RealClearPolitics publishes a continuously updated average of all major national presidential polls — as of this evening, their chart looks like this:
(Note that the RCP chart is updated frequently, so that by the time you read this, it may be slightly different.)
But whenever I check the RCP average, including today, I notice something odd: The larger the polling sample size, the more the poll favors Mitt Romney.
I’ve copied the RCP data and pasted it in here in a format that I can re-order. First, here’s how RCP organizes the chart, which they do chronologically, with the most recent poll at the top:
| Poll | Date | Sample | MoE | Romney (R) | Obama (D) | Spread |
|---|---|---|---|---|---|---|
| RCP Average | 10/15 – 10/23 | – | – | 47.8 | 47.2 | Romney +0.6 |
| Rasmussen Reports | 10/21 – 10/23 | 1500 LV | 3.0 | 50 | 46 | Romney +4 |
| ABC News/Wash Post | 10/20 – 10/23 | 1394 LV | 3.0 | 49 | 48 | Romney +1 |
| IBD/TIPP | 10/18 – 10/23 | 938 LV | 3.5 | 44 | 47 | Obama +3 |
| Gallup | 10/17 – 10/23 | 2700 LV | 2.0 | 50 | 47 | Romney +3 |
| Monmouth/SurveyUSA/Braun | 10/18 – 10/21 | 1402 LV | 2.6 | 48 | 45 | Romney +3 |
| CBS News | 10/17 – 10/20 | 790 LV | 4.0 | 46 | 48 | Obama +2 |
| NBC News/Wall St. Jrnl | 10/17 – 10/20 | 816 LV | 3.4 | 47 | 47 | Tie |
| WashTimes/JZ Analytics* | 10/18 – 10/20 | 800 LV | 3.5 | 47 | 50 | Obama +3 |
| Politico/GWU/Battleground | 10/15 – 10/18 | 1000 LV | 3.1 | 49 | 47 | Romney +2 |
But what if we simply re-ordered the polls not chronologically, but according to sample size, with the largest at the top? This is what it would look like:
| Poll | Date | Sample | MoE | Romney (R) | Obama (D) | Spread |
|---|---|---|---|---|---|---|
| RCP Average | 10/15 – 10/23 | – | – | 47.8 | 47.2 | Romney +0.6 |
| Gallup | 10/17 – 10/23 | 2700 LV | 2.0 | 50 | 47 | Romney +3 |
| Rasmussen Reports | 10/21 – 10/23 | 1500 LV | 3.0 | 50 | 46 | Romney +4 |
| Monmouth/SurveyUSA/Braun | 10/18 – 10/21 | 1402 LV | 2.6 | 48 | 45 | Romney +3 |
| ABC News/Wash Post | 10/20 – 10/23 | 1394 LV | 3.0 | 49 | 48 | Romney +1 |
| Politico/GWU/Battleground | 10/15 – 10/18 | 1000 LV | 3.1 | 49 | 47 | Romney +2 |
| IBD/TIPP | 10/18 – 10/23 | 938 LV | 3.5 | 44 | 47 | Obama +3 |
| NBC News/Wall St. Jrnl | 10/17 – 10/20 | 816 LV | 3.4 | 47 | 47 | Tie |
| WashTimes/JZ Analytics* | 10/18 – 10/20 | 800 LV | 3.5 | 47 | 50 | Obama +3 |
| CBS News | 10/17 – 10/20 | 790 LV | 4.0 | 46 | 48 | Obama +2 |
Notice the unmistakable trend?
All polls with 1000 or more respondents favor Romney; all polls with smaller than 1000 respondents favor Obama (or are tied).
Statisticians will tell you that the larger the sample size, the more reliable the poll. This fact is reflected in RCP’s “Margin of Error” (MoE) column, which shows a lower margin of error, and thus a greater level of reliability, for the large-sample pro-Romney polls. Each and every pro-Obama poll has a higher margin of error, and is thus less reliable.
These are the facts as they currently stand, and they’ve been true like this almost every day since soon after the first debate when Romney surged in popularity.
But Why?
The question we must now ask ourselves is why only the weak polls with low response rates favor Obama.
Where are your columns showing the percentage of democrats, republicans and independents? Could the smaller samples also be skewed by oversampling of democrats? Are all the polls sampling likely voters versus registered voters?
The name of each polling company in my chart is a link to its in-house results, most of which show sample cohorts. Or you can click through to the RCP original page to see more details.
I wanted to keep my charts simple and clean merely to show the trend, and not muddy it up with a lot of micro-details.
And yes — the “LV” in the sample column means “Likely Voters.”
Yesterday’s Time Poll for Ohio had the following internals:
Dem 49.3 Rep 29.7 Ind 21.0 giving a result of 49% Obama 45% Romney 6% Undecided
This poll couldn’t be more biased, yet F. Chuck Todd on Morning Joe was touting the poll as Romney is in big trouble in Ohio and if he doesn’t win Ohio he is finished.
This propaganda is over the top. RCP needs to be blamed for not weighting the polling result to get their averages. You cannot include PPP in your results.
I agree.
RCP is loosing what ever credibility it had. It could recover some by giving party ID-breakdowns in the tables.
So with a PLUS TWENTY Democratic sample Obama ONLY has a 5% lead?
Heck that ought to be news right there that it is so low. Amazing how Democrats and their willing accomplices in the media slant everything so blatantly.
I looked at that Time poll too and it was embarrassingly bad. But I don’t think the numbers were as far off as you are saying. If my memory serves me right, it was 28% Republican, 24% Independent, and 42% Democrat. That is an oversampling of 3 points Democrat from 2008 numbers, an under sampling of 2 points off Republican 2008 numbers, and under sampling of 6 points off 2008 of Independents.
What stood out to me and others was that over 400 people were “unemployed” in this poll. There were more “unemployed” than employed people. And Romney won among the “unemployed”! They also reported that Obama “led 60 to 30 among those who already voted.” Yeah – 80 people said they’d already voted and voted for Obama. But of the others? 15 of them said they “didn’t know” who they’d already voted for! And this was published like “gospel.” Gospel of the Church of Everlasting Manure.
That’s believable! I’m unemployed, and I’m voting for Romney! I may be one of the 47%, but that doesn’t mean I WANT to be! In January, I was laid off from a job with a major defense contractor due to the upcoming sequestration! I hold Obama and the Democrats who refuse to stop spending on social programs, preferring to buy votes with taxpayer funds instead of providing for the common defense like the Constitution tells them to!
Mr. Foster,
Thank you for using your workplace skills to help ensure the defense of the nation.
I hope we will soon have a Patriot in office that is worthy of those efforts, an that your livelihood can be justly restored.
Semper Fi
Where did you get those internal numbers from?
Bruce
frank newport, of gallup, had a rather interesting take on chris wallace’s sunday show. gallup does not weight their polls by party. he went so far as to express skepticism for doing so, because the weighting would have to come from an exit poll, which he felt were less than perfect.
he didn,t specify thier precise sample determination, but he did seem to take pride in determining who was a likely voter. most polling firms find 80% of respondents say they will vote. 25% of that group is lying.
Gallup is right to avoid that, as people without a strong party identification sometimes align their “party preference” to the candidate they are currently favoring. So, weighting by party is to some degree weighting by preference — which means you are distorting the poll.
So, if 55% are for Romney, this might cause the “R” preference to be, say, 4 points higher than “normal” as some of the independents supporting Romney claim to be Republicans for the moment. So you then “weight” the poll and say that Romney has only 51%. Not good.
