Posted on 06/21/2004 3:20:15 PM PDT by Momaw Nadon
Ah, perhaps you do not understand what is meant by "independent events". Events are statistically independent if, and only if, the probability of both/all events occurring is equal to the product of the probabilities of each event occurring.
The notion of statistical independence has little to do with causality, and does not respect normal principles of transitivity. It is possible for events A and B to be independent, and for B and C to be independent, and likewise A and C, and yet for the three events not to be mutually independent of each other. For an example of this, suppose I have two dice, one red and one white, and I roll them. Consider the following three events:
But won't that dependence be reflected in the polls or the "markets" also?
-PJ
I'm sure it is reflected in the markets. I would be surprised if the wagered probability of Bush winning matched and tracked the conditional probability one would compute assumuming all state probabilities are accurate and independent.
While it is theoretically possible that any of the 3*2^51 combinations of electoral votes could occur (the 3x being a consequence of Maine having six possibilites rather than 2: RR+R RD+R RD+D DR+R DR+D DD+D), the actual probabilities of certain combinations occurring do not necessarily track well with what one would combute by multiplying their individual probabilities. For example, what is the probability of all of the following states voting for Bush: Alabama, Alaska, Idaho, Indiana, Kansas, Kentucky, Mississippi, Montana, Nebraska, North Dakota, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, and Wyoming? I for one would consider that much better than an even-money wager, and yet if the state probabilities are to believed, the probability that all those states will voting for Bush would have to be about 40% for the probabilities to be independent.
-PJ
Date | Prob. Bush Win | Mean EVs | Std. Dev. |
01/21 | 96.8% | 341.5 | 41.1 |
01/26 | 95.5% | 334.8 | 40.6 |
02/02 | 92.2% | 323.8 | 39.7 |
02/09 | 83.0% | 307.8 | 40.3 |
02/16 | 78.4% | 300.4 | 39.4 |
02/23 | 76.2% | 298.2 | 39.6 |
03/01 | 74.5% | 295.9 | 39.3 |
03/08 | 68.0% | 289.2 | 39.8 |
03/15 | 68.0% | 288.8 | 39.0 |
03/22 | 68.5% | 289.3 | 38.8 |
03/29 | 69.4% | 290.1 | 38.8 |
04/05 | 71.2% | 292.3 | 39.1 |
04/12 | 70.4% | 290.6 | 38.1 |
04/19 | 68.6% | 288.1 | 36.7 |
04/26 | 64.9% | 284.5 | 36.3 |
05/03 | 66.3% | 285.7 | 36.3 |
05/10 | 65.6% | 285.3 | 36.8 |
05/17 | 65.2% | 284.8 | 36.6 |
05/24 | 60.0% | 280.3 | 36.9 |
05/31 | 61.1% | 281.2 | 36.8 |
06/07 | 60.5% | 280.6 | 36.5 |
06/14 | 65.0% | 285.0 | 36.6 |
06/21 | 63.9% | 284.0 | 36.8 |
Thanks jdege!
I started my paragraph with one thought and rather confusingly munged it with another. Let me ask you a few questions: What would you consider to be the probability that Bush wins all of the following states: Alabama, Alaska, Idaho, Indiana, Kansas, Kentucky, Mississippi, Montana, Nebraska, North Dakota, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, and Wyoming?
Your point is that you'd bet better than even money that all states would go to Bush, meaning that the probability is better than the 32% that the numbers indicate. Are you thinking that because of an inherent dependency in the character of the states, for instance, that New England as a block would go for Kerry, that the South as a block would go for Bush, that the urbanized states as a block would go for Kerry, etc? Is that where you find your dependency? I'm thinking that that dependency is encoded into the individual state probabilities already.
To rephrase your question, out of 100 elections, how many times would you expect all of the states in your list to vote for Bush?
-PJ
Most of them--probably about 60% or so. For Bush to lose any of those states, something would have to go severely wrong for him campaign. While I wouldn't consider such an event impossible, I wouldn't regard it as being even a 50% probability.
I would posit that my simplified model of the election would regard each state as having a "Bush favorability requirement" and a random dither value, and I would regard the nation as having a "national Bush favorability rating". If the national Bush favorability rating exceeds the sum of a state's favorability requirement and its dither value multiplied by a random number -1 to +1, the state will vote for Bush; otherwise the state will vote for Kerry.
If a state like Nebraska votes for Kerry, it will most likely be because Bush messed up badly. A mistake large enough to cost Bush Nebraska would also likely cost him Virginia and Tennessee as well.
The Georgia Loses scenario has a 0.005% probability of occuring, while the other scenario has a 0.003% probabiliy of occuring. That makes the Georgia Loses scenario more likely.
My gut instinct tells me that it is more likely to lose one state in a slate than it is to win all states in a slate, or paramutual betting would never survive as an industry.
-PJ
I disagree. It would take a major surge in Bush's favor for him to have a prayer of winning Rhode Island. Such a surge would be broad-based and national, so much that he cannot possibly lose conservative pro-Bush Georgia.
In both scenarios, we must premise our evaluation on the assumption that Bush wins Hawaii (CERTAINLY KERRY in my estimation but not confirmed by polls), New Jersey (LIKELY KERRY), and Rhode Island (CERTAINLY KERRY).
So which is more likely? Bush wins Delaware (CERTAINLY KERRY based upon 2000 election results but unconfirmed by polling and previously a bellwether), and Washington (POSSIBLY KERRY, a swing state).
OR Bush loses Georgia (LIKELY BUSH--and almost certinly Bush unless Kerry takes Miller as veep).
The latter situation is more likely (provided that Delaware is as certainly for Kerry as we think it is), but only if we ignore our premise. But if we consider the premise, then we give up our "statistically independent" assumption used to derive the probabilities. That assumption presupposes that the state races for the electors for president occur independently of each other. But they don't.
If Kerry cannot hold Hawaii and Rhode Island, he probably will have collapsed entirely, and hence he certainly won't win Georgia. If on the other hand, Bush manages to persuade the liberal voters in Hawaii and Rhode Island to choose him over Bush, he definitely won't have trouble in Georgia, likely will win Washington, and may take Delaware.
Of course, I'm basing my prediction on the assumption of a two-way race, wherein minor-party candidates take only a small portion of the vote in the named states. If enough Kerry voters in Rhode Island, for example, defect to Nader either to throw the state to Bush or to give Nader an outright electoral victory, then a liberal shift to throw Georgia to Kerry (or Nader) might be rational to contemplate.
It almost sounds like you are saying that the fact that a slate of states leans towards Kerry or Bush is the dependency in and of itself, given that if a news event sways one leaning state it is likely to sway them all. My point is that those states independently chose to lean that way in the first place, and it is almost coincidental that a slate of states has the same common leaning as its common bond. I doubt that there was some unmodeled dependency that caused the states, and only those states, to lean in the same direction, and therefore lean in another direction based on some event, together as a dependent block.
-PJ
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.