JFern's "Statistics"

[J. J.]

On another thread, our old friend JFRAUD has raised statistal questions unreated to the topic.  Though he has, charateristically, declined to address the topics, and has declined to start a separate thread, despite repeated requests.  So, it falls to me to address them in the appropriate forum.

JFRAUD asked:

https://uselectionatlas.org/FORUM/index.php?topic=20462.195

You must be logged in to read this quote.

There are several problems.  First, if the sample group were the Kerry delegates to the Democratic National Convention, this would indicate that Kerry was in big trouble as he should win all of them.  The "expected value" here woulld be 1000 for Kerry and zero for Bush.  In that case, with the expected value being 1000 Kerry, there would be no statistical correlation.

The second problem is trying to tie these factors to another event.  Is it because the economy is perceived to be bad, the war in Iraq, voters don't trust Dick Cheney, or that the X-Files are not in first run any more?

The significance here is not how well this describes the population, but how far off the expected value is this number.

[○∙◄☻¥tπ[╪AV┼cVê└]

On another thread, our old friend JFRAUD has raised statistal questions unreated to the topic.  Though he has, charateristically, declined to address the topics, and has declined to start a separate thread, despite repeated requests.  So, it falls to me to address them in the appropriate forum.

JFRAUD asked:

https://uselectionatlas.org/FORUM/index.php?topic=20462.195

You must be logged in to read this quote.

There are several problems.  First, if the sample group were the Kerry delegates to the Democratic National Convention, this would indicate that Kerry was in big trouble as he should win all of them.  The "expected value" here woulld be 1000 for Kerry and zero for Bush.  In that case, with the expected value being 1000 Kerry, there would be no statistical correlation.

The second problem is trying to tie these factors to another event.  Is it because the economy is perceived to be bad, the war in Iraq, voters don't trust Dick Cheney, or that the X-Files are not in first run any more?

The significance here is not how well this describes the population, but how far off the expected value is this number.

I have addressed the topics, but I'm bringing this up to show how intellectually dishonest you are, and why I shouldn't waste my time arguing something less black and white then statistics, where you're clearly wrong.

I said a RANDOM poll of people who are likely voters. If you're prefer, lets change it to an actual 1000 person random sample of people who actually voted. Set up the Diebold machines to randomly recount 1000 votes cast (suppose everyone votes on Diebold). Nice attempt to distract from the real issue, but I'm not letting you get away with that intellectual dishonesty.

OK, so I have a 1000 random vote sample of the 120+ million vote sample (I would not have to word this way if you weren't so focused on trivialness in an attempt to avoid answering the actual question I posed). 940 of those are for Kerry, 60 are for Bush. Is that a statistically significant lead for Kerry?

[○∙◄☻¥tπ[╪AV┼cVê└]

Or if we want a less precise and more common sense wording, let's say a good polling firm like SUSA comes out with a poll with Kerry at 94% and Bush 6%.  Would that be a statistically significant lead by Kerry?

[J. J.]

Or if we want a less precise and more common sense wording, let's say a good polling firm like SUSA comes out with a poll with Kerry at 94% and Bush 6%.  Would that be a statistically significant lead by Kerry?

The lead itself would not be statically valid, as you would have to look at the poll itself.  You are equating the result with the validity of the poll. 

Let me put it this way, if I flicked a light switch 100 times (and after each test, checked the bulb to make sure that it worked), we would expect, if the switch controlled the bulb, to it work 100 time out of 100.  Likewise, I would expect the bulb not to light whan I did not flick the switch.  If the bulb goes on when don't flick the switch, or doesn't go on when I do flick and it does it enough times, I better stop thinking that the switch controls the light.

[○∙◄☻¥tπ[╪AV┼cVê└]

Or if we want a less precise and more common sense wording, let's say a good polling firm like SUSA comes out with a poll with Kerry at 94% and Bush 6%.  Would that be a statistically significant lead by Kerry?

The lead itself would not be statically valid, as you would have to look at the poll itself.  You are equating the result with the validity of the poll. 

Let me put it this way, if I flicked a light switch 100 times (and after each test, checked the bulb to make sure that it worked), we would expect, if the switch controlled the bulb, to it work 100 time out of 100.  Likewise, I would expect the bulb not to light whan I did not flick the switch.  If the bulb goes on when don't flick the switch, or doesn't go on when I do flick and it does it enough times, I better stop thinking that the switch controls the light.