I think it also matters how the names are picked, what time of day they pool is done, what part of the Country is polled, if it is a short one or a long one gathering demographics. While it is true that some polls do not weight to get a balance of D’s and R’s some are weighting to the 2008 D vs R mix.
The Gallup website explains how they try to determine a likely voter.
They ask each voter a set of questions on their past voting history (how often they voted, how enthusiastic they were, etc.), and on their current views: Are they really paying a lot of attention to this campaign? Do they consider themselves politically aware and maybe even politically active? Etc.
IOW, they try to deal with the psychology of the voter, not his political leanings. A voter who thinks that this election is absolutely critical for his own family’s future is more likely to vote than a voter who thinks that it really doesn’t matter that much who wins.
BTW, I saw newport on FNS and I called BS. I think if party ID bounces around that much it has to do with selection methodology. That to me is the unanswered question. As some one that did surveys of IT people in the 90s. I can tall you it is useless. People lie all the time. There is a lot of psychology going on when you get a random phone call. It begins with caller id and pre-begins with how you get and select the number.
Lies, damn lies and statistics.
The whole thing is bogus as far as I am concerned.
weighting by party is an infinitely flexible cheat stick.
Poll are nothing more than mechanisms to manipulate the electorate. That hold true for conservative as well a liberal pollsters, but since pravda controls the information stream, the mainstream is 90% dominated by the liberal manipulators. Pravda learned years ago that the public doesn’t believe their pretend “news”, so they turned to pretend poll as their preferred means of manipulation.
I know that both Gallup and Rasmussen use a rolling average. Gallup uses a full week while Rasmussen uses 3 days. As I understand it that means they poll that number of people every day, so the actual sample is probably several times bigger than the stated number. I’m not sure of Politico but I suspect that nice round number may be a clue that it is also a rolling average. Some of the others are a single poll that was taken over multiple days to allow for call backs of those who didn’t answer the phone the first time. That makes comparisons even harder.
If a poll is kept open too long it is useless. In the last week I saw a sample size of under 500 from a state poll of a US Senate race that was taken over ten days. In the last month of the campaign with all the debates, public appearances and new ads in the news every day it is hard to see how that particular poll had much validity.
The RCP average gives equal weight to all polls. If you weight the poll results by sample size, you get:
Weighted average: … Romney 48.3 … Obama 47.0
Giving a 1.3% difference instead of the 0.6% difference shown.
I was going to point that out, but forget to note it — thanks for doing it for me!
When you account for the 5 day Gallup sample and the 3 day Rassmusen sample, it is Romney +2.55.
Exactly Right!
You have to use a weighted average of all polls, maybe dropping the tie, as I did. All things being equal (they’re not, alas), you get a much more realistic result than the one RCP stated. Mine came to a 1.5% Romney lead, discarding the tie.
No matter what, all of this polling stuff seems to me to be more akin to 16th century attempts at finding the “philosopher’s stone”, or even earlier attempts by Bishop Usher to divine how many angels could “dance on the head of a pin”. — Answer: None, or One, or All. Take your pick. Enjoy.
does this hold true for state polls in ohio – larger sample sizes, better numbers for romney? or is it true only in national polls?
Another factor is how the pollster got the phone numbers. Sometimes pollsters pay the states for lists of the numbers of registered voters, feed those numbers into a computer and as for a randomly generated list of XXXX numbers to call. Others use a random dialing program and trust that the person who answers is telling the truth when the say they are registered to vote. Since many people who aren’t registered will tell pollsters that they are registered that can be a source or error.
On a likely voter poll there is also the issue of the tightness of the screen. Again many people do say they are likely to vote when they are not. The loosest screen is to merely ask Are you likely to vote? The tighter the screen the more specific the questions with the person on the recording end being instructed to not consider someone a likely voter if they can’t answer or seem hesitant on a question like Who did you vote for in the last election? or Can you tell me where is your polling place is?
On a poll of voters in a particular state both the starting place and the tightness of the screen are huge factors in assessing reliability.
Where did you learn about polling information, types of polls, screen tightness, etc….B.U. S.of Comm.? Nice work
re: NC Girl…..here comes the Sun..
I would like to see at what time of day all polling was conducted, and what percentage of the sample were cell phones as opposed to land lines.
If you called land-line home phones from 9-5 am, that would skew the sample.
Two possible answers come to my mind. One is dem vs GOP sampling. If the pro Obama polls all have a higher Dem vs GOP ratio, that would explain it.
Another possibility is to look at who has the best polls for Obama. Ranging from Mitt +1 to Obama +3, all of them are MSM. If a conservative is called by an MSM, are they more or less likely to hang up than a liberal? Yes, some of the MSM are more toward the center than the left, but they are still MSM.
These two aren’t mutually exclusive with MSM identified polls getting a lopsided dem vs GOP ratio because GOP voters are hanging up on the MSM.
I suspect that with declining participation rates, lower trust in MSM and what is shaping up to be a possible embarrassment for some of the polling firms, we are likely to see a “Dewey defeats Truman” re-evaluation of polling occur after the election. Part of why the Chicago Tribune printed their headline was they had polling that showed Dewey was up on Truman. The problem was the polls were only contacting people with private phones (who did support Dewey over Truman). Those with party line phones or no phones supported Truman and the rest is history. How pollsters, especially pollsters tied to MSM outlets, are going to reach conservatives turned off by MSM liberal bias may turn out to be a difficult question to answer.
The size of the sample in this case has more to do with financial constraints than anything else. Sampling is expensive. Now, without examining each and every poll, the margins of error are actually built into the study. The 3-4% you hear about frequently is due to the fact that most samples have around 1,000 people. (This number gives a good idea of who will vote for whom and is affordable.) In any reputable sample, they will tell you their allowed margin of error and the sampling mix that they use….Which is where I think the problem is. (Note. I deal with sampling data daily in my job, but I have not closely examined these polls, the questions they’ve asked, sampling frame etc.) The single largest problem, and this is highly subjective, is the mix they use in their sample: democrats vs. republicans vs. independents. To gauge this correctly, you have to use some “mathemagic.” No one knows exactly who will turn up on election day, though samplers are asked to forecast based on several factors who will turn out to vote and who will stay home. You can automatically discount any poll that states that Obama the incumbent will draw a higher mix of democrats in 2012 than Obama of the church of hope and change of 2008. It isn’t going to happen. You know how I know? Check favourability ratings. Check polls asking if people think the country is going in the right direction. Check consumer confidence levels. Democrats, many independents, and a number of republicans were energized about an Obama administration in 2008. They are not now. It will almost never happen that an incumbent energizes his/her supporters more after several years in office. It certainly won’t happen this time. Obama is a polarizing figure, even more so than GWB. Republicans were ready to vote against Obama 6 weeks ago. After the first debate, suddenly the idea of voting for “President Romney” seemed real. And people began admitting what they had felt for a while….The way to counter this massive shift in independents/women etc? You would only have to include more democrats in the mix. Any poll that estimates more Democrats than in 2008, or more independents than Republicans can automatically be tossed out. My company would laugh me out of the board room if I based a business decision on such a lousy forecast.
“It will almost never happen that an incumbent energizes his/her supporters more after several years in office.”
In other words, Obama is _not_ Reagan.
Exactly. Hell, Obama is not even George W. Bush. Many voters are suffering massive Obama fatigue; more interestingly, a lot of ardent supporters from 2008 have a serious case of post purchase cognitive dissonance. What happens when you buy something and the real benefits do not equal the advertised benefits? This is certainly the case for Obama. Loading a sample with more democrats than in 2008? Not excusable given current favourability ratings.
