This is one of the reasons I don’t like polls and why I didn’t blog about John Kerry’s negative bounce yesterday. I knew that Big Media would come out with a poll showing something different.
According to the Washington Post (registration req.), Kerry’s ahead of Bush.
Any poll-knowledgeable readers out there? How reliable are these things? I’m feeling a little lazy today and don’t want to do the research myself.
{ 1 trackback }
{ 28 comments }
The polls are all over the place, even by the same polling companies depending on if the polls were taken Saturday-Sunday or Sunday-Monday.
If you measure the standard of error, they tells us even less.
This poll was of registered voters. Generally the only poll deemed close to being accurate is one of likely voters. Lastly, This poll was taken over the weekend and will show a known bias towards Democrats. Apparently Republicans have better things to do on weekends then hang around the house answering polls!
So, there’s two parts to consider with an poll — how many people were polled and who was polled. I could poll 10 people on my floor today, but the results would be very skewed towards Bush (then, I would also have to consider the possibility of people lying to me), and 10 people, even randomly chosen, has a huge margin of error.
As for the who: if you look at the graph, all they’re sampling from is =registered= voters. As we know, nowhere near all registered voters will actually vote in the election. Some polling outfits, who spend much more money on getting good samples, will give results for =likely= voters, which is based on what the people themselves say as well as if they voted in the past. One trouble the Dems may have is that there are particular groups that overwhelmingly vote Democratic when they do vote… such as blacks… but really, do you see Kerry firing up that base? Some people may think “Meet the new boss, same as the old boss” and give the election a miss.
As for the how many: this is where the statistics come in. Let’s just assume you’re sampling from a large population of people, who will give one of two answers: Kerry or Bush (it gets more complicated if there’s more answers, so let’s just go with this). Say it’s really split 50-50. As you sample more and more people, the sample percentage will deviate from the real percentage less and less, on average. It’s like flipping a coin a bunch of times – the more you flip, the higher the probability the percentage of tails will be within a certain percentage of the real percentage.
In any event, statisticians can tell you the “margin of error”. In the case of polling about 1000 people, the margin of error is 3%. This means that the true percentage has a 95% chance of being within 3 percentage points of what was given. So if Bush is polling at 44%, the possible range is 41% up to 47% for Bush. That’s a range of 6 percentage points… not too impressive, eh? To get within 1 percentage point, one has to sample 10,000 people. However, and this is reasonable to consider, these polls are being done multiple times. Assuming that most people aren’t flipping between choices, you could possibly average the results from the 10 latest polls to get an answer within 1 percentage point (you have to assume they all are polling the same number of people, and are using the same sampling criteria – registered voter vs. likely voter vs. whatever).
In any case, this is all silly. It’s just a way for a certain type of person to make money (and this is not to say I wouldn’t be one of those people). There’s only one poll that counts. And people do lie to pollsters (I know I have).
More to the point, the popular vote doesn’t count. Suppose the entire states of California and New York went for Kerry. That would greatly raise his popular vote percentage, but would have no net effect in the electoral college — he could have gotten the votes for NY and CA simply by carrying each at 51%. I prefer looking at the breakdown of polls by state, as that gives one a better idea of what’s going on. Of course, 50 polls are much more expensive to run.
when I said “10 people”, I meant a 10-person poll.
Anyway, the stuff I was talking about is generally covered in a introductory statistics course, even at the high school level. I believe stats is a much more important math topic than trig or calculus. Statistics are being used a lot nowadays to make arguments and decisions, so I would like people to be able to interpret them.
Here’s where I go to watch the polls:
http://www.dalythoughts.com/ecb.htm
It’s a state-by-state breakdown, and will even show you how the state went in 2000. Fascinating stuff.
Two words: Registered Voters. I’m sure they have a Likely Voters version of this poll too. And I bet it reads ‘Push’.
Whoever said that weekend polls favor dems are wrong. Weekend polls oversample Republicans, because Democrats tend to have more busy lives in urban areas and are unreachable. While Republicans with families in suburban/rural areas tend to stay home. Therefore, polls done over the weekends tend to skew Republican.
Meep, good breakdown.
OpinionJournal has a cool flash Electoral College calculator at http://www.opinionjournal.com/ecc/calculator.htm It includes historical data going back to 1980 and the kicker is that you can even create and save your scenarios. Combined with the State-by-State polling data from “dalythoughts”, one could amuse themselves to no end.
LOL, the dalythoughts comic strip of the day about John Kerry’s bunny suit and Jimmy Carter’s Killer Rabbit thing is hilarious
Kerry’s numbers this weekend are significant for one reason. He did not get a 15 point ‘bounce’. That is usually what candidates get when their convention comes about, if I understand correctly. There is typically a balloon in the numbers, and this did not happen. Given the chart you posted, Kerry may have had a 2 or 3 point bounce, but not the 15 point that has come to be expected.
I theorize that either people have already made up their minds and are ensconced in their decisions or Bush is about to win in a landslide. The information/internet age has changed the way people think about politics, and 9-11 has also made people more politically aware and… shall we say… stubborn. We shall see which of my theories holds true in November.
