Jump to content
270soft Forum

I ran 270soft's 2020 Presidential Election 100 times -- here's what I learned


Recommended Posts

270soft has a 2020 Presidential election simulator that will automatically track the stats for you.  https://www.270towin.com/2020-simulation/

Out of 100 simulations, Biden won 88 times.  Trump won 11.  And there was exactly one 269-269 split.  

I have a lot more interesting stats at the end -- but first, some pictures!

Here was Biden's most dominant win, 444-94.

833003506_Bidenvictory.jpg.2c6e74e1577db319fbb9403066f7ae28.jpg
Honorable mention to Biden's second most dominant win, 441-97:

B441.jpg.176a1e8b02a3981d545bc78d236fb30b.jpg

Donald Trump's most dominant win. 327-211.
430489966_Trumpsbestvictory.jpg.2949bc066f14c94f34f00acda061d7bd.jpg
And dishonorable mention to Trump's second highest performance, 322-216.1254892732_Trump311.jpg.662faae0e39694ece323f1a79285b9e9.jpg

But now the interesting thing:  Stats!

Biden won Colorado in 99% of the elections.  Nevada in 93%.  Michigan in 84%.  Wisconsin in 76%  Florida in 66%.  Georgia in 53%.  Texas in 30%.  Alaska in 14%.  Mississippi in 1%.  

The greatest predictor for who would win the general election was who won PA (went to the GE winner 93% of the time) -- Biden won PA 89% of the time.
This is followed by Michigan and Minnesota which each predicted the GE winner 92.9% of the time -- they were won by Biden 84% and 92% of the time, respectively.

The closest state was Iowa, which was won by Biden 51% of the time.

And the tipping point state -- the state that put the winning candidate past the 270 threshold when ordered from largest to smallest states -- was PA (16% of the time) AND Florida (16%).

Full stats of all 100 elections:

image.thumb.png.902526f394a8de4984450f94f8a4cd20.png

  • Like 2
  • Thanks 1
  • Upvote 2
Link to post
Share on other sites

I kept the 100 results and added 300 more on top of it, for 400 election simulations total.  That's going to be my max -- I have things to do today, I can't get fired. ;c)

Notably, while Trump did improve his win percentage slightly (he's now at 14.2% instead of 11%), his ceiling stayed at 327 votes.  Meanwhile, Biden's ceiling improved to 449.  Indeed, Biden had significantly more 400+ victories than Trump had 300+ victories.

Our new record:
 NewBiden449.jpg.f1e5acaca7bef1b5c297062fe7922928.jpg

Our new stats:

 

Session Statistics

These will reset when you reload the page

Total Simulations: 400

  Biden Ties Trump Bellwether1 Tipping Point2
Overall Winning % 85.5% 0.3% 14.2%    
Alaska 10.5%   89.5% 24.8% -
Arizona 66.3%   33.8% 77.2% 6.3%
Colorado 96.5%   3.5% 86.7% 2.0%
Florida 73.3%   26.8% 84.2% 24.3%
Georgia 47.5%   52.5% 60.9% 4.3%
Indiana 2.0%   98.0% 16.3% -
Iowa 46.3%   53.8% 58.6% 1.0%
Kansas 0.8%   99.3% 15.0% -
Louisiana 0.3%   99.8% 14.5% -
Maine 95.5%   4.5% 87.7% 0.5%
Maine 2nd 47.3%   52.8% 59.4% -
Michigan 83.8%   16.3% 88.0% 11.0%
Minnesota 89.5%   10.5% 88.2% 4.0%
Mississippi 0.8%   99.3% 15.0% -
Missouri 3.0%   97.0% 17.3% -
Montana 5.5%   94.5% 19.8% -
Nebraska 2nd 55.5%   44.5% 66.9% 0.5%
Nevada 90.3%   9.8% 86.0% 3.5%
New Hampshire 86.3%   13.8% 88.0% 1.3%
New Mexico 99.8%   0.3% 86.0% 0.5%
North Carolina 67.8%   32.3% 77.7% 7.2%
Ohio 41.5%   58.5% 55.4% 4.8%
Pennsylvania 86.8%   13.3% 89.5% 15.8%
South Carolina 7.5%   92.5% 21.8% -
Texas 30.0%   70.0% 44.4% 2.3%
Virginia 97.3%   2.8% 87.0% 4.3%
Wisconsin 74.5%   25.5% 84.0% 6.5%

1 Percentage of simulations where nominee winning state wins the election.
2 The state that gives the election winner 270 electoral votes, when ordering from largest to smallest margin of victory.

