ENTER YOUR OWN EXIT POLL MODEL INPUTS TO HELP DEBUNK rBr.

[TruthIsAll]

DUers: Design your own exit poll optimization ANALYSIS.
I will run the EXIT POLL RESPONSE OPTIMIZATION MODEL based on your input.

Objective:
By running a number of "constrained" optimizations, we can gain insight into the feasibility of the famous rBr hypothesis, which is that Bush voters were less likely to respond to exit pollsters.

The model will determine values for the constrained variables given below in order to derive a target Kerry/Bush percentage split using aggregate exit poll response data.

TARGET GOAL-SEEKING:
1) Enter a target percentage for Kerry or Bush:
For example,
Bush 48.80%
Kerry 51.20%


EXIT POLL RESPONSE INPUT CONSTRAINTS
There were 1250 precincts, ranging in partisanship from strong Bush to strong Kerry:

Strong Bush > Strong Kerry
40 415 540 165 90

2) Enter your range estimate of Kerry's win percentage (KP) for each precinct group.

For example,
Kerry precinct win percentages ranges:
KMin 0% 20% 40% 60% 80%
KMax 20% 40% 60% 80% 100%

3) Enter a range of response rates for each precinct group.
(note that the optimization will determine the rates necessary to match the weighted average 53% exit poll response).

For example, constrain the response rates to a U-shaped curve:
Min 40% 45% 50% 45% 40%
Max 70% 65% 55% 65% 70%

Or an inverted U-shape:
Min 50% 55% 40% 55% 50%
Max 55% 65% 75% 65% 55%

Or go for broke (no constraints):
Min 0% 0% 0% 0% 0%
Max 100% 100% 100% 100% 100%

4) Enter your own Within Precinct Error (WPE) across precincts (you can try all zeros) or let the optimizer match the the reported E-M precinct WPE's:

WPE -10.00% -6.10% -8.50% -5.90% 0.30%



Notes:
a) The optimization will determine the AGGREGATE precinct response rates subject to the constant 53% weighted average constraint.

b) Similarly, the model will determine feasible precinct Alphas (Kerry response/Bush response ratio) to produce the 1.12 weighted average constraint.

[LightningFlash]

1. What I propose is to analyze all the percentage columns.

It's really quite simple. Take each confirmed voting percentage and totality from the U.S. Cencus Bureau's results, and line them up with the 12:33 AM Exit Poll results side by side to the 1:41 AM Exit Poll results.

The comparative analysis will reveal just how severe the bogus early AM weighting was and in all likelyhood the amount of fraud, or it will reveal the opposite and show just how bogus the midnight weighting was and some randomly distributed error.

I'm sure the scientists can manage that. I await their reported findings eagerly.

[TruthIsAll]

2. Did you see this?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x373187

Let's not get off track. Do you have a set of input constraints you would like to run through the model?

[LightningFlash]

3. Just one.

Thanks, I had seen the numbers but I needed a real comparison.

What we need is a breakdown of this data by precinct somehow to get a much more finite idea. My one added measure to this analysis is to increase the margin of sampling error.

Let's say by random chance the "sampling error" was almost .65%. Run it through the slotted charts, how does the affected outcome actually change?

Thanks for the hard work.

[TruthIsAll]

4. That's all the data we have, so we must do an aggregate analysis

in lieu of the individual precinct data.

The failure to release the detailed precinct data is exactly the problem and is a smoking gun in and of itself.

[socal_dan]

5. Why only 5 data points to debunk rBr?

Why are there only 5 data points used to debunk the rBr?

The rBr - reluctant Bush responder - is the hypothesis that the difference between exit polls and actual results in the November 04 presidential election were due to Bush responders being either less reluctant to respond to exit pollers or the exit pollers being less reluctant to ask Bush voters.

Repitively USCV has used a 5 point plot to attempt to debunk the rBr where they show precincts with high amounts of Bush responders had as high rates of being polled by exit pollers as precincts with high amounts of Kerry responders.

The 5 groups of precincts are grouped by "20% or less Bush votes".. "20-40% Bush votes'... etc.

Why only 5? Looking at the 5 point plot, it looks reasonably flat within a margin of error or so... but I think if it were divided into a 10 or 20 point plot (by increments of 10% or 5%) the eventual result would look clearer (naturally, removing points for which there were no precincts, EG - if there were no precincts with less than 5% Bush votes).

This way random errors would look more random - as bulges here and there in the plots - whereas any correlation between the number of Bush votes and number of Bush exit poll responders would stick out as a diagonal line.

[TruthIsAll]

6. I agree, but 5 points is all I have to work with.

I can easily extend the model to include more precinct groupings, but I am "constrained" to the aggregate data that USCV has used.

In any case, read Baiman and Febble for using simulation with the aggregates.

Even with the limited five partisanship data groupings, the optimization (goal-seeking) model returns some very interesting results. We just need to play with various assumptions to test their feasibilities.