I am not sure how reliable a trend you have found, given the small number of polls. But another apparent correlation in your data is that with 2 exceptions, the polls favoring Obama all finished earlier than those favoring Romney. So some of what you are seeing could be voters trending towards Romney over time.
Funny, just a few hours ago, I noticed the same thing you note.
1. Smaller sample sizes do not necessarily mean lower response rates, as you seem to assume. It depends on the numbers they sought to get.
2. Smaller samples should be more variable in BOTH directions, not just in one direction.
3. I wonder whether there is a correlation between sample sizes and the use of cell phones. Maybe organizations that do more cell phone calling do more calling overall. I have no idea.
4. Maybe the larger samples are because the pollers spend less $$ on each call, such as talking to the person who answers the phone, rather than choosing the adult in the household randomly (such as asking for the person with the latest birthday). If so, that would make the smaller polls MORE accurate.
Without looking at the polling methods, it is hard to guess the reason for the pattern you nicely identify.
Theory: The polls done on the cheap (making fewer calls) use a weaker likely voter screen to inflate the apparent size of their sample. This gives them better numbers for Obama, who consistently polls higher among registered voters than among likely voters.
This doesn’t presuppose malice — just greed and/or laziness.
Excellent theory. You may be 100% correct.
The IBD/TIPP poll just released showing Obama with a healthy lead screened 993 registered voters and found 923 likely voters. For that to be accurate, we would need a 93% turnout.
It also came up with a D+7 likely voter sample, which would mean this year will be similar to 2008 on party ID.
The computer guys had an old term for it: Garbage In, Garbage Out.
The size of sample does heavily affect the margin of error, but it is important to know what the margin of error actually means. It means that IF THE SAMPLE IS RANDOMLY DRAWN FROM THE SUBJECT UNIVERSE, that the results lie within the range of the margin of error 95% of the time (or 99% or whatever the “confidence interval” is for the study). It is not a measure of the chances that the sample is in fact random. A biased sample is just plain ole biased and its results are invalid for any kind of sampling statistics. What the margin of error is important for is knowing that a CORRECTLY DRAWN RANDOM SAMPLE gives the incorrect result on occasion. Given the very large number of political polls being taken, EVEN IF THEY WERE ALL TRULY RANDOM, several each week will give the wrong result.
The entire practice of weighting samples by criteria like political party is very suspect. It says that if we mix up results obtained non randomly in the “right” way we will get a random-like result that we can describe with statistics. Of course, this means the bias of the researcher outweighs all else. Properly drawn, a random sample will find the right mix of voters itself — randomly.
True, but with a response rate which slipped from over 70% in the ’80s to 37% by 1997 and only 9% today (according to Pew Research), the probability that a properly drawn random sample of respondents will share any given characteristic with the 90% of the population which declines to participate also goes down.
Unfortunately, the only ways to ascertain that would be to survey those who refuse (and isn’t that a problem!) or exit polls after the fact, which have their own sampling problems with response.
Well, when you only ask two Democrats and a Republican, you can handily go: Obama, 67%; Romney, 33%!
Brilliant explanation! Occam’s razor.
Another cheap theory: if you’re going to publish a biased poll, why spend the money on a big sample size?
Very good point. Since the poll is a statistical Potempkin Village, no need to construct actual buildings — just façades. And no need to go to a lot of effort for a decent sample size — just an offhand gesture at polling is sufficient.
Very interesting observation. Please note that the RCP average is a straight average and does not take into account sample size. Weighting the average for sample size (in a sense treating these disparate samples as unitary sample) I calculate Romney 48.34% vs. Obama 47.02% for Romney +1.32%. FYI.
I wonder what the Margin of Error would be for a poll with a 11,340 Likely Voter sample size (which is what this mega-combo poll “unitary sample” works out to). I imagine that it is very very very low. That means basically that Romney is up 1.32% in a poll that is as close to accurate as we can get, and with a margin of error that’s probably much less than 1.32%.
The MOE for a sample of 11,340 voters would be +/- .92%, so the lead of 1.32% is not outside the margin of error.
That’s not how it works–but you end up being right for the wrong reasons.
Mathematically, you can’t just add up all the voters in all polls as if it were a single large poll—the Margin of Error of the sum won’t be the same. (The standard deviation of a sum is not equal to the sum of the standard deviations, because of the square root involved.)
The Central Limit Theorem states that the various poll results should cluster around to form a normal distribution (bell shaped curve) around the “true” estimate–the choices of the entire voting population. IOW, the average of all these polls will be closer to the true average than any individual poll.
Further, the standard deviation of this distribution (which determines the Margin of Error) will be inversely proportional to the square root of the sample sizes.
Since these samples in these polls are at least several hundred respondents each, the standard deviation will be extremely small. That’s why the Real Clear Politics average of all these polls would be a highly accurate estimate of voter preferences–if it weren’t for the fact that all those polls are taken on different days.
I can only help my daughters with up to 8th grade math. This talk about standard deviations and budget deficits is beyond me…
If you perform a Chi squared check on the polls you find 3.2 for the Romney numbers and 1.49 for the Obama numbers (should be 1 for both). This says nothing about whether Obama or Romney will win (Romney’s larger number could be from polls being over or under). What it does say is that the polls are very unlikely to have been taken randomly from the same ensemble.
As Justin pointed out, the MOE just says 5% of results are going to be outside that range if the sample is correctly drawn. If we start with ClintACK’s theory that the ones with the lower numbers are also being done on the cheap, we also have to assume then that a good portion of their results is infested with garbage. That doesn’t just increase the MOE, it can shift the mean, and MOE just doesn’t account for that. If the sample mean is consistently off from the population mean, the sampling process is inherently broken and no amount of averaging can correct it, since the same error should be in every such poll. Even weighting the average of all the polls to reduce the weights of the suspect ones, the source of error is merely reduced, not canceled out like a true series of random samples would tend to do.
If the smaller polls are cheaper and inherently broken as a result, the only viable option is to discard them. The downside is that there are fewer data points to work with after that, and of course there’s no guarantee a high LV count doesn’t correspond to a weak screen. I think Clint’s right, but figuring out where to draw the line between good polls and bad polls without knowing the ins and outs of each one’s methodology is a sure recipe for confirmation bias.
I saw a comment elsewhere saying that some polls throw all early voters into the likely voter sample. They apply the same likely voter screen to everyone else.
This inflates the democratic proportion of the likely voter sample, because more democrats vote early than republicans.
The theory makes sense to me. Perhaps some more sophisticated polls account for the early voters differently, or adjusts the screen of all the others to account for the Democrat early voting skew.
One thing I don’t know is why Democrats are more likely to vote early.
“One thing I don’t know is why Democrats are more likely to vote early.”
I have seen that before. When I asked for evidence, the poster folded his tent and slunk off into the night. Frankly, when faced with ANY political certainty I ask for evidence. In politics, the chance of the sun rising in the east tomorrow is 98% +/- 4% MOE 3.
A compromise between water and sewage is still sewage
Is it possible that public sector unions (and maybe private sector unions as well) are asking their members to vote early, so that the Democrats can use that as a talking point?
It would seem that a little shoe leather journalism could answer that question.
BTW, if they aren’t doing that, then what are we paying them so much of our tax dollars for?
The early voting center is typically found in the urban center. The urban center is generally populated more heavily by Democrats. Republicans in the suburbs are not going to drive to the early voting center when they can go to a nearer voting booth on election day.
Also early voting is the preferred method for Democrats who are dead or nonexistant but who still vote.
I am inclined to discount the trend you’re seeing here related to sample size. As earlier comments have pointed out, the Republican-Democrat split in the sample is overwhelmingly correlated with the margin between the two candidates.
