Polling-public opinion or propaganda?

[coplinger]

Has anyone else noticed that these (so-called) 'national' polls, that the news media seem to take as the gospel, usually only consist of 300-1000 people. I went through 6 or 7 major polls on the candidates & the largest number polled I found was 1028.

It would seem that in a nation as big as ours, it would take a lot more than a sample of 1000 to get anywhere near accurate poll numbers, yet so much of our political discussion is based around this polling. I bet if you poll the right 1000 people, you could get any result you were looking for.

I think maybe the purpose of these polls is not to be a barometer of public opinion, but to be an influence on public opinion.

This may explain why the voluntary online polls (MSNBC's post debate poll had something like 15,000 responses) have Ron Paul doing so much better than these managed 'national' polls.

Any thoughts on this?

 

[axiomata]

It's simple statistics. If the participants are chosen truely randomly then a sample size that large is scientificically accurate to the +/- percentage the poll lists.

Though not having a landline of my own, I know I will never be polled so its technically not perfectly random, but if we are interested in measuring Paul's odds in the primaries, these polls are the most acacurate.

The good news is that our online work will carry over to the scientific polls slowly but surely. Paul is up to 3% from 1% this week.

 

[coplinger]

In a country of 300 million, a poll of 1000 people can be considered scientifically accurate??

That's one three thousandth of a percent!

I find that hard to believe.

 

[axiomata]

http://whyfiles.org/009poll/math_primer.html

A pretty basic explanation without too much math.

 

[Brandybuck]

It's simple statistics. If the participants are chosen truely randomly then a sample size that large is scientificically accurate to the +/- percentage the poll lists.

There's several problems with those polls. First, you can't get truely random samples. The demographics of people who hang up on pollsters is different than the demographics of those who participate. Phone calls during hours convenient to the pollsters reach pick different samples than those made during inconvenient hours.

Second, how do you trust the participants? How do you know they're answering accurately? That sounds like a strange question on my part, but it's valid to ask. Maybe they're answering "Obama" or "Clinton" because they don't want to seem like a racist or sexist. Maybe they answered "Rudy" because they don't want the pollster to finish the list of a dozen names. Maybe they got to go whee and are trying to hurry the poll along!

Third, pollsters tend to lead the participants. They're not supposed to, but they do. I just got interviewed on a telephone poll an hour ago (for a local ballot measure). The pollster was using very leading language, and was trying to be folksy and friendly.

Third, there are the questions themselves. Do they randomize the order of names each time? Are they asked in a truly neutral manner? Are they preceeded with statements of "fact" intended to put the question in "context"?

Finally, who are the participants? Were they viewers of the debate? If not, then these polls are essentially referendums on name recognition and media exposure. Are they people who will actually get out and vote in the primaries, or just someone who bothered to answer the phone during dinner?

The media is calling the the online internet polls "unscientific". They're right, they are unscientific. But so are the offline polls!

 

[axiomata]

Your criticisms are valid ... especially questioning the randomness of the sample and pollsters leading their questions. I was just primarily pointing out how theoretically, assuming a perfectly random sample and unbiased pollsters, such a small sample can scientifically represent millions of people which I read as coplinger's main question.

 

[coplinger]

It would seem to me that these polls would be more accurate for the candidates who had a higher percentage, but that the lower the percentage, the less accurate they would be.

I work with audio and this seems to be comparable to bitrate:
16 bit is pretty accurate for a loud signal, but not nearly as accurate for quiet, low volume stuff. Whereas 24 bit is accurate at much lower volumes.

It would seem that to get accurate numbers for those that poll less than say 20%, you would need a much larger sample.

I can see if you were voting on 2 or 3 candidates that 1000 might give you relatively accurate numbers, but for 10 candidates....?

 

[axiomata]

The thing to remember is that the margin of error is the same +/- percent no matter the candidate. So a candidate who is at 1+/-5% could actually have six times as many true supporters as the poll indicates. For a candidate at 30+/-5% if, for example, he/she really has 35% support this is only a 1.14 times more. I think that is what you were getting at with your audio analogy.

 

[coplinger]

Yeah, that makes sense.