So far, no one has responded with some inputs. Why don't you be the first?

[socal_dan]

7. Okay, just for fun.

I was actually trying to draw out the more datapoints I want - but I can't top post yet.

TARGET GOAL-SEEKING:
1) Enter a target percentage for Kerry or Bush:
Bush 51.25
Kerry 48.75

That's the CNN results Bush/(bush+kerry) and kerry/(bush+kerry). Those two are all over each other.. har har har.

2) Enter your range estimate of Kerry's win percentage (KP) for each precinct group.

Alright, I have no good idea on this.. so I'll hit center:
10% 30% 50% 70% 90%

3) Enter a range of response rates for each precinct group.

I don't know, this seems reasonable.. eliminating any possibly outliers while allowing for a reasonable rBr:

Min 30% 30% 30% 30% 30%
Max 70% 70% 70% 70% 70%

4) Go with the reported rates

WPE -10.00% -6.10% -8.50% -5.90% 0.30%

[TruthIsAll]

8. Here goes.

EXIT POLL RESPONSE OPTIMIZATION MODEL				
				
Objective: Determine values of constrained variables
required				
to derive a target Kerry/Bush percentage split using aggregate
exit poll response data.				
				
TARGET INPUT				
Bush 	51.25%			
Kerry	48.75%			
				
Precinct Range Variables:				
1-Response Rates constrained to equal the input weighted
average within the (Low, High) range				
2-Kerry win percentages constrained to the precinct interval
range (Min, Max)				
3-Alpha (K/B response rate) constrained to the weighted
average 				
WPE across precincts were constrained to match E-M				
				
				
								
RESPONSE INPUT CONSTRAINTS								
								
53.0%	Wtd Average Response rate 							
1.120	Wtd Average Alpha (K/B )							
								
1250	Precincts	Strong Bush	 Strong Kerry			
	Kerry	40	415	540	165	90		
	Min	10%	30%	50%	70%	90%		
	Max	10%	30%	50%	70%	90%		
								
ResponseMin	30%	30%	30%	30%	30%		
Rate	Max	70%	70%	70%	70%	70%		
								
-6.77%	WPE  	-10.00%	-6.10%	-8.50%	-5.90%	0.30%		
	Act	-10.00%	-6.10%	-8.50%	-5.90%	0.30%		
								
								
OPTIMIZER OUTPUT								
								
Precinct	40	415	540	165	90		Avg
Response:	30.0%	70.0%	36.1%	70.0%	55.0%		53.0%
								
Alpha(K/B):	0.93	1.18	1.07	1.07	1.30		1.12
								
								
Kerry wins	51.99%	of exit poll responders						
Bush wins	48.01%							
Bush needs	55.57%	of refusers to match his vote						
								
Target								
Kerry 	48.75%							
Bush	51.25%							
								
								
	OPTIMIZATION MODEL							
								
		Strong Bush			Strong Kerry			
Precincts	40	415	540	165	90		1250
	K Min	10%	30%	50%	70%	90%		47.6%
	K Max	10%	30%	50%	70%	90%		47.6%
								
	Kerry%0%	30%	50%	70%	94%		47.6%
	Alpha	0.93	1.18	1.07	1.07	1.30		1.12
								
	Resp%	30.0%	70.0%	36.1%	70.0%	55.0%		53.0%
	Resp	12	291	195	115	49		663
								
	K Resp 	0	103	104	87	61		355
	B Resp	12	187	91	29	0		319
	K Resp	0.0%	35.5%	53.4%	75.1%	100.0%		52.0%
	B Resp	100.0%	64.5%	46.6%	24.9%	0.0%		48.0%
								
	K Ref	0	32	161	32	29		254
	B Ref 	28	93	184	17	0		322
								
	TotalK 	0	135	266	119	90		609
	TotalB 	40	280	274	46	0		641
								
	B Total	100.0%	67.5%	50.8%	27.9%	0.0%		51.2%
	K Total	0.0%	32.5%	49.2%	72.1%	100.0%		48.7%
								
	Kv-Bv	-100.0%	-35.1%	-1.6%	44.3%	100.0%		-2.5%
	Kp-Bp	-100.0%	-29.0%	6.9%	50.2%	100.0%		4.0%
	WPE	0.0%	-6.1%	-8.5%	-5.9%	0.0%		-6.5%
								
	B Exit%	100.0%	64.5%	46.6%	24.9%	0.0%		48.0%
	B Ref%	100.0%	74.7%	53.2%	34.8%	0.0%		55.6%
								
	WPE  	-10.00%	-6.10%	-8.50%	-5.90%	0.30%		-6.77%
	Tol	10.00%	0.01%	0.01%	0.01%	0.30%		0.29%

[LightningFlash]

9. Astonishing.....

Bush would require nearly twice to three times the amount of refusers/liars in every Cencus taken precinct to actually win the election. That is stunning....Why oh why does Mitofsky not release the precinct data?

:wow:

[socal_dan]

10. What - is it Mitofsky that won't give me my 10-20 point plot?

That is... instead of dividing exit poll response rates among precincts divided by 0-20% bush result, 20-40%... I want MORE POINTS!

Is Mitofsky the rotten egg in this computation?

What's he scared of!

[LightningFlash]

11. Re-submit this following set of numbers.

TARGET GOAL-SEEKING:
1) Enter a target percentage for Kerry or Bush:
Bush 51.25
Kerry 48.75


2) Enter your range estimate of Kerry's win percentage (KP) for each precinct group.

I will assume the following increased response rate (Based on Kerry voters being "less" reluctant)
70%

3) Enter a range of response rates for each precinct group.


Min 50% 50% 50% 50% 50%
Max 70% 70% 70% 70% 70%

4) Go with the reported rates

WPE -10.00% -6.10% -8.50% -5.90% 0.60%

Just for kicks I have doubled the WPE basis of sampling error. Please, plug the numbers through and let's see what the outcome is.

[TruthIsAll]

12. Results

EXIT POLL RESPONSE OPTIMIZATION MODEL			
			
Objective: Determine values of constrained variables required
to derive a target			
Kerry/Bush percentage split using aggregate exit poll response
data.			
			
TARGET INPUT			
Bush 	51.25%		
Kerry	48.75%		
			
Precinct Range Variables:			
1-Response Rates constrained to equal the input weighted
average within the (Min, Max) range			
2-Kerry win percentages constrained to the precinct interval
range (Min, Max)			
3-Alpha (K/B response rate) constrained to the weighted
average 			
4- WPE across precincts default to E-M actuals			
			
RESPONSE INPUT CONSTRAINTS			
							
53.0%	Wtd Average Response rate 						
1.120	Wtd Average Alpha (K/B )						
							
							
1250	Strong Bush			Strong Kerry			
Prcts	40	415	540	165	90		Avg
							
Kerry Win Pct:							
Min	10%	20%	40%	60%	80%		37.92%
Max	20%	40%	60%	80%	100%		57.60%
Response rates:							
Min	50%	50%	50%	50%	50%		50.00%
Max	70%	70%	70%	70%	70%		70.00%
							
WPE  	-10.0%	-6.1%	-8.5%	-5.9%	0.6%		-6.75%
Actual	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
OPTIMIZER OUTPUT							
							
Precincts							
Number	40	415	540	165	90		
Resp.	50.0%	59.0%	50.0%	50.0%	50.0%		53.0%
K/B	1.41	1.20	1.08	1.04	1.00		1.12
							
EXIT POLL							
Kerry 	52.13%	of exit poll responders					
Bush	47.87%						
Bush needed	55.07%	of refusers to match his vote					
							
Target							
Kerry 	48.75%						
Bush	51.25%						

							
							
OPTIMIZATION MODEL							
							
Categ.	HighB	Bush	Even	Kerry	HighK	.	Total
Prcts	40	415	540	165	90		1250
Kerry	14%	20%	55%	78%	98%		48.31%
							
K/B	1.41	1.20	1.08	1.04	1.00		1.12
							
RESPONDERS							
Total	20	245	270	82	45		663
Pct	50.0%	59.0%	50.0%	50.0%	50.0%		53.00%
							
Kerry	4	59	161	67	44		335
Pct	20.2%	24.1%	59.6%	81.0%	97.9%		52.13%
Bush	16	186	109	16	1		328
Pct	79.8%	75.9%	40.4%	19.0%	2.1%		47.87%
							
REFUSERS							
Kerry	2	28	138	62	44		275
Bush	18	142	132	21	1		313
Pct	89.8%	83.4%	48.9%	24.9%	1.5%		55.1%
							
VOTE							
Kerry	6	87	299	129	88		609
Pct	15.2%	21.0%	55.4%	78.1%	98.2%		48.75%
Bush	34	328	241	36	2		641
Pct	84.8%	79.0%	44.6%	21.9%	1.8%		51.25%
							
WPE							
Kv-Bv	-69.5%	-58.0%	10.7%	56.1%	96.3%		-2.50%
Kp-Bp	-59.5%	-51.9%	19.2%	62.1%	95.7%		4.25%
WPE	-10.0%	-6.1%	-8.5%	-5.9%	0.6%		-6.75%
-Input	-10.0%	-6.1%	-8.5%	-5.9%	0.6%		-6.75%
Diff	0.0%	0.0%	0.0%	0.0%	0.0%		0.00%

[LightningFlash]

14. Haha....Thanks for proving it again TIA.

I already suspected as much, that even a double outside the Cencus Bureau's margin of error would not be able to account for a Bush win.

Of course, even though it is hilarious, there are those who don't do "statistics" who simply aren't grasping it. Up to the top for them.

Apparently, Bush still requires exactly 55% of all voters to REFUSE saying they voted for him! That's almost the same amount of Iraqis who hate the U.S. army. ;)

[kster]

13. kick.nt