Here's how Repub state pollsters work for Bush ( Election-Model results)

[TruthIsAll]

That's why I trust the Indies.

9/27 electoral-vote.com (includes Repub leaners Survey-USA,
Gallup, Mason-Dixon, Strategic Vision,Badger)
Kerry: 
255 EV, 
Proj. 49.92%, 
Win Prob 29%, 
Current Weighted National equiv: 
Kerry 45.20%, 
Bush 46.94%



9/23 Zogby/ARG (Independents)
Kerry: 
300 EV, 
Proj. 50.85%, 
Win Prob 91%, 
Current Weighted National equiv: 
Kerry 46.49%, 
Bush 46.24%


9/27 electoral-vote.com....................9/23 Zogby/ARG     
      
Kerry	Bush	electoral-vote.com		STATE	Kerry	Bush
39	55	Rasmussen-14		          AL	40	54
30	57	American Res. Group 		AK	30	57
43	54	Survey USA-3		          AZ	43	49
45	48	American Res. Group-3		AR	45	48
48	43	Strategic Vision (R)-3	CA	52	41
44	52	Survey USA-3		          CO	45	46
54	39	American Res. Group-3		CT	54	39
50	41	American Res. Group-3		DE	50	41
78	11	American Res. Group-3		DC	78	11
49	48	Rasmussen-7		          FL	46	45
42	53	Rasmussen-14		          GA	42	53
51	41	American Res. Group 		HI	51	41
30	59	American Res. Group 		ID	30	59
49	40	Market Shares-4		    IL	51	42
39	54	American Res. Group-5		IN	39	54
45	48	Opinion Dynamics-2		IA	45	47
35	57	American Res. Group-4 	KS	35	57
38	53	Bluegrass Poll		          KY	39	57
42	50	American Res. Group-5		LA	42	50
46	47	Survey USA-3		          ME	48	44
48	48	Survey USA		          MD	52	43
64	27	American Res. Group 		MA	64	27
48	44	Rasmussen-7		           MI	48	40
46	46	Rasmussen-7		          MN	47	45
42	51	American Res. Group-4		MS	42	51
44	50	American Res. Group-4 	MO	44	50
36	54	Mason-Dixon-3		          MT	36	54
30	61	American Res. Group 		NE	30	61
46	48	Gallup-4		           NV	45	47
46	46	Research 2000-4		     NH	46	46
47	43	Quinnipiac Univ.-4		NJ	50	42
43	47	Mason-Dixon-2		      NM	49	44
51	31	Sienna College-4		   NY	52	40
44	49	American Res. Group-4 	NC	44	49
33	62	American Res. Group 		ND	33	62
44	48	Opinion Dynamics-2		OH	46	48
33	64	Survey USA-3		          OK	38	55
47	48	Survey USA-3		          OR	47	45
48	45	Rasmussen-7		           PA	48	45
55	37	Survey USA-3		          RI	55	37
42	51	Rasmussen-7		           SC	42	51
39	58	American Res. Group 		SD	39	58
41	55	Survey USA-3		      TN	43	50
37	58	Survey USA-3		      TX	36	58
27	64	American Res. Group 		UT	27	64
50	40	American Res. Group 		VT	50	40
42	53	Survey USA-3		      VA	43	49
46	45	Strategic Vision (R)-3	WA	51	44
42	51	Gallup-4		            WV	46	46
38	52	Badger Poll-7		      WI	50.30	47.90
29	65	American Res. Group 		WY	29	65
43.63	48.84		Average                  44.65	47.82

[Heath.Hunnicutt]

1. You told me Nader doesn't matter, but...

You predict a 91% confidence of a 50.85% victory by Kerry, using Independent polls and a 2-party model.

Surely the amount of influence Nader is going to have will average out to at least 0.85%.

What happens when you change it to a realistic model that includes Kerry?

[TruthIsAll]

2. Its only Kerry vs. Bush that counts. Project Kerry, leave the rest..

You mean Nader, right.

Nader is a non-issue. He'll get less than 1%. Don't get lost in the trees. View the forest.

I am projecting Kerry. Leave the rest for Bush and Nader.

I only care how Kerry is doing versus Bush.
He is beating Bush in the projection of the pop. vote, that's what counts.

[Heath.Hunnicutt]

3. Are you using poll data that includes Nader, then?

When you write you are projecting Kerry and leaving the rest for Bush and Nader, are you saying that you are using poll results that somehow indicate the fraction of people who will vote for Kerry instead of either Bush or Nader?

Or are you using polling data from polls where only two choices (Bush/Kerry) were giving during the poll, even though the poll respondent lived in a state where Nader was or would be on the ballot?

Because when you write "Nader is a non-event," I don't believe you.

I understand Monte Carlo analysis somewhat, I understand statistics, probabilities, and models. Your model reminds me of that old saw, "There are lies, damn lies, and statistics."

Without accuraretly representing the race that is actually going to happen, you cannot accurately model its outcome.

The problem with your forest/trees analogy is that if there are 100 trees in the percentage forest, there is only 0.85% of one standing between us and Bush. Which side of the line is that middle tree on?

