9/28 ELECTION MODEL (TIA): Obama@ 340EV; state/nat’l polls, projections, win probs & fraud scenarios

[tiptoe]

2008 ELECTION MODEL
A  Monte Carlo  Electoral  Vote  Simulation

TruthIsAll

Updated: September 28

Press REFRESH after linking to a graph to view the latest update

Chart   State Poll Aggregate + Projection Trend

Chart   National 5-Poll Moving Average Projection

Chart   State vs. National: Vote Share Projection Trends

Chart   Battleground-State Polls

Chart   Battleground-State Win Probability

Chart   Obama Electoral Vote Simulation Frequency

Chart   Electoral Vote + Win Probability Trend

Chart   Electoral Vote + Projected Vote Share Trend

Chart   Undecided Voter Allocation + Win Probability

Chart   Monte Carlo Electoral Vote Simulation Trials

 2008 Election Model Fraud Analyzer 

Uncounted  &  Switched Votes

Chart   Effect on Obama Projected Vote Share

Chart   Effect on Obama Projected Electoral Vote

 

“I’m glad that TIA raises the questions. A Top 10 list.”  — Sancho

 

This
State
National
State
National
Monte  Carlo
Simulation

Update
Poll
5-Poll
2-party
2-party
Expected

9/28/2008
Aggregate
Average
Projection
Projection
Electoral Vote

Obama
McCain
 47.96 (51.40) 
 45.34 (48.60) 
 48.60 (53.29) 
 42.60 (46.71) 
51.98
48.02
53.88
46.12
 340 
 198 

    

15-Poll

End

Sample

Poll

NATIONAL MODEL

 

Pre   Undecided Voter Allocation
5-Poll Mov Avg

 

2-Party Projection  (60% UVA)
5-Poll Mov Avg

Trend
                            
Gallup
Rasmussen
Hotline/FD
Research2k
CBS/NYT

ABC/WP
NBC/WSJ
LAT
FOX News
Ipsos

CNN
Quinnipiac
CBS/NYT
Zogby
Ipsos

Registered V
vs     Likely V
Poll Averages

Date
            
9/27
9/27
9/27
9/25
9/24

9/22
9/22
9/22
9/22
9/22

9/21
9/16
9/16
9/15
9/15

Size   
              
2719 RV
3000 LV
914 RV
1100 LV
844 RV

780 LV
838 LV
1085 RV
900 RV
923 RV

697 LV
987 LV
800 LV
1008 LV
1046 RV

RV avg
LV avg
Total

MoE
         
1.88%
1.79%
3.24%
2.95%
3.37%

3.51%
3.39%
2.98%
3.27%
3.23%

3.71%
3.12%
3.46%
3.09%
3.03%

Obama
            
50
50
47
49
47

52
48
49
45
44

51
49
49
47
45

46.7
49.4
48.1

McCain
            
42
44
42
43
42

43
46
45
39
43

47
45
44
45
45

42.6
44.6
43.7

Other
            
8
6
11
8
11

5
6
6
16
13

2
6
7
8
10

10.7
6.0
8.2

Spread
            
8
6
5
6
5

9
2
4
6
1

4
4
5
2
0

4.1
4.8
4.5

 

Obama
            
48.6
49.0
48.6
49.0
48.2

47.6
47.4
47.6
47.6
48.0

48.2
47.2
46.6
45.6
44.6

McCain
            
42.6
42.8
43.2
43.8
43.0

43.2
44.0
43.8
43.6
44.8

45.2
45.0
45.2
46.0
46.0

Spread
            
6.0
6.2
5.4
5.2
5.2

4.4
3.4
3.8
4.0
3.2

3.0
2.2
1.4
(0.4)
(1.4)

Win Prob
               
100.0
99.8
99.5
99.7
97.8

91.5
88.1
90.0
91.1
88.7

90.6
89.1
82.2
36.3
20.1

 

Obama
            
53.9
53.9
53.5
53.3
53.5

53.1
52.6
52.8
52.9
52.3

52.2
51.9
51.5
50.6
50.2

McCain
            
46.1
46.1
46.5
46.7
46.5

46.9
47.4
47.2
47.1
47.7

47.8
48.1
48.5
49.4
49.8

Spread
            
7.8
7.8
7.0
6.6
7.0

6.2
5.1
5.5
5.8
4.6

4.3
3.8
3.0
1.3
0.5

Win Prob
               
100.00
99.99
99.97
99.98
99.6

97.4
96.2
96.9
97.4
96.0

97.1
98.2
97.7
86.9
61.3

 

 

The TRUE Math

Assume that current state polls reflect the will of the electorate and a fraud-free election is held today. The 2008 Election Model indicates that given the following undecided voter allocations (UVA), Obama’s projected 2-party vote share and expected (mean) electoral vote are:

UVA
50%
60%
75%
80%

Vote
51.3%
52.0%
53.0%
53.3%

Evote
325
340
359
364

EV Wins   (in a Monte Carlo simulation of 5000 election trials)
4966
4996          (base case scenario)
5000
5000

 

Note: the polls were taken prior to the debate — so Obama’s numbers should improve as a result.