I'm assuming a completely random sample. Whether Kerry has a statistically significant lead is determined by:

 Assume that Kerry doesn't lead. The best case is then Kerry and Bush are tied at 50%. Does the poll differ statistically from this?

You can let your reject p=5% or p=0.1%, you can do one-sided testing, or two-sided testing, but none of that matters in this example. Any non absurdly small reject probability  p is going to make you decide that 94% - 6% is a statistically significant lead by Kerry.

[○∙◄☻¥tπ[╪AV┼cVê└]

Random forum poster: "Holy, sh**t look at this new SUSA poll showing Kerry up 94-6 nationwide!"

J.J: "Who cares, that's not a statistically signifcant lead."

[J. J.]

Random forum poster: "Holy, sh**t look at this new SUSA poll showing Kerry up 94-6 nationwide!"

J.J: "Who cares, that's not a statistically signifcant lead."

No, but I might say, "That's Zogby."

In answer to your question, you would be looking at the results of the poll directly.  The statistics question is not what are results of the poll, but what is likelthood that the poll is wrong.  Even if the pollster do does everything right in poll construction, and impliments it perfectly, there is still a chance that a certain percentage that the poll itself is going to be statistically invalid. 

In other words, let's that this was a series of 100 polls, all conducted at the same time, and all with the same implimentation and construction.  Also assume that we know the actual result is 51% Bush, 49% Kerry.  Would we expect to get a really bad poll like this one as one of those 100?  Yes.

I think that happened during the 2004 campaign the day that the Vorlon changed his name to "The Shadows."

I think we can add a bit to one rule to Silent Hunter's list:

https://uselectionatlas.org/FORUM/index.php?topic=20676.msg442894#msg442894


"6. Remember a little thing called 'margin of error,' " and never trust just one poll.

[○∙◄☻¥tπ[╪AV┼cVê└]

Random forum poster: "Holy, sh**t look at this new SUSA poll showing Kerry up 94-6 nationwide!"

J.J: "Who cares, that's not a statistically signifcant lead."

No, but I might say, "That's Zogby."

In answer to your question, you would be looking at the results of the poll directly.  The statistics question is not what are results of the poll, but what is likelthood that the poll is wrong.  Even if the pollster do does everything right in poll construction, and impliments it perfectly, there is still a chance that a certain percentage that the poll itself is going to be statistically invalid. 

In other words, let's that this was a series of 100 polls, all conducted at the same time, and all with the same implimentation and construction.  Also assume that we know the actual result is 51% Bush, 49% Kerry.  Would we expect to get a really bad poll like this one as one of those 100?  Yes.

I think that happened during the 2004 campaign the day that the Vorlon changed his name to "The Shadows."

I think we can add a bit to one rule to Silent Hunter's list:

https://uselectionatlas.org/FORUM/index.php?topic=20676.msg442894#msg442894


"6. Remember a little thing called 'margin of error,' " and never trust just one poll.

If the actual population is 51% Bush, 49% Kerry, the odds are extremely small that one of 100 polls would give something like 94% Kerry, 6% Bush. In fact, the MOE is pretty small on a 94% Kerry, 6% Bush poll (it would be 0.75%)

Clearly, if you think that if the race is close, and a random sample of 1000 giving 94% Kerry, 6% Bush is likely, you're not understanding the statistics. Check all of the nationwide opinion polls from the 2004, and tell me if Kerry or Bush was at only 6% in any of them.

[Jake]

There was a poll that had Kerry near 50% in Utah.

[○∙◄☻¥tπ[╪AV┼cVê└]

There was a poll that had Kerry near 50% in Utah.

If it wasn't a joke, link to it.

[Alcon]

There was a poll that had Kerry near 50% in Utah.

Uh...no there wasn't, at least not that I'm aware of. It must be a very minor one if so.

The Bush numbers ranged from 64% (American Research Group, 2/19/04 and 9/13/04) to 69% (Salt Lake Tribune, 9/27/04 and Dan Jones 10/28/04).

The Kerry numbers ranged from 21% (Salt Lake Tribune, 9/27/04) to 31% (American Research Group, 2/19/04).

Any poll near 50% would have been laughed at.

[Sam Spade]

There was a poll that had Kerry near 50% in Utah.

Uh...no there wasn't, at least not that I'm aware of. It must be a very minor one if so.

