Posted on 10/25/2004 11:14:54 AM PDT by You Dirty Rats
Polls for Pols. All we hear day after day, almost minute-by-minute are polls, polls, polls! Exactly what good are they?
If you, like me, have had the masochism to take graduate-level statistics, then you know just how much nonsense is spewing from pollsters these days.
Classic example -- "Statistical Dead Heat" If Dubya is up 3 point on Kerry and the "margin of error" is plus or minus four points, the media claims it is a "Statistical Dead Heat". Is that true?
Well, no! Here's why. All of these "margin of error" numbers derive from probability laws of mathematics. The usual assumption is a 95% confidence interval. Why 95%? Arbitrary, really. 1.96 "standard deviations" from the average represents exactly 95%, so two standard deviations represent about 95% also. If you want 99%, that's three standard deviations. That's how the mathematics of the "bell-shaped curve" work out. When teacher grade "on a curve", that's the curve they reference -- even if they don't really understand the details.
OK, so what's the stat geeks point, you ask. Here it is. If we sample from a large population, and use so much of a sample per size of the population, mathematical tables will tell you how wide the "confidence interval" is. Forget voters for a moment; think red and blue marbles. If there are 1,000 marbles in a bag and you pull out 10, and 40% are red, then the average is 40%, of course, but the margin of error is very wide, since your sample was so small. If you pull out 100, and 40 of them are red, same 40% average, but now the margin of error will be less. If you pull out all 1,000 and get 400, then 40% is still right and the margin of error is zero! The sample size, then, combined with the observation of the sample and the size of the total population, gives you a point estimate and a margin.
Bake to Dubya and Kerry. If a media poll says Dubya 48-45 and the margin of error is plus or minus four, what they are really say is that, if the underlying assumptions are accurate and the sample is a fair sample, and not biased, then there is a 95% chance the end result will be between Kerry by 1 percent and Dubya by 7 percent. The chance that Dubya will get more is actually very high, since only a small part of the "bell-shaped curve". Opening my statistics book, and using this example, Kerry wins only if the end result is more than 1.5 standard deviations in his favor. What his percent chance in this case? 6.68%!!!!Feel better, everyone?
However, there is a BIG problem. The mathematics of what I described are bulletproof when sampling red and blue marbles. They are NOT bulletproof if the sample is statistically biased. By ""statistically biased" I don't mean that the statistician necessarily wants to force a result he wants; it means that for some reason the sample drawn is not random across the entire population. All of us know this -- if you ask 1,000 voters in Utah, you'll get a very different answer than 1,000 voters in DC.
OK, nothing you don't already know. Here's the REAL problem. How do you get an unbiased sample? After "Dewey defeated Truman", pollsters only sampled marginal precincts. They look at swing areas, and use those to decide. How do they know what are marginals and swings? By looking at HISTORY. No problem, if nothing much has changed since the last election.
Think nothing has changed? NO!! How about the reported shift of Jews and African-Americans? What about the changes from 9-11? How about how energized different groups are? How about cell phones not getting calls?
No Republican has ever won without winning Ohio. No Democrat has ever one with less than four Southern states. The Red Sox have never beaten the Yankees in the postseason. Well, times change. The Yankees are home. Kerry won't win four Southern States. Dubya may lose Ohio. History is nice, but it isn't 1856, 1860, 1932, 1960, 1968, 1980, 1992, or even 2000. Too much has changed just since September 11, 2001.
Finally, you might ask why the polls are all over the place. How can Kerry be up several points in Ohio in one poll, and down several in another? The ACTUAL position right now, remember, is clear if you could accurately get the opinion of every voter. You can't -- so pollsters sample.Different results come from dissimilar samples, either intentionally or accidentally.
With all of the changes and special factors in this race, and with the wildly-changing polls over the same states and nationally, we know exactly one thing: For the vast majority of us, the polls are absolutely worthless!!
The only good polls are taken by the absolute best experts in the field, who know better than other how to sample. In an election this wild, with all the changes, even these guys can be wrong. Think they are working for a Networks? NO!! The best work for the Candidates!!!
Bottom line -- forget the polls. They tell us nothing reliable. On average, Dubya is up -- so it's likely that he is. But it's not be a lot, so nobody knows anything yet. We just have to bear down, do what we can to help the Bush-Cheney Campaign, and VOTE FOR W!! WE WILL WIN WITH W!!!!!
A little primer on the boring but critical details of statistics and polls. Every time I hear "Statistical Dead Heat" I want to strangle those ignorant morons in the media.
Yes, brother. Preach it, brother. This deserves and Amen. A Hallelujah. No sarcasm intended.
Great post. Statistics can be manipulated.
BTTT!!!!!!
What about Dumbies? ;0)
The headline is repetatively redundant.
Exactly 78.63% of all statistics are made up on the spot.
The best work for the Candidates!!!
And that is important to remember.
When 9 out of 10 polls show Bush in the lead, you can be pretty confident that Bush is REALLY in the lead. Not certain, but pretty confident.
3 out of 4 people make up 75% of the human population...
9 out of 10 dentists think the last guy is just being a jerk...
Actually.. I believe the margin of error applied to the polling results of both candidates?
If Bush has 48, Kerry has 45.. with a 4 point MOE, The possible extreme results are not Kerry +1, Bush +7.
The most extremely possible results within that 95% confidence level is Bush 44, Kerry 49.. Kerry +5.. or Bush 52, Kerry 41, Bush +11.
The 95% Confidence level of a Bush 48-Kerry 45 poll is that it's somewhere between Kerry +5 or Bush +11. 5% of the time it will be outside that margin.
I believe so anyway, unless someone wants to correct me.
Do keep "vanity" posts to a minimum - Free Republic is primarily a place to discuss news, articles, and editorials. Vanity posts, creations of the poster him or herself, should meet a high standard of quality before one is even considered worthy of posting. Often a relevant current thread or general announcement, catch-all thread is a much better choice for a brief question or comment.
Please, no vanities in news / current events.
The average voter in the United States has 1 testicle and 1 ovary.
Mathematically, if the Margin of Error is 4 and Dubya is up 4, then the chance he'll win the popular vote is about 97.5%. If he's up 6, MOE 4, then it's 99.5%. Hehe.
Thanks for the well written and cogent analysis. Can I pose one question. Most everyone here seems to put a great emphasis on the RealClearPolitics poll averages..even Rush talks about it often..It's become the hot site, it you will...But, isn't it the classic example of GIGO..
Now all we have to do is gather up the will to prosecute voter fraud, which we should have done last election, so it won't haunt us next election.
OK, so what you're saying is-- all other things being equal -- Whatever the margin of error is, if the poll shows Bush ahead by x percent, there's as much chance that it's wrong in his favor (x+1) as there is that it's wrong in Kerry's favor (x-1).
ping
I've also noticed the media bias in the description of the polls; when Kerry is up by 3 points he is "pulling away" from Bush, but when Bush is up by 3 in the same poll it's a statistical dead heat.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.