The other point I’d make is that it’s relatively useless to compare, say, IBD vs. Gallup vs. Rasmussen. What you want to compare is e.g. Rasmussen today vs. Rasmussen a week ago vs. Rasmussen two weeks ago and pay attention to the trends within the one poll.
If I could get deep access to some of these tracking polls, I’d actually tend to ignore partisan affiliation at first, and I wouldn’t care which particular poll baseline I am starting from (i.e. Rasmussen vs. IBD vs. Gallup). Instead, I’d take any of these polls’ data and start with a sample of about 529 people who said they voted for Obama in 2008, as well as 457 McCain voters, and I’d try to line these numbers up as closely as possible to match the 2008 D-R-I partisan split. This would indicate for the portion of the 2012 electorate that is representative of who showed up to vote in the 2008 election, how much change is occurring from that baseline. For example, I could estimate fairly accurately how many are switching from Obama to Romney, or from McCain to Obama. I could also answer the question, what percentage of 2008 Obama or McCain voters are likely to sit out this election?
I would call this group of 2008 voters “Subset 1″, and removing the 2008 voters from this group who say they are sitting out the 2012 election, I’d label this altered sample “Subset 2″. Then I could compare these numbers against the overall poll’s results for Likely Voters (“the superset”) for Democrats, Independents, and Republicans and draw some conclusions about the demographics of the 2012 electorate that didn’t vote in 2008 but say they plan to vote this time around — I’d call this group “Subset 3″.
Why would I manipulate the poll data in this manner? It is my belief that Subset 2 would be the best basis for “horse race” tracking, that is, trying to track and predict the momentum of each candidate at any given point in time. I would also want to track Subset 3 over time, but for a different reason, which is to gather a large sample of data and trends with which to predict who is going to show up in 2012 that didn’t vote in 2008. IOW, I’d use the data trends from tracking Subset 3 along with the data trends from tracking Subset 2 to then project the final composition of the electorate (i.e. the turnout), and who they’d be voting for (i.e. the final vote projection). IMO, slicing and then recombining the data in this fashion would lead to more accurate poll numbers than any of the methodologies used by these individual tracking polls (either on their own or aggregated a la RealClearPolitics or 528/Nate Silver).
This just in: a new PPP poll has provided good news for Team Obama. While Mitt Romney still leads overall nationally, the President is now +4 on Mars, +5 on Mercury, a whopping +12 throughout the Jovian moon system and is rapidly pulling even on Neptune and Uranus. When asked for comment, David Axelrod said, “This is very encouraging news–I see a pathway to victory! Uh, by the way, how many votes DOES the solar system have in the Electoral College?”
This just in: in the critical swing states a new poll shows Obama leading by a whopping 100% among dead voters, 75% among multiple voters, 90% among voters convicted of a felony, 100% among those voting under the names of various disney and sesame street characters, and 88% among illegal immigrant voters. When asked to speculate on why these groups were voting, top Obama campaign officials said: “It’s ok, everyone does it. In fact it’s hilarious. Vote early and often, see your local OFA office for help filling out the paperwork. And if anyone tries to ask you for ID be sure to call them a racist.”
I knew pi$$ing off Big Bird would cause problems.
Obama ahead on Uranus? On his, too, or more precisely inserted within.
Funny, I noticed the same thing earlier today when I was looking at the RCP polls.
Has anyone checked out the RCP electoral college map? Oh yeah, Romney is ahead in 7 out of 10 toss up states. We have this in the bag so there’s no need to worry. Yay!
The cockiness…do not acquire it yet.
/Yoda Solo
And yet the RCP ‘no state toss up’ has Obama 281 over Romney 257.
Fraud all over. Voting machines in N.Carolina pre-programmed with Obama results.
Obama in last debate forgoing his “crutch” (teleprompter) for an earphone: yes, check it out (you decide)
http://ppsimmons.blogspot.com.br/2012/10/was-obama-wearing-earphone-during.html
see the eraphone is in his left ear because he’s left-handed.
Nov 6 Election Voter Tabulations will all be processed by a foreign company (SKYTL) in a foreign land (Barcelona, Spain) organized by a group of investors (one of whom is the fabled “Puppet Master” – George Soros).
And there are those Americans who insist voter fraud doesn’t exist. Just is incredible! Wake-up America. Pray. Amen.
Yes indeed…Fraud is the bigger issue and one reason for a lot of ‘early’ Democratic votes. Easier to fake a mail-in than having somebody show up. Obama has had 4 years to develop dead drops, phoney adderesses and multiple voters. But, I am sure the Justice Dept will be on the lookout for them. Not!
larger polls tend to have the ability to self correct…
take ibd/tipp with obama +3, and a small sample.
if you run their numbers using the religious demo against cnn exit poll of 2008…
obama’s down 6.05 net pts among protestants,
down 2.7 net pts from catholics,
unchanged among jews(.02%),
down 3.24pts among ‘other’,
and down 1.2 for ‘none’.
this breakdown would suggest a 13.19 pt net swing to romney from the 2008 result.
so how good is their sample with this much variability about what they are reporting to measure v. what has been measured accurately?
not very. they painted three sides of the house, and hoped no one notices that they didn’t finish the last.
The time has come to forget the polls. Study them if it comforts you. Nov 6 will have the largest possible “sample”, depending on turnout, numbering in the tens of dozens of millions.
Also, no matter the sample sizes of polls, they all show a large enthusiasm gap in favour of Republicans. Time to pour all your spare millions of dollars into GOTV.
As said by others, the guy who pays the bill tends to want his biases confirmed.
I’ve been searching for — and cannot find — the quote by a senior editor at Time or Newsweek to the effect that media support is worth 6% or 7% on election day. Think it was in the 1980′s, which means pre-internet. Anyone know who it was?
>> the quote by a senior editor at Time or Newsweek to the effect that media support is worth 6% or 7% on election day <<
IIRC, it was Evan Thomas a few years ago on the "Inside Washington" PBS TV show.
Thank you. You may be right, though Evans could be a bit young (b.1951). Will try to verify.
Er…Evan Thomas
Sounds about right. If the media treated this administration with even a fraction of the rigor they treated Bush, Obama would be down 10% in the poles right now instead of 1-4%. I’d guess that most Americans don’t know about most of this administration’s scandals. Media portrayal makes a HUGE difference to the uninformed voter who votes on the basis of their emotion or “gut”. Imagine if the papers had given Fast and Furious front page coverage with pictures of murdered Mexican children and demanded answers when the administration invoked “executive privilege”? That’s what they would’ve done to any republican administration. Instead they tried to portray the whole thing as a “partisan witch hunt” by house Republicans. These things matter.
I saw another article with the headline “Republican house easily defeats Obama budget” absolutely no mention that not one democrat voted on it either. Think about that, not one member of congress voting on the president’s budget is huge news, but they deliberately try and mislead the uninformed into blaming the Republicans.
Medical malpractice can kill patients, enough journalist malpractice can doom nations.
enough journalist malpractice can doom nations.
Yes. Agree with the rest of what you say, too. Cleaning up our corrupt media belongs very high on the To Do list, regardless of who wins the election. No walks, they’ve done too much damage.
When it comes to polling I always think of the people who have the dilemma that their spouse will vote one way and they know they will vote the other way but don’t want it to come out. In normal times that sort of thing evens out, but we have an emotionally and racially charged election here where some people who intend to vote for Romney are playing possum. A lot more, I suspect, than the reverse, primarily because they don’t want to appear racist. Let’s call it the Chris Matthews effect – even thinking about voting against Obama is racist. It may be negligible or it may not. but I think it could be distorting the polls significantly.