The Bush numbers ranged from 64% (American Research Group, 2/19/04 and 9/13/04) to 69% (Salt Lake Tribune, 9/27/04 and Dan Jones 10/28/04).

The Kerry numbers ranged from 21% (Salt Lake Tribune, 9/27/04) to 31% (American Research Group, 2/19/04).

Any poll near 50% would have been laughed at.

It was a Democrat-sponsored poll.  It was on electoral-vote.com (which used even partisan polls).

It was pretty funny, I must admit.

[Alcon]

There was a poll that had Kerry near 50% in Utah.

Uh...no there wasn't, at least not that I'm aware of. It must be a very minor one if so.

The Bush numbers ranged from 64% (American Research Group, 2/19/04 and 9/13/04) to 69% (Salt Lake Tribune, 9/27/04 and Dan Jones 10/28/04).

The Kerry numbers ranged from 21% (Salt Lake Tribune, 9/27/04) to 31% (American Research Group, 2/19/04).

Any poll near 50% would have been laughed at.

It was a Democrat-sponsored poll.  It was on electoral-vote.com (which used even partisan polls).

It was pretty funny, I must admit.

The only electoral-vote.com Utah polls I can find in their CSV file is 64-69% Bush. I can't understand.

[○∙◄☻¥tπ[╪AV┼cVê└]

It was a Democrat-sponsored poll.  It was on electoral-vote.com (which used even partisan polls).

It was pretty funny, I must admit.

I don't see it. I see a 69-24, a 68-23, a 64-27, and a 65-25 here:
http://www.electoral-vote.com/2004/info/allpolls.csv

[J. J.]

The thing is, even if the pollster does everything right, he's still going to get one absolutely wrong result out of the margin of error in X amount of polls.  Now, with a sample size of 1000, X will be lower than if the sample size is 300, but it's still going to happen.

[○∙◄☻¥tπ[╪AV┼cVê└]

The thing is, even if the pollster does everything right, he's still going to get one absolutely wrong result out of the margin of error in X amount of polls.  Now, with a sample size of 1000, X will be lower than if the sample size is 300, but it's still going to happen.

Yeah, well he's never going to get 94-6 with a sample of 1000 when it was really a tie, and that's why 94-6 is a very statistically significant lead.

[J. J.]

The thing is, even if the pollster does everything right, he's still going to get one absolutely wrong result out of the margin of error in X amount of polls.  Now, with a sample size of 1000, X will be lower than if the sample size is 300, but it's still going to happen.

Yeah, well he's never going to get 94-6 with a sample of 1000 when it was really a tie, and that's why 94-6 is a very statistically significant lead.

That you cannot say, statistically.

We are not referring to margin of error directly, but a problem that the poll will just be wrong.   When at a poll, statistically, it is accurate to say, this is the corect number, within the margin of error, X number of times out of 100.  X is usually 95 to 99.

[○∙◄☻¥tπ[╪AV┼cVê└]

The thing is, even if the pollster does everything right, he's still going to get one absolutely wrong result out of the margin of error in X amount of polls.  Now, with a sample size of 1000, X will be lower than if the sample size is 300, but it's still going to happen.

Yeah, well he's never going to get 94-6 with a sample of 1000 when it was really a tie, and that's why 94-6 is a very statistically significant lead.

That you cannot say, statistically.

We are not referring to margin of error directly, but a problem that the poll will just be wrong.   When at a poll, statistically, it is accurate to say, this is the corect number, within the margin of error, X number of times out of 100.  X is usually 95 to 99.

With a 1000 sample poll, the margin of error is always less than 1.582%.
We have 96-4, which is a z-score of 27.8.

If you know anything about z-scores you'll realize there's no way you'll ever get that.

[J. J.]

The thing is, even if the pollster does everything right, he's still going to get one absolutely wrong result out of the margin of error in X amount of polls.  Now, with a sample size of 1000, X will be lower than if the sample size is 300, but it's still going to happen.

Yeah, well he's never going to get 94-6 with a sample of 1000 when it was really a tie, and that's why 94-6 is a very statistically significant lead.

That you cannot say, statistically.

We are not referring to margin of error directly, but a problem that the poll will just be wrong.   When at a poll, statistically, it is accurate to say, this is the corect number, within the margin of error, X number of times out of 100.  X is usually 95 to 99.

With a 1000 sample poll, the margin of error is always less than 1.582%.
We have 96-4, which is a z-score of 27.8.

