Forums | MacLife

D'Eyncourt wrote:

resedit wrote:

D'Eyncourt wrote:

Wow. You really do not understand statistics, do you?

Yes. I do.
Where you take the exit polls may be representative of the sample of THAT polling station - of voters that happen to respond to the exit poll at the times the exit polls are done. Different demographics vote at different times and at different polling stations (as they live in different areas) - not to mention the fact that different demographics also have different turnout ratios.

When you decide where to do your exit polls, you are creating a biased result. It may be biased towards the norm or it may be biased towards an extreme, but it is biased.

Do find some perverse satisfaction in assuming that people are automatically biased against you?

Um, I never said any such thing.

Do you somehow think that pollsters are unaware of your objections and thus do not try to adjust for them?

You can't always adjust for them.
You can analyze differences from exit polls and the official count, and try to find what caused those differences, but it isn't static.

But I understand that in the resedit universe nothing important happens without your input; thus if you never have been polled then it must be because of bias.

Sheesh.

I never said that because I haven't been polled there is bias.
There is always bias in any statistical analysis based upon a sample of the population. Sometimes that bias is towards the norm of the population, that is the goal - but by choosing the polling stations with which to do your exit polls, there already is bias - because they are chosen for a reason, they are chosen because they are what the pollsters think will give them the most accurate representation of the bigger population. Sometimes they are accurate, sometimes they are not, but the sites are chosen and are therefore biased just in that aspect.

Even if they randomly chose where to do the exit polls, they would be biased towards the random selection.

Flip 10 pennies. Statistically, half of them will be head and half of them will be tail. However, it is not uncommon to get 7 or even 8 heads, hell - even 9 heads can happen.

If your exit polls are chosen because of analysis from past election trends, then they are biased for that reason. The bias may be the norm, it may not be.

If the exit polls are chosen randomly, they are also are biased towards the random selection.

Either way, exit polls are not a valid reason to say an election was rigged.

You can however have mathematicians try to figure out what the liklihood of the disagreement of exit polls with actual polls is - but even that analysis ultimately is subjective and prone to both error and possible bias, and even a low liklihood that the difference is legit does not invalidate an election.

An online friend purchased two honduran milksnakes that were sold to him as normal phenotype but guaranteed HET for a rather cool morph. Statistically (recessive gene) 25% of the young should have expressed that morph.

First mating, he got 10 eggs. He expected 2 possibly 3 morphs. He got none, and complained that he was ripped off. However, in reallity, there was a 1 in 20 chance that his results would happen.

Guess what - the next breeding, he got his morph (proving both were HET), and he had to apoligize to the breeder for publicly stated he was a scam artist.

Point being - randomness doesn't always follow the norm, and even randomly selected exit polls will be biased - either towards the norm, or to an extreme. Thus - drawing conclusions based upon them is dangerous.

With the snake example, you can figure out the statistical liklihood of what happened with precice mathematics. With an exit poll, you can't because things are not that simple, it requires subjective analysis (differences in demographics that cause votin differences can be hard to identify) to try and figure out what the liklihood of the disagreement from expexted is.

But do you understand why not everyone doesn't need to be polled for a sample to be taken?

Yes. Larger the sample size, the more accurate - theoretically.
Where the theory breaks down is that where the samples are taken (and how they are taken) can have a signifigant impact on the result of the exit poll. It no longer is simple lower division statistics, because the data points are not random, the data points are biased based upon the local populations.

You take 10 test scores at random from a lecture, and you will probably be closer to the average than if you take 10 test scores from the students with the most absences - yet both have the same sample size.

Do you understand the issue?

It is very difficult to construct exit polls that truly are a valid random sampling of the population.

Last edited by resedit (2006-11-01 8:27 am)

after-life wrote:

resedit wrote:

But do you understand why not everyone doesn't need to be polled for a sample to be taken?

Yes. Larger the sample size, the more accurate - theoretically.

So it's settled. You really don't understand the basic rules of statistics.

Once you go past a certain (low) threshold, increasing the sample size DOES NOT make it more accurate.

Actually, what happens is the increase in accuracy become marginal, and mathematically insignifigant - assuming your method of collecting data is good. Often the method of collecting data is not good, and a larger sample size can hedge against that error.

btw - just so you are aware, I was a math major.
The statistics courses I took required an understanding of multi-variable calculus.