Then there are the Jacksonians. People think that Jacksonians are mainly limited to the crescent that runs from West Virginia to Oklahoma and they would be right to a point. Remember those coal miners who voted for anyone but Obama in the Democratic primaries including a guy in jail in Texas? I think these guys are going to vote Republican in droves this election, much like Nixon won over a chunk of the Joe Sixpack working class back in the day. It wont make any difference in most of the states in my Jacksonian crescent, but it will make a difference in Ohio and Pennsylvania. And Jacksonian are everywhere and they never vote Republican until they do. And I’ll tell you something else, a lot of Blacks are Jacksonian Democrats and and some of them are as mad as hell at the President and some just disappointed. My congressman Alan West of FL22 routinely gives Obama heaps as only a Black man can. My absentee ballot got mailed off the day before yesterday and I voted with conviction for West and Romney in that order. My head is Jeffersonian like the President – the part I refer to as my inner Tom Friedman who believes so many things that just aren’t true. But my heart grew up in the back woods of New Hampshire and is Jacksonian. So when I saw the Democrats sending General Petraeus off to lose in Iraq my heart went cold and I vowed I would never vote for a Democrat again. Jacksonians are like that and if you don’t understand that, its not my problem.
Zombie scoop: “Romney Wins.”
This year, it’s not ‘Dewey Beats Truman.’ The results are in through the polls, as a result of the lopsided debate outcomes, from the skewing towards Dems and the attendance at the various rallies. Romney wins.
I see a 52-45 or 53-44 outcome, a landslide and an end to the destructive policies of the Wahn.
‘They all look winable to me’ Ronald Reagan in 1980 to one of his staffers before election day. Predictions were a 32 state win – it was 44. I’ll go out on the limb and say that some of the solid blue states might actually go for Romney. I’d say no more likely that a 1 in 10 chance, but still plausible, unliekly, but possible. btw, 1 in 10 is better then the chance of drawing 1 type of card out of the deck (1 in 13).
These polls from CBS, ABC and the like all seem to mirror each other; the thought occurred to me that they aren’t really polling but just recycling an old poll. Who would know?
All the polls are “Weighed” (rigged). The factors used to ‘weigh/rig’ the polls should be adjusted by sample size. Weighing for a 2000 sample should NOT be the same as weighing for a 1000 sample. The smaller size tends to magnify the weight used.
I have never done polling work, but real Statistics such as used to design aircraft, hydraulic flow systems, Stress factors, etc. Use various sorts of formulae to account for that.
OIL rushing through a pipeline at 70 PSI does not behave the same as OIL rushing through a pipeline at 85 PSI. Air flow over a wing at 120 knots is not the same as airflow over a wing at 200 knots.
“Oh, people can come up with statistics to prove anything, Kent. Forty percent of all people know that.” — Homer Simpson.
Why? The same reason some are using D+10 and higher polling samples? I hope you’re smart enough that the “why?” part of this was rhetorical.
It’s really all in how you get to “Likely Voter” from “Registered Voter” (or “Dude answering the phone”). If your data model is built around the 2008 election then Obama automatically wins. In 2008, Obama won the popular vote 52.9% to 45.7%, you’ll note this result falls within the MOE for the various “Obama WINS!” polls. Once you know the answer it’s pretty simple to get the correct result.
My guess is that Obama will get the same 42-43% of the popular vote that Clinton captured and go home wondering “how did I lose to that guy”.
Interesting. Besides the weighted average, which a couple of commenters have noted, I think RCP can be faulted for using averages — actually, “means” — at all. With the small number of polls, and especially because there are often clear outliers, they would do better posting the median, rather than the mean. It doesn’t make much difference in the example given in this case — the median is +1 for Romney — because there isn’t a strong outlier, but I’ve seen it make a notable difference in other cases.
But, to be really clever about it, you’d do a weighted median, and in this case we get Romney +2.9.
Opps, +2.1, not +2.9.
It would be interesting to plug the multiple poll data into SAS and try to do a meta-analysis, which is the right way to combine different studies into a single result (although it’s very easy to do it badly). Simple averaging of sample means is really sloppy and would only give a true overall mean by coincidence. But to do it right, you’d have to spend a lot of time looking at the construction of the likely voter screens.
The more I look at these polls, the more I am convinced that non-response bias is the real gorilla in this room. I work with health survey data, where non-response is a huge problem. Health surveys are very intrusive, and people are reluctant to reveal such intimate information. Yet we get far better response rates than 9%! And just like health data, you almost have to assume that those who refuse to respond differ in some systematic way from those who participate, and probably on those very factors that are correlated with your outcome variables of interest.
The problem is, there really is no good way to analyze non-response bias unless you can get non-respondents to…agree to be surveyed! If the pollsters aren’t sweating poop nuggets over this they should be, because they’re all going to look pretty foolish on the morning of November 7.
Ha!
I do research in healthcare IT. I NEVER trust polls or blind surveys. KLAS is the worse. They ask people if they are happy with their IT selection. Who is going to admit they made a mistake. Smaller vendors that get knocks look at the comment section and can detect the comments are NOT from actual users. The whole thing is bogus.
One, example: go and try to find an actual practicing doctor (primary care) that is happy with EHRs. NONE. But the official stats from government surveys claim most docs are happy with their EHRs. I can make a survey to say anything I want. Not hard.
If our family habits are any indication, non-respondents make extensive use of caller id, and if any unknown numbers slip past that net, we promptly hang up when we hear the beeeeep! Maybe we’re just a grouchy bunch or maybe our voting habits are nobody else’s business!
My father had a saying about religion: Northern protestants like to keep their religion to themselves, Southern baptists never want to shut up about it. Seems maybe for some people talking about politics is something they’d keep to themselves, but get them started on their medical concerns and they’ll talk about their bunyons til your ears fall off. I’d guess that this may also correspond to conservative vs. liberal likelyhood of response.
Why do you believe editorially conservative outlets are immune to leftist bias on the “news” pages? The WSJ is notorious for the “reporting” being conventional left pablum; the same could be true for the Washington Times and IBD. They’re still staffed by “journalists” — which is to say, corrupt know-nothings with the independence of prairie dogs.
W.r.t. your theory #3: the reason “all” (actually 3 out of four) of the smaller polls favor Obama may be random. If we assume equal likelihood of error to ‘favor’ Obama’ and ‘favor Romney’, the likelihood of getting 3 of four going one way is 3/8.
More than that … Sort by margin of error, I think, eyeballing it, then that all the Romney results are up top, followed by the tie, followed by the Obama results.
Zombie, A few of your earlier commenters have hit upon some of these highlights, but what you have noticed matches a similar trend I wrote about yesterday: polls with tighter voter screens tend to be more Republican. I referenced your post in an update to mine along with this comment:
Good polling is very expensive. To get a sample that is representative of the voting population means that you have to make many times more calls than you need actual respondents. The best polling organizations ask about your voting history and from that they determine your likelihood of voting. If you voted in the primary and the mid-term election in 2010, you are more likely to vote again this year. If you didn’t vote then and you don’t know the location of your polling precinct, you are unlikely to vote now. Polls conducted on the cheap ask people to self-identify whether or not they are likely to vote and take people at their word. Culling those unlikely to vote means turning away a potential respondent that you finally got on the phone and who is willing to answer your questions. It is a lot easier and cheaper to use a looser likely voting screen and a smaller sample size than it is to do polling right.