If you know anything about z-scores you'll realize there's no way you'll ever get that.

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

[ATFFL]

If the actual population is 51% Bush, 49% Kerry, the odds are extremely small that one of 100 polls would give something like 94% Kerry, 6% Bush. In fact, the MOE is pretty small on a 94% Kerry, 6% Bush poll (it would be 0.75%)

Are you saying that the MOE is determined by the results?  Or did you mean to say the chance of a poll showing 94-6 when the actual population is 49-51 is .75%?

Also, what happened on Electoral Vote.com is that when ARG came out with their polls the method the site used predicted a Kerry win.  All they did was assume that whatever movement occured from the last two polls would continue indefinately.

[○∙◄☻¥tπ[╪AV┼cVê└]

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

The error on a sample of 1000 is a bit over 3%. That determines statistical significance. If you don't understand that, sign up for freshman statistics.

[J. J.]

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

The error on a sample of 1000 is a bit over 3%. That determines statistical significance. If you don't understand that, sign up for freshman statistics.

You seem to be talking about the margin of error within the sample size, the maxim percetage that the numbers within the poll will vary.  I'm addressing the possibility that the sample was outside the second or third standard deviation (95% and 99%, respectively).

[○∙◄☻¥tπ[╪AV┼cVê└]

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

The error on a sample of 1000 is a bit over 3%. That determines statistical significance. If you don't understand that, sign up for freshman statistics.

You seem to be talking about the margin of error within the sample size, the maxim percetage that the numbers within the poll will vary.  I'm addressing the possibility that the sample was outside the second or third standard deviation (95% and 99%, respectively).

You're hopelessly confused. Time for you to read some Statistics 101 like this:
http://www.measuringusability.com/sample_old.htm

Of course, knowing you, you'll miss the point, again.

[J. J.]

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

The error on a sample of 1000 is a bit over 3%. That determines statistical significance. If you don't understand that, sign up for freshman statistics.

You seem to be talking about the margin of error within the sample size, the maxim percetage that the numbers within the poll will vary.  I'm addressing the possibility that the sample was outside the second or third standard deviation (95% and 99%, respectively).

You're hopelessly confused. Time for you to read some Statistics 101 like this:
http://www.measuringusability.com/sample_old.htm

Of course, knowing you, you'll miss the point, again.

Here is a link from the site you posted:

http://www.surveysystem.com/sscalc.htm

Assuming 120,000,000 million voters, you could get a result within 3% of the actual amount, 95% of the time, with a sample size of 1067.  To illustrate the difference, to get a result within 1% of the actual amount, 99% of the time, you would need a sample size of 16639.

Here is the actual explanation from the site:

You must be logged in to read this quote.

That's pretty much what I've been talking about and what Tedrick has been talking about.  It's from the link on the site that you quoted.  Why can't you understand it?

[○∙◄☻¥tπ[╪AV┼cVê└]

The problem is, you cannot say that the 94% is significant or not.  What the result is not going to determine if the poll itself is statistically significant. 

The error on a sample of 1000 is a bit over 3%. That determines statistical significance. If you don't understand that, sign up for freshman statistics.

You seem to be talking about the margin of error within the sample size, the maxim percetage that the numbers within the poll will vary.  I'm addressing the possibility that the sample was outside the second or third standard deviation (95% and 99%, respectively).

You're hopelessly confused. Time for you to read some Statistics 101 like this:
http://www.measuringusability.com/sample_old.htm

Of course, knowing you, you'll miss the point, again.

Here is a link from the site you posted:

http://www.surveysystem.com/sscalc.htm

Assuming 120,000,000 million voters, you could get a result within 3% of the actual amount, 95% of the time, with a sample size of 1067.  To illustrate the difference, to get a result within 1% of the actual amount, 99% of the time, you would need a sample size of 16639.

Here is the actual explanation from the site:

You must be logged in to read this quote.

That's pretty much what I've been talking about and what Tedrick has been talking about.  It's from the link on the site that you quoted.  Why can't you understand it?

Yep, I had a good feeling that you'd miss the point.

If we have a sample of 1000, the 95% confidence interval is around +-3%. That means if Kerry has a lead of 6 or fewer points, it's not a statistically significant lead. If Kerry has a lead of 88 points (94-6), then it's quite statistically significant.

I know a lot more statistics than you, you are a fraud for blaming your lack of knowledge of statistics on my supposedly being wrong.