A really good poll is Iowa Market, at http://www.biz.uiowa.edu/iem. Really good because people actually put their money on the market. And they seem to be very accurate, so far. Notice especially how the lines diverge just after the Dem convention (click on the President Winner-Takes-All link and view the graph).
“Generally the only poll deemed close to being accurate is one of likely voters. ”
The problem is that they usually mean ‘voted last time’ when they say ‘likely voters.’ There’s arguments now that that the registered voter polls may be more accurate, as people are expecting previous non-voters to participate more.
meep:
“I prefer looking at the breakdown of polls by state, as that gives one a better idea of what’s going on. Of course, 50 polls are much more expensive to run.”
http://www.electoral-vote.com updates daily with state-by-state polls. includes downloadable excel file with all the date.
I have seen similar poll numbers from other news sources in the past couple of days (Fox and CNN). However, those same polls, when conducted among *likely* voters, show the numbers almost exactly flipped: Bush leading Kerry 50% to 40-some%.
Those poll numbers are taken from this data I belive,
< a href="http://www.washingtonpost.com/wp-srv/politics/polls/trend_080104.html">http://www.washingtonpost.com/wp-srv/politics/polls/trend_080104.html
This Washington Post-ABC News poll was conducted by telephone July 30-August 1, 2004 among 1,200 randomly selected adults nationwide, including 940 self-identified registered voters. Margin of sampling error for overall results and for registered voters is plus or minus three percentage points. Fieldwork by TNS of Horsham, Pa.
I would recommend to anyone to read “Off with their Heads” by Dick Morris. He has a good section that in detail discusses how the New York Times conducts polls and how they skew the results. Very informative and you will never believe a poll from them again as to they always add 3 points to the liberal side.
You know the old saying: “There are three types of lies. Lies, damn lies, and statistics.”
I don’t think the polls mean anything until after the Republican National Convention. But KEEP IN MIND, the CA recall election where Arnold was supposedly in a tight race with ___[What was his name again?]
I think that the media is so in for Bush that they have been “push polling” for some time now. They just can’t help themselves. Push polling is an intentional phrasing of questions and sample selection to achieve a desired result in the hopes that the polling will push people toward that view/candidate.
The only polls that count are the internal campaign polls. You can tell more about how they are doing by observing the tactics and spending patterns of the campaigns than any poll will tell you.
In CA the Dems paraded every Democrat they could find to help Gray Davis, that should have told you just how bad things were. It also tells you just how unpredictable the Latino vote is.
Let’s remember one thing: the election is still three months away. The Democratic Party is in disarray. All the Kerry supporters I know (and that’s most of the people I know) are growing less sure of themselves and more desperate. Their ideological edifice is set to crumble. Now is the time to step out and start talking to people. There is time.
The only poll that counts is the one in November.
I’m no expert, but I’ms sure we all recall similar fluctuations in the polls four years ago. I think polls are only as good as the day on which they are taken. What I mean is, if the vote was taken today the poll may, in fact, be accurate. However, as predictors of future voter behavior, polls seems to be just another way to generate headlines.
As far as the 15 point bounce is concerned, the estimate was from a Republican source who was intentionally trying to build expectations (that no Kerry supporters bought). The 15% bounce, while not unheard of, is not close to the average.
It’s gonna be a close one. Look for a few key states (thanks, Andy) as being pivotal: Ohio (most important); also Oregon, Wisconsin, and New Hampshire.
Kerry’s poll numbers are worse than they appear.
He had two tried-and-true opportunities to boost his numbers: the veep selection and the convention. Between them, they should have given him near a double-digit lead. They didn’t come close. He may have lost ground.
This may mean that his support has already maxed out. Meanwhile, the economy improves and the Republican convention is yet to come. The unexpected could happen, but this is not a good trend for the Kerry campaign.
Incidentally, the Washington Post usually oversamples democrats by a very wide margin. 50-44 in favor of the democrat probably means it’s a dead heat.
Well, Carter got a 16 point bounce, and Dukakis had a 7 point bounce. I did just read, however, that there is a very low number of undecideds, and that is where the bounce generally comes from. So, the more people that have their minds already made up, the lower the bounce.
I don’t believe any polls due to the inherent selection bias.
Where is the evidence that people who voluntarily choose to answer a poller’s question accurately model the overall population…
Fairmodel’s been updated. W 57%.
Iowa Markets W’s up over 50 since the dem convention.
There were supposedly all these undecideds these past few months, Kerry doesn’t get a bounce and now there’s no undecideds.
Rigght.
Republicans vote more often and that skews the results. Subtract about 5-15% (depending on the state) from the Kerry side and you’ll have a more accurate picture.
The polls thought McGovern/Nixon was close (both with 41%).
One thing to consider about phone polls is … I lie to them all the time.
My goal is have polling so discredited that nobody cares what they say. With that in mind I NEVER answer a pollster truthfully. Even if the poll is on cookies I lie.
Comments on this entry are closed.