(This, BTW, explains why most polling organizations don’t change to even their porous version of a likely voter screen until after Labor Day.)
http://www.bobkrumm.com/blog/?p=2449
None of these polls are worth anything. With just a 9% response rate, the pollsters simply have no idea who’s going to vote for whom. As Iowahawk has been telling us for weeks: these polls tell you nothing more than the voting patterns of the vanishingly small number of people who agree to be polled.
We might as well be looking at egret tracks in the sand, and arguing over what they portend for the future.
The party and candidates are calling everybody. That is at least my sense here in FL. They sound like a poll but are gathering GOTV lists. See below.
That 9% figure bothers me, too. What are the characteristics of the 9% who can be reached by land-line and/or cell phones AND agree to be polled? How do they differ, if they do, from the other 91% ? Some respondents to this article argue that the many polls can somehow be averaged to get the benefits of a very large sample size. But, if all the polls are sampling the same sub-population of the 9% who can and will respond, how does that averaging help? Imagine a poll that contacts people and wants to ask them intimate questions about their sexual practices. Would anyone place any faith in the responses of the very small percent who would answer those questions, even if the questions were answered truthfully? Would it help to conduct many such polls and average the results?
I commented on this the other day. It appears to me that NO poll is really valid, because with only a 9% response rate, the “experiemental units,” are self-selected and therefore some sort of bias must be assumed. In real research, experiemental units are completely randomly selected out of a larger population to eliminate selection bias.
Zombie,
One more thought.
There are at least three different purposes for polling.
I live in FL and have gotten called several times a week for months but the volume has dropped of significantly.
One poll, which is sample sized based, are those trying to figure out how will I vote for statistical reasons, they sometimes ask a lot of demographic info. I must have gotten a hundred of these.
A second poll is a push-poll, which is a call that sounds like a poll but is just direct advertising (I don’t think I got a single one this seasons).
The third poll is a poll of all registered GOP/Dums by the parties. They are trying to figure out if I am going to vote and which way. They are a little hard to figure out but tend to be person to person (not robo) and have only two questions (who are you voting for and are your likely to vote). I got a handful of these.
The polling is way, way off in FL. I think the statistical types have given up (FL is in the bag for Romney). The push-pollers may be holding fire for now. The party polling is also way down. They know how likely I am (100%) and who I will vote for.
Voting opens on Saturday. And I will be there at 8AM. I am expecting the gotv polling to start on Monday. I am going to say I did not vote for a few days to see what happens. I expect as soon as they figure out I did vote, or I tell them I did, the polling will be, thankfully, over.
It seems I am one of a dying breed. I have a landline phone with a listed number.
I get polling calls almost daily, sometimes several a night.
I always ask the name of the polling company and google them after the survey. The results are illustrative.
Many are push polls. “If you learned candidate A ate babies, would it change your opinion?” I go out of my way to skew such polls with contradictory answers. The real polls I treat much differently and try to give honest answers.
Why they spend this much money polling me, in Calif of all places, is beyond me. There is simply no way Calif will go any other way than deep blue. The population centers along the coast skew hard left and the remainder of the state with 90% of the land mass are voiceless.
I have noticed that the pollsters are now screening for likely voters only. They are now going out of their way to ask about prior voting, how often and for whom. Oddly, even though there are huge issues on the ballot here in Calif, only a few have even asked about them.
In closing, I offer the following advise to polling companies calling me;
If you want someone to take your poll, use someone who speaks American English. Using a caller that sounds like they just stepped out of the “hood” is a sure way for me to decline. Ditto if they sound like they are calling from India.
Try to brief your survey takers to know how to pronounce the candidates name. I had one a few days ago that managed 3 different pronunciations of the same name. I had to correct her each time.
At the first sign it is a push poll, I hang up. At least give the modest illusion that its a legitimate poll.
Stop calling me during Monday Night Football. My responses are usually more profane and ittitable when I an interrupted during the game.
Cheers,
knewshound
I’ll go with #2: pollster bias. You wrote, “But a flaw in this theory is that two of the four pro-Obama polls were published by conservative-leaning media — IBD and Washington Times — who presumably have no interest in helping Obama.” Presumably, the conservative-leaning media engaged a polling firm and did not conduct the poll themselves. The polling firms are not bound to the ideological leanings of their client and might be/probably are using their conservative-leaning clients as cover.
From a Democratic Party perspective, if the President is behind and Romney is trending inexorably ahead, it is essential the polls not display the true state of affairs. I read recently the Democrats have an army of attorneys waiting in the wings to contest this election. Polls showing the President within the margin of error in the few weeks leading up to the election will bolster that effort. Polls indicating a Reagan/Mondale-like blowout, on the other hand, simply will not do. To successfully pull off election fraud, one must make the outcome seem plausible.
Ask yourselves: based upon all we know about the Chicago Way, do these people seem like the types to give up power on the basis of a free and fair election?
My non-scientific rule-of-thumb is to subtract 10 points from the Democrat (on all polls) to get the true result.
Seems to work pretty well.
What is the purpose for polling anyway? Seems to me it’s to encourage people to think that their candidate is a lost cause and stay home rather than vote. Better take polls with a grain of salt anyway. Living in a “battle ground” state (Colorado), I’ve been polled 5 times in the past two weeks. It’s gotten so tiresome that I have begun to lie on all of them just to throw my personal monkey wrench into the numbers.
What if you average the data points rather than just average the polls? That is a much more statistically correct way to approach a meta-analysis.
Or simply weigh the polls to reflect the amount of data they contain.
As Romney grows in “plausibility” and “acceptability” the likelihood that a Romney voter will respond increases. With the media bashing now countered by Romney’s debate performance, it is now more acceptable to admit support for Romney.
seems obvious that many people, esp men, are refusing to tell anon pollster when asked that they are voting for Romney. Perhaps they [wisely] don’t want to start a fight with their wife. Perhaps they are afraid they’ll be investigated like Joe the Plumber was.
Interestingly, the gender gap in most polls has disappeared. Half of women are voting Romney and the other half are f$$&ing idiots.
The same groups who are cheap and pay for a smaller poll size are the ones that use outdated voter breakdown models, such as 2008, thus giving more edge to Obama. The pollsters aren’t to blame, the groups that pay them for half-hearted effort are to blame.
Just had to say: Wow, a whole lot of rationale, polite conversation here. What gives?
Oops, not trying to imply people on this site aren’t generally polite. I just rarely see that on any political response threads. Kudos!
Election fatigue. It has been a long grind. Sometimes I wonder if bullets instead of ballots doesn’t have an upside. With bullets the really obnoxious won’t make it to the next election. Then I remember how obnoxious liberals find me and appreciate ballots.
Haji can’t shoot.
After a quick and very dirty run at the numbers, it appears that RCP does not weight individual polls to account for sample-size differences. If so, the RCP average is wrong. Again, after a VERY quick and dirty attempt to weight the data, it appears the RCP average should be about +1.9 for Romney. Please correct me if I’m wrong here.
Rather than sample size, I see a correlation simply between the weighting of party affiliation and Obama lead. I have some charts posted here:
http://nomoon.org/forum/showthread.php?tid=15&pid=54#pid54
I am a retired teacher. When questioned by a supposed authority figure, a substantial but unknown percentage of people will either lie and say the “right answer” or lie just because they’re feeling naughty and want to stick it to The Man. In the case of political polls, the “right answer” is “Yes, I’m a likely voter” and “Yes, I support Obama.” Some of the options for naughty answers include giving contradictory answers, claiming to be undecided, saying the opposite of how one actually intends to vote. All of these options are handy ways to get revenge for having dinner interrupted by an idiotic phone call. For example, a friend of mine always answers polling phone calls. He tells the call center employee, “I’ll answer all your questions but I’ll lie every time.” The poll proceeds just as if he hadn’t disqualified himself! It takes a very large sample to compensate for our natural human cussedness.
