again for nothing other than the pride that progressives have in being able to virtue signal their good thinker status.

So that's part of it. There's also the truth, another big part of it. And have others have pointed out here: I love a good fight.

Eric newhill

Shepard, Utter nonsense. The polls are of typically *registered* Ds and Rs and Indies. There is no language in the questions about how they feel they identify at the moment. I myself have been polled recently and the questions were "are you a registered voter?" "Yes." "What party?" "If the election was today, who would you vote for?" This is line with exactly what the polling methodology says it asks.

You're just making stuff up so it fits some marketing approach you think you understand - or you're confusing the polling with the idiotic bathroom/gender issue.


Physician says Hillary has Parkinson's disease:


Make of that what you will. Supposedly written by a board-certified anesthesiologist, but no names given.

Edward Amame


"And have others have pointed out here: I love a good fight." Finally it appears we have something in common.


EA & Tyler

Fight clean lads. Fight clean. pl

Edward Amame


First off, since 2013 China and India have beat Mexico as the most common immigrant countries of origin who've been in the US for a year or less. More Mexican immigrants have returned to Mexico than have migrated to the US since 2009.

If they are here legally, why not? Back in the day, my Irish Catholic and Italian ancestors undercut the cohesiveness of communities, so on that point... As to standards of living, that's tougher. It's been shown that immigration has hit families w/o college diplomas by lowering their incomes from some 7-8%. So that needs to be addressed, though I doubt congress as it is currently made up really gives two s*^$s about that.

The comments to this entry are closed.

My Photo

February 2021

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
Blog powered by Typepad