Obama is projected to win 53.9% of the two-party vote based on the most recent 5 national polls. The national tracking polls include one-day of sampling post-debate. Since the national polls lead the states, we can expect a rise in Obama’s expected EV. With a 53.3% state poll aggregate, he could expect 364 EV (see the table).

View the State vs. National vote share projection Trend.

But what if the election is not fraud-free? Assuming 3% of votes cast are uncounted (75% of which are traditionally Democratic) and 4% of Obama’s votes are switched electronically to McCain, Obama would lose by 277–261 EV with 49.4% of the two-party vote (see the election fraud sensitivity table below).

This was the electoral vote.com map on Nov 1, 2004.

The expected electoral vote is calculated by the formula:

EV  =  Σ P(i) * EV(i),   where i = 1,51

P(i) is the probability of winning state (i) with EV(i) electoral votes.

The probability P(i) of winning state (i) is based on the projected state vote share V(i).

The projected vote share is equal to the latest poll plus the undecided voter allocation.

V(i)  =  Poll(i)+UVA(i)

The state win probability is calculated using the Excel Normal distribution function, assuming a 4.0% MoE for a typical 600-sample poll:

P(i)  =  NORMDIST ( V(i),  0.5,  .04/1.96,  true )

The electoral votes, current poll numbers, projected vote shares and win probabilities for all the states are given in the table below.

Electoral-vote.com and RealClearPolitics have Obama leading by 286–252. They assign the state electoral vote to the poll leader, regardless of the spread. This is mathematically incorrect, since win probabilities are dependent on the spread — which they do not use in calculating the EV. Furthermore, undecided voters are NOT allocated.

On the other hand, FiveThirtyEight has Obama leading by 318–220. But the win probability of 79% is much too low — it should be close to 98%. This indicates a problem with the simulation algorithm and/or the state win probabilities. The electoral vote distribution chart is not a continuous normal distribution bell-curve; there are discrete jumps in the bin totals. The curve should be smooth and symmetric around the mean (expected EV) which is within 1 EV of the median in a 5000 Election trial Monte Carlo simulation. The Election Model Electoral Vote Simulation Frequency chart is a continuous, bell-shaped EV frequency histogram.

THE 2008 ELECTION MODEL

Last
S T A T E   M O D E L
 
N A T I O N A L   M O D E L
 
MONTE CARLO
SIMULATION

Update
L A T E S T  S T A T E–P O L L  A V E R A G E
 
L A T E S T   P O L L S   M O V–A V E R A G E
 
EXPECTED

9/28/2008
Aggregate
2-party
Projection
 
5-Poll
5-Poll 2-party
Projection
 
ELECTORAL VOTE

 
 
 
60% UVA
 
 
 
60% UVA

Obama
McCain
47.96
45.34
51.40
48.60
51.98
48.02
  
  
48.60
42.60
53.29
46.71
53.88
46.12
  
  
 340 
 198 

 

Nov 1, 2004 Final Election Model
75% UVA
 
 
 
75% UVA

Kerry
Bush
47.88
46.89
50.52
49.48
51.80
48.20
  
 
47.80
46.60
50.64
49.36
51.77
48.23
  
 
 337 
 201 

 


Sensitivity Analysis — Impact of Uncounted and Switched Votes on Obama

Uncounted
1%
 
2%
 
3%
 
Switched
2%
4%
6%
Vote%
50.7
49.7
48.6
EV
309
280
249
Vote%
50.5
49.4
48.4
EV
303
274
243
Vote%
50.2
49.2
48.1
EV
296
268
237

 

Sensitivity Analysis — Impact of Aggregate State Projected Vote Share

Undecided Voter Allocation Scenario
Base Case

Obama
40%
50%
60%
75%
80%

 

Projected 2-Party Vote Share

Obama
McCain
50.6
49.4
51.3
48.7
51.98
48.02
53.0
47.0
53.3
46.7

 

MoE
Popular Vote – Obama Win Probability (Normdist)

1.0 %
2.0 %
3.0 %
89.5
73.5
66.2
99.5
90.0
80.4
99.99
97.4
90.2
100.0
99.8
97.4
100.0
99.9
98.5

 

Electoral Vote – Obama       (Monte Carlo Simulation:  based on state win-probabilities)

Mean
Median
308.5
308.0
325.1
325.5
339.9
340.0
358.6
360.0
364.3
366.0

Maximum
Minimum
377
233
404
249
420
263
420
284
420
293

 

Electoral Vote – Obama Win Probability

Trial Wins
Probability
4808
96.2
4966
99.3
4996
99.92
5000
100.0
5000
100.0

 

95% EV Confidence Interval
Upper
Lower
352
265
368
282
381
299
393
324
396
333

 

States Won
Obama
 
24
 
25
 
30
 
30
 
30
 

 

 

 

2008 POLLING ANALYSIS AND PROJECTIONS

National Model — see atop
State Model
(2-party vote shares)

 
 