They could use portable lie detectors;
http://www.pimall.com/nais/e.pse.html
If they know you are lying, they can just reverse your answer. I grunt when on the phone to somebody I don’t want to hang up on. Let ‘em try and figure out if that was an honest grunt or a lying grunt.
“Oh, people can come up with statistics to prove anything, Kent. Forty percent of all people know that.” — Homer Simpson.
After a quick and very dirty run at the numbers, it appears that RCP does not weight individual polls to account for sample-size differences. If so, the RCP average is wrong. Again, after a VERY quick and dirty attempt to weight the data, it appears the RCP average should be about +1.9 for Romney. Please correct me if I’m wrong here. Math/stat skills a little rusty.
IF OBAMA WINS THIS ELECTION, THE UNITED STATES MAY NEVER RECOVER. Urgent Message From Bruce Tefft, former CIA Counter-Terrorism Officer I devoted my life to defending America from foreign threats including the former Soviet Union to dangerous Islamic radicals. When I was sworn in as a CIA officer, the oath included a promise to defend the United States from … READ MORE: http://bwcentral.org/2012/10/if-obama-wins-this-election-the-united-states-may-never-recover/
Interesting it is, but not all that baffling. I remember once seeing an advertisement by Ford which included a survey of owners of a particular class of car. The survey was used to show a strong preference for the Ford model. The survey claimed to be based on 100 subjects, but there was an asterisk at the 100. In the footnote, it said that the survey was actually based on two samples of 50 subjects each.
As the article states, the smaller the sample size, the less reliable the results. This is, in essence, the law of large numbers stated in the contrapositive. The law of large numbers says that the larger the sample size, the more accurate (reliable) the results. However, this does not explain why small samples favor one party or the other so, let’s suppose the following:
Let’s say that, rather than conducting a survey of 1,000 subjects, the survey organization takes 5 surveys of 100 subjects each. We’ll call the 1,000 subject survey a “large” survey and the 100 subject surveys “small” surveys. And, let’s say that the results are that 3 small surveys favor Romney and only two small surveys favor Obama. Now, here’s the trick. Rather than reporting all the surveys or even reporting any of the surveys that favor Romney, the organization reports only the better of the two Obama favoring surveys. If no one never asks the question of exactly how many surveys were taken in toto, no one finds out that the small surveys as well as the large ones actually favor Romney. This is a version of the magician’s trick, the explanation is in what you don’t get to see — out of sight, out of mind.
Why not do it with the larger surveys? It’s too expensive and, since the larger surveys are more accurate and since Romney is the favored candidate, it would be exceedingly costly to pump out enough large surveys to finally get one that favors Obama — if indeed one does get even one. For example, the cost of conducting ten surveys with 100 subjects each is probably not much (if any) more than the cost of conducting a single 1,000 subject survey. If one is really dishonest (not that I would ever suggest anything of the sort of our Democrat party), one could take a single 1,000 subject survey and organize the results into ten (or or even more) smaller surveys and pick the ones that most favor the desired outcome. One can divide up and group the results anyway he likes. That’s one way to work backward to get the result one wants.
So, I would say that Romney is leading in the majority of surveys both large and small. However, because of the higher error range in the small surveys, one is far more likely to find a small survey, rather than a large survey, that favors Obama. That particular survey is the one that the tendentious survey organization will present as proof that it’s candidate is leading. So, it comes down to less a matter of survey and more a matter of highly selective presentation.
Now go back up to the Ford survey. Perhaps, what Ford did was to conduct ten or 20 surveys of 50 subjects each, rather than just two (which is what they hope the reader would infer). Next, suppose Ford picked only those two most favorable surveys and combined the results. That would significantly improve the chances of the getting the desired result over any number of separate 100 subject surveys — especially if the Ford car were not actually the preferred one.
Its not too difficult to work out, but you have to dig into the numbers a bit more. When the Pew Research Center announced it was only getting results from 10% of its calls, it posted lots of data on who was more likely to respond to pollsters and who was not. In essence, the poorer, less educated, and more dependent on government money… the more likely people are to respond to polls.
That means smaller polls will have a smaller chance of getting less Democrat respondents, and it also explains why so many polls are sampling Democrats so heavily: they don’t have a lot of choice. That’s who’s answering the calls for the most part.
The bigger the poll sample, the more Republicans and independents you can actually get to answer, so your poll will tend to reflect a more balanced response group.
the answer is simple. Polls are not in any way scientific sampling of the public’s voting intentions, they are corrupt tools to manipulate the public. The manipulators can make any poll of any size come out the way they want simply by tweaking the numerous internal cheat parameters that they have. Gallup may be the exception, although that isn’t clear either. Rasmussen seems to want to be a counterweight to the the mainstraim media’s mass manipulation, so he is pretty careful not to show too big of a continuous gap between his numbers and the flagrantly liberal one. His goal isn’t to tell the truth, just to be a believable counterweight. Gallup and Rasmussen (and maybe a couple of others) seem to actually become reliable close to the end, to make sure they burnish their reputations, but we aren’t quite all the way there yet. Add 3% to Romney’s total for those two, and a lot more for the other polls.
The reason the smaller polls uniformly favor the marxists is cost. Once you realize that the polls are de-linked from reality, you understand that there is no reason for the manipulators to pay more than they have to pay. Obviously, 100 respondents would be viewed sceptically; but the manipulators have decided that a number between 500 and 1000 passes the average poll consumer’s smell test. Many of the less well know polls probably poll theor relatives and the closest Starbucks rather than random people. Then they just manipulate the data to be what they want it to be.
Even corrupt media manipulators need to be cost effective.
I learned under Frank Newport. No poll is accurate unless there is a sampling of at least one tbousand.
Polls must have interviewed only Likely Voters.
The first question is whether the respondent thinks rhe country is going in the right sirection.
The poll asks if the respondent is strongly, leans toward being Democrat/Republican.
The Republican and Democrat are reversed equally.
The first question differentiates between party affiliation and actual perception.
Margin of error over 3% is usually not worth being designated a viable poll.
You’ll notice that the last three polls are all MSM. Media polls always show Obama ahead or very close. Indeed if any shows Romney ahead it can be construed that he is FAR ahead. In short part of the answer is that the MSM poll results should be interpreted as their attempt to re-elect dear leader rather than a statistical quirk.
I remember once seeing an advertisement by Ford which included a survey of owners of a particular class of car. The survey was used to show a strong preference for the Ford model. The survey claimed to be based on 100 subjects, but there was an asterisk at the 100. In the footnote, it said that the survey was actually based on two samples of 50 subjects each.
As the article states, the smaller the sample size, the less reliable the results. This is, in essence, the law of large numbers stated in the contrapositive. The law of large numbers says that the larger the sample size, the more accurate (reliable) the results. However, this does not explain why small samples favor one party or the other so, let’s suppose the following:
Let’s say that, rather than conducting a survey of 1,000 subjects, the survey organization takes 5 surveys of 100 subjects each. We’ll call the 1,000 subject survey a “large” survey and the 100 subject surveys “small” surveys. And, let’s say that the results are that 3 small surveys favor Romney and only two small surveys favor Obama. Now, here’s the trick. Rather than reporting all the surveys or even reporting any of the surveys that favor Romney, the organization reports only the better of the two Obama favoring surveys. If no one never asks the question of exactly how many surveys were taken in toto, no one finds out that the small surveys as well as the large ones actually favor Romney. This is a version of the magician’s trick, the explanation is in what you don’t get to see — out of sight, out of mind.
