You got it. The professional hairsplitters here are utterly missing, or perhaps simply denying, the big picture - that a majority of the population has so lost confidence in Bush that they’re at or on the verge of thinking he needs to be removed, NOW.
Yet even the Washington Post, whose preeminence today is due in large part to their Watergate reporting, refuses to get into it. Their reasons amount to no more than “It isn’t being seriously discussed because we’re not seriously discussing it.” Damn liberal media … :rolleyes:
Because it’s too easy to get a sample that is larger swinging to one direction or another. As monolithically as we portray politics, it’s not actually that monolithic. The people who answered that poll were taken by the luck of the draw.
I don’t necessarily know what sample size I would say would be useful, though I think at least 50,000, 1000 from each state would probably be the best way to determine something like this, not 24 from each state. We don’t even know which states he took his sample from either. We don’t know what demographics he hit with his poll.
1% of the American population is 2.8 million people. So this statistic has a sample size of roughly .0005%. As this statistic is completely unverifiable, it is basically useless.
Don’t get me wrong, I think Bush should not only be CONSIDERED for impeachment, but that he should be impeached, I just think that it’s kind of naive to think that you can judge the opinion of the entire populace on such a small sample size. I am happy that the children are enjoying their play as the field of statistical analysis matures more fully, but I still find it dubious that it is reflective of any reality, especially since most polls like this cannot be verified at all. It’s one thing to take a poll and be right about who will be elected, because it’s verifiable, but for determining the opinion on specific issues it’s far more difficult. When dealing with a determination like the outcome of an election it’s easier because there are fewer variables. When presented with a dirth of choices for president, people are like to vote a certain way based upon party affiliation. However, party affiliation has less impact over single issues, than it does on the choice of candidate.
Lemur866 That’s utter crap. You can miss whole swaths of demographics by not taking a large enough sample. I think it is our dependency on such statistical sets that is what is fucking us up. Too many one size fits all mentalities. I disagree with your assessment that you can take a sample and be reasonably assured that one person can speak accurately for 200,000 people each. If sample size doesn’t matter why do they use 1,216 rather than 100?
Lot’s of modifiers there. Language is a subtle thing, and the OP draws a misleading conclusion.
Now, if the question were: “Do you agree or disagree that President Bush should be impeached?”, then the OP would be correct.
As for the sample size issue, if one draws a nice, well-behaved random sample of US Resident Adults, then the basic formula for a 95% confidence interval for a proportion P is approximately:
[p-2sdp,p+2sdp], where sdp=square_root(p*(1-p)/n). As it turns out, sdp is maximised over the domain [0,1] by p=.50=(1/2), so the upper bound for sdp is 1/(2*(square_root(n)), which is quite small for n>1500.
There is no crap in what lemur said. 1000 people - chosen carefully (stratified pseudo-random sample - make sure it represents all groups according to how they appear in the population) is plenty - actually, the marginal return for each additional person polled goes down remarkably after 30 people. And good polls say what the margin of error is anyhow.
Agree that they shoulda worded the question differently, if they wanted to know what percent of Americans think he should be impeached.
If Bush wiretapped American citizens without the approval of a judge should Congress consider holding him accountable? No brainer…yes, of course they should consider holding him accountable. If President Bush raped baby ducklings while lighting $1000 cigars lit from burning minorities and the poor all the while howling with mirth should Congress consider holding him accountable? Certainly.
Too many qualifiers IMHO to be meaningful. Why didn’t they just come right out and ask “Should President Bush be impeached?”?? Wouldn’t that be a question that cuts right to the heart of the issue?
FTR…should President Bush be impeached? I’m not sure. Certainly if he’s done something demonstrably illegal he should be. Should Congress consider holding him accountable through impeachment? I’m not sure right now…has he done something demonstrably illegal? If so, certainly they should do more than consider it…they should DO it. “If President Bush wiretapped American citizens without the approval of a judge, do you agree or disagree that Congress should consider holding him accountable through impeachment.” Certainly they should look into it to determine if what he’s done is in fact illegal…and if so they should proceed to impeachment.
The interval form of the estimate is [52-2.9,52+2.9], which simplifies to
[49.1,54.9].
Now, the true proportion can lie anywhere in the interval. It could be in the [49.1,50) portion.
The hidden assumption in the poll is that if POTUS did what they say he did, then he should be up for impeachment. The foregone conclusion is that they would also want him convicted in the process.
If you want a depressing poll/quiz, imagine the same respondents answering questions about the underlying constitutional law in play here, about the impeachment process itself. I doubt that a supermajority of the respondents actually know enough about the relevant law to actually have an intelligent opinion one way or the other.