[TruthIsAll]

4. My response

You:
When you write you are projecting Kerry and leaving the rest for Bush and Nader, are you saying that you are using poll results that somehow indicate the fraction of people who will vote for Kerry instead of either Bush or Nader?

Me:
I use national and state poll data which includes Kerry, Bush and undecided/other. As a base case assumption, I allocate 60% of the latter group to Kerry.

You:
Or are you using polling data from polls where only two choices (Bush/Kerry) were giving during the poll, even though the poll respondent lived in a state where Nader was or would be on the ballot?

Me:
No, see above.

You:
Because when you write "Nader is a non-event," I don't believe you.

Me:
You don't believe me? You mean you don't agree with me. I said it. Nader pulled 2.7% last time. He will get under 1% this time and is not on the ballot in a number of states. Nader is not a concern. I know he pulled 96,000 votes in Florida. I know he was the difference in NH. But he is not going to be a factor this time.

You:
I understand Monte Carlo analysis somewhat, I understand statistics, probabilities, and models. Your model reminds me of that old saw, "There are lies, damn lies, and statistics."

Me:
You understand it "somewhat"? That's not good enough. I use the latest poll data, calculate national and state probabilities based on the MOE, with the data input to the Normal Distribution. I KNOW what I'm doing. You apparently don't.

You:
Without accurately representing the race that is actually going to happen, you cannot accurately model its outcome.

Me:
The best models do not clutter themselves up with factors which
have no discernible impact on the outcomes. Have you ever written financial, economic, engineering or political models? If you did, you would understand my point.

You:
The problem with your forest/trees analogy is that if there are 100 trees in the percentage forest, there is only 0.85% of one standing between us and Bush. Which side of the line is that middle tree on?

Me:
As I said before, KISS. Let's use a simple case.

Assume the following in a State poll:
Kerry 46
Bush 46
Nader 2
Other 6

I project Kerry = 46 + .6*(100-46-46)= 46 +.6*8= 46+4.8 = 50.8
I assume Kerry takes votes from both Nader and other (undecided)

I give Bush (and other) the rest = 46 +.4*8= 46 + 3.2 = 49.2

So, yes, the projected Bush total includes Nader (.80)

The final result is:

Kerry 50.8
Bush 48.4
Nader .8

As I said, Nader is a non-factor. My analysis is actually CONSERVATIVE, because Bush's total includes Nader's .80%

Hope that you're satisfied with this.

TIA

[Heath.Hunnicutt]

5. Well, your answer reveals your mistake

You state that you are projecting that 60% of people who say they will vote for Nader will actually vote for Kerry instead. Do you have any evidence to back that up other than your optimism?

Look, I'm just going to ignore the pejorative aspects of your reply. I understand you want Kerry to win, and we all want him to be winning. I also know what a normal distribution is, and other distributions of random variables. I do know how Monte Carlo analysis works, and I am certain it depends on the assumptions you feed into it.

I am not a professional statistician, and I sense that you aren't, either. However, I do have professional statisticians at my disposal, if you'd like some experts to look at this.

The problem is not with your math. It is with the way you are using it.

TruthIsAll:
I project Kerry = 46 + .6*(100-46-46)= 46 +.6*8= 46+4.8 = 50.8
I assume Kerry takes votes from both Nader and other (undecided)

(Emphasis mine)

That was my point exactly. Let's take the state poll you used as an example. If you took the 60% of undecided voters only from actually undecided poll respondents, and left Nader's vote as reported, you would get:

Kerry = 46 + (60% of 6) = 49.6% Kerry
Bush = 46 + (40% of 6) = 48.4% Bush
Nader = .............. = 2% Nader

In short, you moved 1.2% of the vote from Nader to Kerry because you wanted to. Your model would make just as much sense if you asserted that Nader was going to get 10% of the poll respondents who said they are voting for Kerry. If you want to view the world as Kerry and non-Kerry, this changes your result to 49.6% Kerry, 50.4% non-Kerry.

There is no reason to believe the Nader voters are going to act like the undecideds.

To rehash, here are your assumptions:

60% of undecided poll respondents will vote for Kerry.

60% of poll respondents who claimed they will vote for Nader also will actually vote for Kerry.


The first is plausible (although why choose 60% and not 50%), but the second is not.

[Heath.Hunnicutt]

6. And why I care...

I'd like to add: at first I couldn't have been more excited to see a statistical model being used to predict the election outcome. What with electoral-vote.com swinging in the breeze, something based on science would be great.

The problem with a mathematical model is that you have to feed it assumptions. You are feeding yours an assumption I don't believe, so I don't believe your results. I wish I did, because the whole reason I read over your post carefully is that I thought it had inherent value.

[DemocratSinceBirth]

7. As You Are So Making Clear Polls Are Mere Snapshots...

The polls that count and on which the credibility of the pollsters rest is the final pre-election polls...

IMHO, the best approach is to do a poll of polls that way the "negative" and "positive" outliers would cancel out each other...

[Heath.Hunnicutt]

8. No, TruthIsAll made that clear

The original comparison TIA made is valid: some polls are biased and give different results than others.

His approach to modelling this election should be fairly accurate if it is based on accurate assumptions.