Election simulation

[jacob_101]

I wanna play:    : )

ME CD2 60%
NH 50%
NM 15%
IA 25%
MN 50%
WI 50%
OH 45%
PA 80%
FL 25%

[Inverted Things]

I wanna play:    : )

ME CD2 60%
NH 50%
NM 15%
IA 25%
MN 50%
WI 50%
OH 45%
PA 80%
FL 25%

Kerry's Chances: 19.6%
Chance for tie: 2.2%

[Erc]

Aveleval...a Monte Carlo simulation is nice if you wanted to include dozens of states, but if you're just doing that limited of a number, it's easy enough to outline the probabilities directly--which is how I calculate my Kerry-O'Meter (see sig) at noon and midnight every day.

With your initial values, Kerry has a 23.2% chance of winning (not far off from your 22%)--note that I can't easily get tie probabilities, as I naturally assumed Bush would win on them.

[Erc]

Tested it out for a few other values--your 19.6% prediction for the last one was dead on.  Well, with an n of 1,000,000, you should be pretty on target.

[Inverted Things]

Aveleval...a Monte Carlo simulation is nice if you wanted to include dozens of states, but if you're just doing that limited of a number, it's easy enough to outline the probabilities directly--which is how I calculate my Kerry-O'Meter (see sig) at noon and midnight every day.

With your initial values, Kerry has a 23.2% chance of winning (not far off from your 22%)--note that I can't easily get tie probabilities, as I naturally assumed Bush would win on them.

Agreed. I did this more or less to teach myself how to use linked lists.

[Inverted Things]

My program also outputs a histogram of the number of times Kerry gets each # of EVs. If someone tells me how to use images on this forum, I can post a few.

[King]

This looks interesting...

ME CD2--82%
NH--91%
NM--18%
IA--10%
MN--58%
WI--26%
OH--41%
PA--98%
FL--40%

[Erc]

This looks interesting...

ME CD2--82%
NH--91%
NM--18%
IA--10%
MN--58%
WI--26%
OH--41%
PA--98%
FL--40%

26.6% chance of a Kerry win...

[Inverted Things]

This looks interesting...

ME CD2--82%
NH--91%
NM--18%
IA--10%
MN--58%
WI--26%
OH--41%
PA--98%
FL--40%

26.6% chance of a Kerry win...

26.6 Confirmed
7.7% chance for a tie

[Mr. Morden]

Avelaval--I really like your idea, however I think that the biggest source of error is likely to be your assumption that the outcomes of all of the swing states are uncorrelated.  I think there's a high probability that on Tuesday we're going to be surprised by one candidate or the other systematically outperforming our expectations across the board.  This happened with the Democrats in 1998 and 2000, and with the Republicans in 2002.  One side or the other does somewhat better than expected nationwide, perhaps because of some systematic bias in the polls, or because one side turns out to have a much better turnout machine than expected.  This could easily be enough of an effect to overwhelm the state-by-state variations.

An easy way to incorporate this into your simulation would be to have an additional random variable that shifts the probabilities for *all* the swing states in one direction by a fixed amount.  For example, have the simulation randomly choose a number between +10% and -10%.  If it picks 8% in the Bush direction, then the probabilities for *all* the swing states get shifted by 8% towards Bush.  If it picks 5% in the Kerry direction, they all get shifted by 5% towards Kerry.  You could even break it down in the end to see what percent of the time a national break towards Bush still results in a Kerry win or vice versa.

[Inverted Things]

Avelaval--I really like your idea, however I think that the biggest source of error is likely to be your assumption that the outcomes of all of the swing states are uncorrelated.  I think there's a high probability that on Tuesday we're going to be surprised by one candidate or the other systematically outperforming our expectations across the board.  This happened with the Democrats in 1998 and 2000, and with the Republicans in 2002.  One side or the other does somewhat better than expected nationwide, perhaps because of some systematic bias in the polls, or because one side turns out to have a much better turnout machine than expected.  This could easily be enough of an effect to overwhelm the state-by-state variations.

An easy way to incorporate this into your simulation would be to have an additional random variable that shifts the probabilities for *all* the swing states in one direction by a fixed amount.  For example, have the simulation randomly choose a number between +10% and -10%.  If it picks 8% in the Bush direction, then the probabilities for *all* the swing states get shifted by 8% towards Bush.  If it picks 5% in the Kerry direction, they all get shifted by 5% towards Kerry.  You could even break it down in the end to see what percent of the time a national break towards Bush still results in a Kerry win or vice versa.

Like the idea, but I don't see how illuminating it will be. In 2002 Gallup and Mason-Dixon did incredibly well at predicting senate and gubernatorial races despite the right-wing breeze that seems to blow across the country. I'm of the opinion that the polls can be trusted, which is why I don't have NV, WA, OR, MI, etc. in my simulation. Perhaps I should add HI to the simulation as well. If anyone is interested in the source code let me know.

[ilikeverin]

My program also outputs a histogram of the number of times Kerry gets each # of EVs. If someone tells me how to use images on this forum, I can post a few.

[img*]http://INSERT YOUR IMAGE URL HERE[/img]

(minus the *)

Example:

[img*]https://uselectionatlas.org/FORUM/Themes/classic/images/english/reply.gif[/url], minus the * is:

[dougrhess]

Since you're running a Monte Carlo simulation, couldn't you include all the states to account for the odd surprise?
Perhaps someone can calculate the chances more systematic based on a combination of state polls.

I could, but in my opinion that would give less than reliable results. Among my biggest assumptions is independence of states. If I give Bush a chance at Washington, for instance, Bush likely has Florida all locked up, which would violate my key assumption of state independence.  I might add CO, WV, NV, MI, OR, NJ and ME at large, but I feel a little sketchy about those as well.

What kind of software do you do to run this? I have some stats packages (Stata, SAS, SPSS) but not sure if those are sufficient. Interesting topic!

[I spent the winter writing songs about getting better]

ME CD2 73%
NH 53%
NM 46%
IA 41%
MN 53%
WI 50%
OH 51%
PA 82%
FL 39%

[I spent the winter writing songs about getting better]

How about with TradeSports values?

NH 65%
NM 32.8%
IA 42%
MN 61.5%
WI 49.7%
OH 47.1%
PA 74.1%
FL 41.2%

ME CD2 I'll guess about 75%

[Inverted Things]

ME CD2 73%
NH 53%
NM 46%
IA 41%
MN 53%
WI 50%
OH 51%
PA 82%
FL 39%

Kerry's chances: 35.9%
Chance for Tie: 2.8%

[Inverted Things]

How about with TradeSports values?

NH 65%
NM 32.8%
IA 42%
MN 61.5%
WI 49.7%
OH 47.1%
PA 74.1%
FL 41.2%

ME CD2 I'll guess about 75%

Kerry's chances: 34.3%
Chance for tie: 3.4%

[Inverted Things]

Since you're running a Monte Carlo simulation, couldn't you include all the states to account for the odd surprise?
Perhaps someone can calculate the chances more systematic based on a combination of state polls.

I could, but in my opinion that would give less than reliable results. Among my biggest assumptions is independence of states. If I give Bush a chance at Washington, for instance, Bush likely has Florida all locked up, which would violate my key assumption of state independence.  I might add CO, WV, NV, MI, OR, NJ and ME at large, but I feel a little sketchy about those as well.

What kind of software do you do to run this? I have some stats packages (Stata, SAS, SPSS) but not sure if those are sufficient. Interesting topic!

All you need is a C compiler.