  • Like 2
  • Upvote 2
Link to post
Share on other sites
1 minute ago, Actinguy said:

I kept the 100 results and added 300 more on top of it, for 400 election simulations total.  That's going to be my max -- I have things to do today, I can't get fired. ;c)

Notably, while Trump did improve his win percentage slightly (he's now at 14.2% instead of 11%), his ceiling stayed at 327 votes.  Meanwhile, Biden's ceiling improved to 449.  Indeed, Biden had significantly more 400+ victories than Trump had 300+ victories.

Our new record:
 NewBiden449.jpg.f1e5acaca7bef1b5c297062fe7922928.jpg

Our new stats:

 

Session Statistics

These will reset when you reload the page

Total Simulations: 400

  Biden Ties Trump Bellwether1 Tipping Point2
Overall Winning % 85.5% 0.3% 14.2%    
Alaska 10.5%   89.5% 24.8% -
Arizona 66.3%   33.8% 77.2% 6.3%
Colorado 96.5%   3.5% 86.7% 2.0%
Florida 73.3%   26.8% 84.2% 24.3%
Georgia 47.5%   52.5% 60.9% 4.3%
Indiana 2.0%   98.0% 16.3% -
Iowa 46.3%   53.8% 58.6% 1.0%
Kansas 0.8%   99.3% 15.0% -
Louisiana 0.3%   99.8% 14.5% -
Maine 95.5%   4.5% 87.7% 0.5%
Maine 2nd 47.3%   52.8% 59.4% -
Michigan 83.8%   16.3% 88.0% 11.0%
Minnesota 89.5%   10.5% 88.2% 4.0%
Mississippi 0.8%   99.3% 15.0% -
Missouri 3.0%   97.0% 17.3% -
Montana 5.5%   94.5% 19.8% -
Nebraska 2nd 55.5%   44.5% 66.9% 0.5%
Nevada 90.3%   9.8% 86.0% 3.5%
New Hampshire 86.3%   13.8% 88.0% 1.3%
New Mexico 99.8%   0.3% 86.0% 0.5%
North Carolina 67.8%   32.3% 77.7% 7.2%
Ohio 41.5%   58.5% 55.4% 4.8%
Pennsylvania 86.8%   13.3% 89.5% 15.8%
South Carolina 7.5%   92.5% 21.8% -
Texas 30.0%   70.0% 44.4% 2.3%
Virginia 97.3%   2.8% 87.0% 4.3%
Wisconsin 74.5%   25.5% 84.0% 6.5%

1 Percentage of simulations where nominee winning state wins the election.
2 The state that gives the election winner 270 electoral votes, when ordering from largest to smallest margin of victory.

Wow! South Carolina and Alaska. Weren't FDR and LBJ, respectively, the last Democratic Presidential Candidates to carry each? @pilight? @vcczar?

Link to post
Share on other sites
3 minutes ago, Patine said:

Wow! South Carolina and Alaska. Weren't FDR and LBJ, respectively, the last Democratic Presidential Candidates to carry each? @pilight? @vcczar?

LBJ lost SC

Link to post
Share on other sites
1 minute ago, Patine said:

Wow! South Carolina and Alaska. Weren't FDR and LBJ, respectively, the last Democratic Presidential Candidates to carry each? @pilight? @vcczar?

No, while you’re correct about Johnson being the last (and only) Democrat to carry Alaska. However, the most recent time a Democrat has carried South Carolina was by Jimmy Carter in 1976. Kennedy and Stevenson also carried it post- FDR in 1960, and 1952 and 1956 respectively.

  • Like 1
  • Thanks 2
Link to post
Share on other sites
1 minute ago, vcczar said:

LBJ lost SC

 

Just now, vcczar said:

and AK wasn't a state when FDR was president @Patine

He said respectively. As in LBJ referring to Alaska and FDR to SC. I’ve responded to both claims. He was correct about Alaska but wrong about South Carolina.

  • Like 1
  • Thanks 1
Link to post
Share on other sites
Just now, vcczar said:

LBJ lost SC

But the last to hold Alaska, in his case. Notice, I said, "FDR and LBJ, respectively,"  as in each being the last to hold each state.

Link to post
Share on other sites
15 minutes ago, Wiw said:

By that logic, the map should look like this.

jelOz.png

Is this the average result of the 400 elections?  If so, thanks!  I was planning to get around to making that, saved me some time.

Link to post
Share on other sites
On 8/5/2020 at 7:18 AM, Actinguy said:

Biden won Colorado in 99% of the elections.  Nevada in 93%.  Michigan in 84%.  Wisconsin in 76%  Florida in 66%.  Georgia in 53%.  Texas in 30%.  Alaska in 14%.  Mississippi in 1%.  