How about this smart guy. How can we verify if the opinion of one person on this subject actually reflects the other 200,000 people that his/her opinion is representing?
If you can’t do that, then maybe you need to lay off the superiority for a minute.
If you can, then I’ll go enroll in a statistics class post haste. I see this argument all the time about how sample size is irrelevant, yet statistics are so regularly wrong, making them nearly completely useless for questions such as this. If you want to talk to me about marketing trends based upon who purchased what and when, then I’ll agree that statistics have their place, but in the realm of public opinion about issues based politics they are wrong as often as not, and I don’t know anyone who works in statistical analysis that wouldn’t tell you it’s a crap shoot.
I don’t need a statistics class to know the logic of it. If your sample set misses a certain demographic, then no one in that demographic is represented. It’s pretty simple really. Like if I don’t interview any black people, then I don’t know what black people think on this subject. However, “Black” people are not a monolithic entity, so again it’s a crap shoot about WHICH black people I interview. If I don’t hit Black Baptists, then I don’t know what Black Baptists have to say on the issue.
When you were trying to be high and mighty and tell me to take a statistics class you didn’t address my question as to why the sample size wasn’t simply 100 people if the sample size is not relevant. So while I may not be an expert in how statistics are derived, I seriously doubt that if I took this statistics class, that I would suddenly be convinced that people can speak for 200,000 people on this issue. All that this tells me is that they spoke to 9% more people who support CONSIDERING impeachment, than who don’t think it should be considered. It doesn’t tell me a damn thing about the whole country. For instance, there are more people in California and New York, than any other three states in the union. Those are two states that lean heavily to the left. 20% of people live in either New York or California, which lean heavily liberal. Do you think that 200 of the people he interviewed are from those two states? Both of those states are also incredibly ethnically diverse, ethnically diverse states tend to lean toward the liberal side of things also, so how do you hit all the demographics within that ethnic diversity? Were middle class arabs represented? What about poor arabs? How many Mormons did he interview? What about liberal activists?
So don’t give me this shit about the sample size not mattering, because it does.
mswas, you are correct when you state the sample size matters. You are actually in agreement with Lemur866 on this point.
Bolding mine.
Where you seem to be confused is in thinking that there must be a direct relationship between sample size and population size. Please take another look at **Cerebus’ ** equation in post 23. The sample size needs to be 1,000 and not 100 because a sample size of 100 will not have a confidence level of 95%.
If you had had such a class you would understand that what seems like “logic” to you isn’t actually accurate in such a case.
But I suspect instead you’ll just hurl at invective at people who really know quite a bit more about the subject than you do.
It’s kind of like certain other areas of science (say, Quantum Physics) - just because the average layman can’t understand them does NOT make it bullshit. No matter how much your brain would prefer it would be so.
The only real requirement vis-a-vis sample size in terms of 95% confidence is that sufficient sample size is needed to ensure approximate normality.
If such sample size is met, then the “2” in my earlier statements ensures approximate 95% confidence.
The stronger role played by sample size is the precision in the estimate:
Margin of Error = 2sdp = 2square_root(p*(1-p)/n).
In practical terms, sdp is maximised at p=1/2, so an upper bound for sdp is simply square_root(1/4n), so an upper bound for our margin of error is approximately
2*square_root(1/4n) = square_root(1/n)…
As for the sampling issue and representing a large frame.
We need a random sampling scheme, where in the simplest case, every single US Resident Adult has an equal probability of selection.
In practice, stratification and sometimes also clustering is employed in practice, which complicate the sampling design.
If it is, well…nuts. :smack: I thought I got it. Could you say that again - as if you were talking to a slow, dull child, which is apparently what I am?
I will contribute something worthwhile to Great Debates someday! I *swear * it.
I can’t see anything in Cerebus statement that conflicts with what I am saying. What I was saying is that with a sample of 1,216 you are very likely to miss demographic clusters in a population of 280,000,000. I don’t believe that 1 person can accurately represent 200,000 people.
Though it was nice to have Cerebus and Just1Lurk try to explain things to me rather than just calling me an idiot and telling me to take a class.
I am disputing whether or not a sample size of 1216 is enough to achieve approximate normality.
Thanks Cerebus and SaintCad It’s nice that there are some people who will give you some examples when you are having trouble believing something rather than just telling you you’re an idiot and need an education.
Oh wait, I’m sorry, I meant to thank Cerberus and typed SaintCad instead. My mistake. I apologize to both Cerebus and SaintCad for misrepresenting them. Cerberus was nice and helped me with my ignorance. I am terribly sorry for misrepresenting you as that type of person SaintCad.