National Polls are Worthless

2008 Presidential Election Polls


Home
What's New
AI Products
Alzheimer's, beat it
Android Eyes
Android Fingers
Android Hands
Animatronic Products
Animatronic Sites
Asimov's Laws
Baby Androids
Bipedal Projects
Books
Business Plan
Competitions
Conferences
Digital Gyro Board
Domestic robots
Education
Engineers Recommended
Entertainment robots
Future of Androids
Global Warming Fix
Globes of planets
Greatest Android Projects
Gyro/Accelerometer board
Haptic Sensor
Head Projects
Historical Projects
In the Movies
Kill Viruses/Trojans
Live to 100
Mecha Projects
NASA Projects
Planetary Globes
Personal projects
Philosophy of Androids
PRODUCTS
Robo-prize $5M
Robotics Sites
Secret Projects
Smaller projects
Sub-assembly projects
Superintelligence
Suppliers Recommended
Tactile Sensor
Touch Sensor
Valerie Android
Video cameras (smallest)
What's New
Home

Why National Polls are Worthless

     The US presidential election is NOT one election.  It is really 51 separate elections which are held on the same day.   The pollsters obviously have forgotten this fact.   The winner of the popular vote count does NOT necessarily win the election.   Just ask AL GORE.  Four candidates for president have won the popular vote and yet have lost the presidential election.  Here they are:

Year     Popular Vote      % of          Election          % of         Others %
           Winner          vote           Winner           vote         of vote
2000      Albert Gore      48.38      George W. Bush       47.87          3.75
1888    Grover Cleveland   48.63     Benjamin Harrison     47.80          3.57
1876      Samuel Tilden    50.92     Rutherford B. Hayes   47.92          1.16
1824     Andrew Jackson    41.36     John Q. Adams         30.92         27.72

A quick lesson in statistical sampling

      The uncertainty in a sample or the error range is given by 100/ (square root of sample size).  Thus we obtain the following table of error ranges for various sample sizes.

     Sampling errors in polls
     Samples             error %
        10                31.6
        20                22.4
        30                18.3
        40                15.8
        50                13
       100                10
       200                 7
       300                 5.8
       400                 5
       500                 4.5
       600                 4
       700                 3.8
       800                 3.5
       900                 3.33
      1000                 3.2
      1225                 3.0
      2000                 2.2
      2500                 2.0
      5000                 1.4
     10000                 1.0

         In 48 states and DC, the winner of the popular vote receives ALL the electors of that state (or DC). Only Maine and Nebraska give electoral votes proportionally.   The candidate who receives a majority of the electoral votes wins the election - NOT the candidate who receives the largest number of popular votes.  

         Let us examine the results of the 2004 & 2000 elections for the states of California, New York, and Texas.

    2004           Kerry       Bush       Difference    Kerry % Bush % Diff
    California    6,745,485   5,509,826   1,235,659     54.3    44.4   K+10%
    Texas         2,832,704   4,526,917   1,694,213     38.2    61.1   B+23%
    New York      4,314,280   2,962,567   1,351,713     58.4    40.1   K+18%

    2000           Gore        Bush       Difference    Gore %  Bush % Diff
    California    5,861,203   4,567,429   1,293,774     53.5    41.7   G+12%
    Texas         2,433,746   3,799,639   1,365,893     38.0    59.3   B+21%
    New York      4,107,907   2,403,374   1,704,533     60.2    35.2   G+25%

We see that Democrats win California by about 11% and New York by about 22% while Republicans win Texas by about 22%

  Suppose the pollster selects a sample size of 1000 likely voters

    If the sample is divided among 50 states & DC = 19.6 voters each.
    Thus each individual poll would have an error of 22.5%.
    I believe this would be WORTHLESS.

    Suppose the 1000 votes are divided proportionally
    (by population) among the 50 states & DC. Then we would
    get the following:
                                                          2004      2000
    California     = 12.12% = 121 votes  (9% error)       K+10%     G+12%
    Texas          =  7.93% =  79 votes  (11% error)      B+23%     B+21%
    New York       =  6.4%  =  64 votes  (12.5% error)    K+18%     G+25%
    Florida        =  6.05% =  61 votes  (13%)            B+5%      B+0
    Illinois       =  4.25  =  42 votes  (15.4%)          K+10%     G+12%
    Pennsylvania   =  4.12  =  41 votes  (15.5%)          K+2.5%    G+4%
    Ohio           =  3.80  =  38 votes  (16%)            B+2.0%    B+3.5%
    Michigan       =  3.34  =  33 votes  (17%)            K+3.4%    G+5%
    Georgia        =  3.16  =  32 votes  (18%)            B+17%     B+13%
    North Carolina =  3.0   =  30 votes  (18.5%)          B+12%     B+13%
                              --------
                              541 votes = 54.1% of the voters

   1000  541 = 459 divided by 41 remaining states & DC give 11 each (30 % error)
   
   This gives an average error of about 20% - which is really WORTHLESS.
   
   

   Suppose the pollster selects a sample of 1000 likely voters from California
   
   Then the Democrat will lead by 11% =  WORTHLESS
   
   Suppose the pollster selects a sample of 1000 likely voters from New York
   
   Then the Democrat will lead by 22% =  WORTHLESS
   
   Suppose the pollster selects a sample of 1000 likely voters from Texas
   
   Then the Republican will lead by 22% =  WORTHLESS 
    

          Comments?   Email me at crwillis@androidworld.com