Why not do it with the larger surveys? It’s too expensive and, since the larger surveys are more accurate and since Romney is the favored candidate, it would be exceedingly costly to pump out enough large surveys to finally get one that favors Obama — if indeed one does get even one. For example, the cost of conducting ten surveys with 100 subjects each is probably not much (if any) more than the cost of conducting a single 1,000 subject survey. If one is really dishonest (not that I would ever suggest anything of the sort of our Democrat party), one could take a single 1,000 subject survey and organize the results into ten (or or even more) smaller surveys and pick the ones that most favor the desired outcome. One can divide up and group the results anyway he likes. That’s one way to work backward to get the result one wants.
So, I would say that Romney is leading in the majority of surveys both large and small. However, because of the higher error range in the small surveys, one is far more likely to find a small survey, rather than a large survey, that favors Obama. That particular survey is the one that the tendentious survey organization will present as proof that it’s candidate is leading. So, it comes down to less a matter of survey and more a matter of highly selective presentation.
Now go back up to the Ford survey. Perhaps, what Ford did was to conduct ten or 20 surveys of 50 subjects each, rather than just two (which is what they hope the reader would infer). Next, suppose Ford picked only those two most favorable surveys and combined the results. That would significantly improve the chances of the getting the desired result over any number of separate 100 subject surveys — especially if the Ford car were not actually the preferred one.
“Moving averages” decrease the error and you get a better estimate, but you pay a price, you lose some time resolution. So, weekly averaged estimates are more robust than 3-days averaged estimates, they have less variance and are thus more reliable. But 3-days estimators response quicker to the trends than the weekly estimator.
Overall trends are better detected by Gallup (weekly)
Changes in the voter preference are detected faster by Rasmussen (3 days)
MoE is lower for Gallup (weekly) than Rasmussen (3-days).
All of the small sample polls that show Obama with a lead are within the MOE. So tautologically and statistically speaking you can’t actually say that “Obama has a lead” in these polls.
So what you have actually unearthed is the fact that in large sample polls Romney is leading outside the MOE and in smaller sample polls the results are inconclusive.
Most polling is simple propaganda pretending to be science. It’s long past time to eliminate this bias from the public sphere. And it’d not be hard to do. We simply have to borrow a tactic from Obama – sure, you can publish biased polls but it will bankrupt you.
Every publishing organization should be required to also publish odds derived (by whatever means) from their results then take all bets up to some certain maximum amount at risk. They should also then be required to publish a ranking based of their relative results with each published poll. The combination of loss of money and prestige will quickly drive bias from polling.
Is such a scheme permissible? Certainly. The courts have long allowed essentially unlimited regulation of business speech and publishing poll results is commercial and certainly *not* an opinion.
Its always helpful too to consider what the questions are and how they are phrased, such as “Are you an enlightened person who will vote for the Messiah or a racist Neanderthal who will vote for that clod Romney?”
Waiting for the largest poll on Nov 6.
Vote early, vote often.
Notice WHO is doing the small sample polling. They call at least 500 people, to make it look like a real poll, then continue calling until the fluctuation within the MOE happens to show an Obama lead. Then they stop calling.
Never fully trust statistical result unless one is sure of the methodology used, samples targeted, content and wording of questionnaire. For such poll, even 1000 respondents cannot be considered as sufficiently large sample. (vzc1943)
I was right earlier, that the “weighted median” of the polls in Zombie’s article was +2.0 for Romney.
Updated for today, RCP is still computing the “average” at +0.9. But now, the weighted median is +3.0, for Romney. It happens to be easy to check/verify this. As of this writing, there are 10 polls, with a total sample size of 12181. Four of the polls have Romney +3.0, and those four polls have a total sample size of 6988, which is over half of 12181. Therefore, the median is in the group of polls at +3.0 for Romney.
I’ve checked this phenomenon out with the RCP averages of 10 battleground states, and the bias is across the board, without exception. Generally, the bias shaves 0.5-1.0 off what would be Romney’s spread using a weighted median, but in a couple of cases, NH and MI, the bias was 2.2 pts. This (numbers are based on yesterday’s polls), puts Romney at +1.4 in NH (instead of RCP’s average of Obama +0.8), and shaves Obama’s lead in MI down from +4 (per RCP) to +1.8. Virginia goes from a tie (RCP) to Romney +1.
Can’t wait for the largest sample to come in on Nov.6.
I think it likely that all pollsters are sampling approximately the same number of adults, but for some pollsters, the distribution of those samples tend to net fewer registered voters and so fewer likely voters.
I suspect that areas with lower percentage of registered voters in the adult population tend to be very urban or very rural. Unfortunately, I’ve not been able to find the census data (voter registration by metro area population) that would confirm what I think is likely true.
If true, the use of a sample containing more urban voters would account for fewer likely voters and also skew more democrat despite adjustments for party id and that’s why the polls with fewer likely voters favor Obama over those with a sample that includes a more representative distribution of population areas and a larger number of likely voters.
Unskewedpolls.com has a realistic electoral map
Since the end of the first debate there have only been 7 polls that favor Obama. That in itself is too small a sample size to attempt to determine a trend assuming that the hypothesis is:
“Polls that favor Obama have sample sizes = 1000 (one that is > 3000).
So I don’t think there’s enough data to really determine a trend.
Somehow my comment go mangled.
The point I was trying to make is that because there’s only 7 polls that favor Obama since the end of the first debate, the sample size is too small to determine a trend assuming the hypothesis is “Polls that favor Obama have sample sizes < 1000".
The American media is heavily biased to the right. Just today, we have 2% growth and Obama up by 4% in Ohio – all the media could mention was a Romney “special speech” on change. Actually, the speech wasn’t any change at all from his stump speech. Right wing propaganda is rife here – this would never happen in Europe. Try looking at the pre-election polls of Bush and Gore (2000) – they had Bush quite a bit ahead, right before election day. A Karl Rove strategy & that’s what we are going through again. Try reading Nate Silver – he’s likely to get it right.
Rebecca–Your dry, sarcastic wit is fabulous. Let’s have more of it. MSM tilts right? Only if you are standing on your head…a plausible hypothesis here.
Polls are like main stream media. You simply have got to stop listening to them. 78% of polls are biased as per the contractor’s needs (i know the number, I invented it myself!). As one famous quote has it, “There are lies, damn lies and statistics! For all the idiots waiting for the published statistics in order to know how to vote, ,media operators know this fact and therefore skew their outcomes as necessary.
Note that some polls say they are “weighted” to X,Y,Z, a practice that most of them follow. What this means is that they take a sampling, take note of the D/I/R distribution of responders, and then re-multiply them according to whatever distribution they predict for election day. So if they happen to get 10% of responders as D, 60%% I and 30% R, then they will multiply the D, I and R answers to be whatever election day distribution they want. Remember, “weighting” means the numbers have been messed with…..
Fellow veterans beware and be aware that…
Mitt avoided draft during Vietnam (went on a mission to convert Frenchies to Mormonism)
5 of Mitt’s strapping sons avoid military service after 9/11Mitt’s father George avoided military service during WWII.
The Romney’s are a family of self-serving unpatriotic cowards. Mitt is a s**tbag who only loves $$$.
Andy Z. USMC Sgt. 67-70
Sure is tough to beat Obama’s military credentials, so well displayed by his firm response to the Benghazi massacre.
Please let me know if you’re looking for a article author for your weblog. You have some really good posts and I think I would be a good asset. If you ever want to take some of the load off, I’d
love to write some material for your blog in exchange for
a link back to mine. Please shoot me an e-mail if interested.
Regards!