Stick that in your pipe and smoke it.

Here's a page that discusses some of the issues with polling:

http://en.wikipedia.org/wiki/Sampling_(statistics)

after-life wrote:

If your data collection methods are good, you don't need a huge sample.

And there lies the problem.

bratboy wrote:

Do we all agree that rigging elections is bad, and that it shouldn't happen, and that steps should be taken to keep elections from being fixed and to keep people from attempting to wrongly suppress the vote of others?

Of course. However, using exit polls - which are much more easily manipulated w/o being a federal crime in order to determine if an election was rigged or not just isn't reliable.

No election results are going to be tossed out, no one is going to go to jail, no one is going to prevail in court based upon "exit polling" alone.  It's one element that could suggest more investigation should be done in certain cases.

The solution is quite simple.
Voting machines with a paper record that can be hand counted if necessary.

You vote - hardware inside the machine records your vote on paper in such a way that two identical copies are made - like carbon paper. One copy is kept in machine, one copy is given to voter as a receipt they can look at to see that the paper trail in the machine reflects how they voted.

This isn't rocket science, and it baffles my mind that this isn't done.

It never will happen, not just because of politicians. Much of Middle America is frightful of the idea of dirty unwashed inner city masses being able to have a big influence at the polls, so they give tacit acceptance to all those tricks.

The problem is that a lot of people don't really care about their fellow citizens being able to vote.

Look at how many people have jumped on this "voter ID" bandwagon so fast, when everyone knows Mexicans are sneaking into this country for jobs, not to influence our elections.

[Tycho?] wrote:

resedit wrote:

Even if they randomly chose where to do the exit polls, they would be biased towards the random selection.

What?

I really didn't understand any of that discussion....I think someone was failing to take the margin of error into effect.

resedit wrote:

[Tycho?] wrote:

resedit wrote:

Even if they randomly chose where to do the exit polls, they would be biased towards the random selection.

What?

Flip a penny ten times. If you get 7 heads, it is a random result, but not a normal result.
Random selection does not guarantee a normal selection.

Now granted, flip a penny 100 times and it would be incredible to get 70 heads.

But the point is still that random selection can result in skewed data. As can biased selection.
In fact the laws of probability predict that random selection will sometimes bring out extremes. If you never get extremes, then your process is not really random.

Your logic is dizzying.

It's just the point that a random sample is not necessarily representative of the population, thus exit polls based upon a "random" sample are not necessarily going to be representative of the population.

If you get 7 heads out of 10 flips - random chance says that would happen, and the conclusion would be that the penny is biased towards head, when in fact - it may be a fair coin without bias to head or tail (or maybe even have a bias towards tail).

That is where increased sample size reduces the chances of false conclusions, flip a penny 100 times and you are much more likely to get an accurate measurement of the pennies bias then 10 flips.

But anyway - first you select where the exit polls will be done, and then there is further selection in who actually responds to the exit polls. I suspect many people do not want to bothered by them, and it may not be evenly distributed who does and who doesn't. In some areas, democrats may be more willing, in others, republicans may be more willing, in some areas - single moms may be more willing, in some areas, hispanics may be more willing. If the person conducting the exit poll is a hot chick, women may feel intimidated and men may be more willing.

Most of the time they are fairly accurate, but there are WAY too many variables to determine voter fraud based upon them.

Last edited by resedit (2006-11-01 11:52 am)

btw - does anyone have any stats as to how often exit polls agree with elections historically when there are surprise winners (IE not predicted by pre voting polls)?

It would be interesting to see if exit poll statistics are more or less accurate in surprise wins versus expected wins.

resedit wrote:

It's just the point that a random sample is not necessarily representative of the population, thus exit polls based upon a "random" sample are not necessarily going to be representative of the population.

My point is, you shouldn't confuse a completely random event (flipping a coin) with voters' choices.

And you shouldn't bring sample size into the arguement in such a simplistic fashion.  A respectable poll doesn't just throw that +/- of 3% (or whatever) on at the end, they use some good statistical methods to sort it out.  I never payed attention in statistics, but they do a LOT of math and logic to sort out what works, and my understanding is that its exhaustively tested.


Now, the type of people that answer exit polls does have some "selection bias," but I understand exit polls in America are historically pretty accurate.