L A T E S T   S T A T E   P O L L 
 
KEY STATES
(within MoE)
 
2004 EM  KERRY  VOTE–PROJECTION  vs
EXIT POLL  &  RECORDED VOTE–COUNT
 
2008  vs  2004
PROJECTED  VOTE

 

 
 
Pre-Undecided Voter Allocation
 
60% UVA
Projection
 
MC Exp EV
Win Prob
 

Resource
 

Allocate
 
Vote
Projected
WPE (IM)
 Exit Poll 
Vote
Counted
 
Kerry Projection
deviation
 
 
2008 Obama
diff
 
StatesEV
Flip To(*)
 

     

AL
AK
AZ
AR
CA

CO
CT
DC
DE
FL

•
•
•
       
 
Poll
 Date 
       
9/23
9/21
9/14
9/22
9/20

9/23
9/19
9/13
9/23
9/25

       
 
VoteShare
 Popular 
 Electoral 
       
9
3
10
6
55
       
9
7
3
3
27
       
       
Obama
 48.0 %
 301 
       
34
38
39
37
52

51
54
90
57
47

 
McCain
 45.3 %
  190  
       
60
55
56
47
36

42
40
9
37
47

      
 
Spread
2.62 %
 111  
     
(26)
(17)
(17)
(10)
16

9
14
81
20
0

 
 
Obama
51.98 %
 375 
       
37.6
42.2
42.0
46.6
59.2

55.2
57.6
90.6
60.6
50.6

 
 
Obama
99.92 %
 339.9 
       
0.0
0.0
0.0
4.8
100.0

99.5
100.0
100.0
100.0
61.6

 
 
Percent
 100 
          
       










23.9

    
 
Rank
 
          
       










1

    
 
Final  Kerry
51.75 %
 337 
       
42.0
39.8
48.8
50.5
55.8

50.8
56.5
86.3
57.8
52.3

       
JK Unadj
52.51 %
 337 

42.1
41.6
44.9
45.7
60.9

50.7
63.4
91.9
61.9
51.4

Kerry
48.76 %
 252 

37.2
35.9
44.8
45.0
54.9

47.5
54.9
90.1
53.9
47.6


    
 
Exit Poll
0.75 %
   0   

0.1
1.8
(3.9)
(4.8)
5.2

(0.0)
6.9
5.7
4.1
(0.8)

    
 
Vote Cnt
( 3.0 ) %
 (85) 
 
(4.8)
(3.9)
(3.9)
(5.5)
(0.9)

(3.3)
(1.6)
3.8
(3.9)
(4.7)

 
   Final  Kerry  
0.23 %
   3   

(4.4)
2.5
(6.8)
(3.9)
3.5

4.5
1.1
4.3
2.9
(1.7)

    
 
Obama
10
 123 

AL
AK
AZ
AR
CA

CO*
CT
DC
DE
FL*
    

•••


Full 9/28 EM update:
9/28 ELECTION MODEL (TIA): Obama @ 340EV; state & nationall polls, projections, win probabilities & election fraud scenarios

[opihimoimoi]

1. Lemme toast to the 340 EVs

:toast:

K&R

[Febble]

3. Well, TIA is wrong here:

On the other hand, FiveThirtyEight has Obama leading by 318–220. But the win probability of 79% is much too low — it should be close to 98%. This indicates a problem with the simulation algorithm and/or the state win probabilities. The electoral vote distribution chart is not a continuous normal distribution bell-curve; there are discrete jumps in the bin totals. The curve should be smooth and symmetric around the mean (expected EV) which is within 1 EV of the median in a 5000 Election trial Monte Carlo simulation. The Election Model Electoral Vote Simulation Frequency chart is a continuous, bell-shaped EV frequency histogram.

The reason that Nate Silver's histogram has "discrete jumps in the bin totals" is that the electoral college consists of discrete bins. If a candidate wins a state in any scenario the bin total "jumps" to include all the electoral votes of that state.

And you certainly wouldn't expect a symmetric plot, because the electoral votes aren't symmetrically distributed between red and blue states.

I don't know how TIA is getting his curve to be "smooth and symmetreic around the mean" but he's clearly doing something wrong, because it can't be.

[OnTheOtherHand]

4. it ISN'T "smooth and symmetric around the mean"

The ogive is pretty enough, but the histogram is not symmetrical. It just isn't.

It does appear smoother than fivethirtyeight's, however, because it uses bins that are 5 electoral votes wide. (Whoever, if anyone, taught TIA that the resulting histogram could be described as "continuous" has much to answer for.) The effective resolution of the fivethirtyeight.com graphic appears to be 2 electoral votes per pixel.

As for why fivethirtyeight's win percentage is smaller, well, maybe TIA should read the FAQ.

[Febble]

5. Right.

So TIA's curve is smoother because he uses wider bins, and he doesn't get symmetry anyway.

And neither has anything to do with why fivethirtyeight gets a smaller percentage win.

TIA probably gets a larger percentage win because he uses a different weighting for the polls.

I wonder if he still likes Zogby.

heh.

[Marnieworld]

2. kick

:hi: