Sure, but that’s not what’s counted in an election is it? What’s counted in an election is the opinion of those who give enough of a damn about the issues and candidates to make some personal effort.
A nationwide phone poll would change the nature of elections from an opt in format to an opt out format.
Just to play devil’s advocate here: it doesn’t have to be a phone poll to be a sample. It could be a ballot, just as at present, but with only x% of those normally eligible to vote actually voting.
One way to do this would be to use the SSN to get a sample. A really simple way would be to randomly pick a number between 0 and 9, and then say that all the voters with an SSN ending in that digit will actually vote. (Whether that gives an unbiased sample, I don’t know: I do, however, know that using the first digit would give a seriously biased sample, because that depends on which state you were in when you got the SSN.)
One thing that would happen then, of course, is that the political campaigners would do their best to find who has those SSNs, and get them to vote. You would probably have a significantly higher tuirnout than at present, because of this highly targeted campaigning on the eliugible SSNs. (And that might be a good thing – or not, as you see fit, and as you see that advantaging or disadvantaging your poilitical party).
Can I stand for president on the platform of doubling taxes for everyone else and eliminating them for that year’s voters?
From what I understand of sampling, you have to have a certain number of responses to assure that you actually heard from a representative sample. Not contacts, but actual responses.
So let’s say you’re polling the entire U.S. to see who they want for President. You start with a hypothetical 100 million voters. Let’s say 50,000 responses constitutes a valid sample.
But then you have Senate race in a state with 10 million voters. Statistically, you might need 20,000 for a valid sample.
Then you have a Congressional race with 200,000 voters. You might need, say 5,000 for a valid sample. But 435 Congressional districts would net more than 2 million responses.
By then, you have to oversample the entire electorate to get valid samples for the smaller subsets. At that point, you might as well throw the whole thing out and just ask everyone – i.e., an election.
Statistics is only as good as the experimental design and implementation. All those pitfalls you see in current elections are still there a sample of 1 million people. How do you choose the 1 million people? How do you know they aren’t dead, fictitous, fraudulent? How do you know they haven’t been bribed? Are they really random? How do you record their answers – chads, voting screens, paper? Do you take account that certain groups vote less than others when you are constructing your sample?
If you could solve all the problems needed to get a true scientific poll of 1 million people that was beyond reproach, there would be no reason not to extend that methodology to the full nation. This is a garbage in, garbage out problem. Statistics is fine, but it is only as good as the inputs. If you can get the inputs to be good enough, than you might as well count them. Counting has a much better error rate than statistics.
As Richard Feynman sort of said, 3 million people guessing the length of nose of the Emperor of China (who no one ever sees), and taking an average, and calculating p-values… none of that makes it science. It’s still based on garbage, no matter how much math you do on the garbage.
That’s why I wrote in a previous post, “Of course one can always argue that if your don’t care enough to vote you shouldn’t be represented.”
However, I’m not all that sure about the assumption that all people who vote “give enough of a damn about the issues.” I have a sneaking hunch that many don’t know a damned thing about them and vote for a party because their family has always been _________________.
Whoops, I missed that, sorry. Still, the difference between self-selection and random-selection procedures needed to be emphasized anyway.
It’s called, affectionately, the Electoral College.
Question answered, thread closed.
No, I really want to follow this.
Peace,
mangeorge
I certainly am not advocating using a statistical sample rather than a general election. My point is that it is a sort of “conservation of difficulty” thing.
The sample used in the general election is large but self-selected and isn’t ipso facto representative of the population as a whole. It might be so, but it isn’t guaranteed. Added to that is the difficulty of knowing that there was an accurate vote count in close elections since the only control seems to be the exit interviews of the media. There is, of course, the recount process but again, that is open-loop with the only control being the use of representatives of the two major parties to do the recounting.
Sampling, though, also has its difficulties. Getting a proper sample is a problem, and since elections are by states there would have to be a sample in each state. The sampling design would have to be carefully watched because it could be manipulated. Making up the sample would probably require some test polls and analyzing the meaning of the results would also be an open-loop process because the desired result is hard to determine.
I think if proper samples, and that’s a big if, could be constructed for the states, statistical polling would be at least equal to a general election.
In any case, it’s all hypothetical. It just ain’t goin’ to happen.