Yeah I don't know where they're getting their "seed" probabilities from but I'm not buying this. NV and MI were close states on either end in 2016. You can say Florida and Wisconsin are toss-ups but 66-76%? I would like to know what their methodology for this simulation is. They say they run 1,000 simulations per day. Okay fine but even those simulations have to be based on something. I know they're not conducting 1,000 polls per day, so this would be subjective at best. Though the simulations I ran didn't have anywhere near that big a victory for Trump either. To have such a wide range of outcomes implies too much randomness in the sample. The standard deviation of these results is too high relative to the mean for this to be a reliable experiment.

Link to post
Share on other sites
On 8/5/2020 at 8:09 AM, Patine said:

Wow! South Carolina and Alaska. Weren't FDR and LBJ, respectively, the last Democratic Presidential Candidates to carry each?

That immediately raises red flags. I'd like to know what their methodology is. When they run a simulation, it has to be based on which candidate has which probability of winning each state. With a sample size this large I have to assume their base probabilities are about equal to the percentage that you're experiencing. Still, where did THOSE probabilities come from? I call bullshit on that. Unless they're conducting a poll per day in each state this can only be subjective. At best, they have to have some formula for translating polling averages (which are questionable to begin with) to probabilities of victory. I'd like to see what that is, too. I also recall, when I ran similar experiments, there being a very large variance in the sample too, I'd like to see what the 95% confidence interval is. These samples seem to have a very high coefficient of variation to them too, which renders such simulation virtually worthless.

Link to post
Share on other sites
3 hours ago, servo75 said:

Yeah I don't know where they're getting their "seed" probabilities from but I'm not buying this. NV and MI were close states on either end in 2016. You can say Florida and Wisconsin are toss-ups but 66-76%? I would like to know what their methodology for this simulation is. They say they run 1,000 simulations per day. Okay fine but even those simulations have to be based on something. I know they're not conducting 1,000 polls per day, so this would be subjective at best. Though the simulations I ran didn't have anywhere near that big a victory for Trump either. To have such a wide range of outcomes implies too much randomness in the sample. The standard deviation of these results is too high relative to the mean for this to be a reliable experiment.

It’s 25,000 not 1,000.

As for the seed, they state in the opening paragraph:

The simulator is largely based on a data-driven presidential model created by JHK forecasts. However, it also takes into consideration the election forecast of one or more long-time qualitative forecasters.”

As for this being to random to be predictive...I mean...yeah.  I don’t know what you were expecting for a simulation on how people will behave in the future when even some of those people haven’t yet decided how they are going to behave.

The more simulations you run, the closer you get to a likely outcome (which is not to say the only possible outcome.)

As Wiw noted above, the average of my 400 simulations came out as 314-214 in favor of Biden.  There’s nothing glaring in that map such as Biden winning Mississippi or Alaska.  You’ll have outliers in any simulation, the point is to run it a lot so that the outlier noise dissipates.

 

  • Like 1
Link to post
Share on other sites
3 hours ago, servo75 said:

That immediately raises red flags. I'd like to know what their methodology is. When they run a simulation, it has to be based on which candidate has which probability of winning each state. With a sample size this large I have to assume their base probabilities are about equal to the percentage that you're experiencing. Still, where did THOSE probabilities come from? I call bullshit on that. Unless they're conducting a poll per day in each state this can only be subjective. At best, they have to have some formula for translating polling averages (which are questionable to begin with) to probabilities of victory. I'd like to see what that is, too. I also recall, when I ran similar experiments, there being a very large variance in the sample too, I'd like to see what the 95% confidence interval is. These samples seem to have a very high coefficient of variation to them too, which renders such simulation virtually worthless.

I mean, you could just look at their polling data.  It’s on their site.  
 

Their polling average for Iowa actually is 50/50, which is indeed how things played out in my simulations.

They have Colorado at 53% Biden; 40% Trump.  That’s a 13 point margin, difficult to overcome.  And indeed, in my simulations, Trump only managed to beat that margin about 3% of the time.

They have Trump polling at 50% in Mississippi, Biden polling at 41%.  Still an uphill battle, but only a 9% deficit instead of 13%.  And yet in 400 simulations, Biden only pulled it off 0.8% of the time.

They have Alaska polling at 50% Trump, 46% Biden.  The state isn’t as red as it used to be, I remember people saying Hillary could win against Trump there in 2016.  Didn’t come to pass there, but it was a point of Discussion.  Here, after 400 simulations, Biden won 10% of the time.  I think it’s reasonable that Biden could overcome a 4 point margin in 1 out of every 10 elections.

Link to post
Share on other sites
15 hours ago, Actinguy said:

The simulator is largely based on a data-driven presidential model created by JHK forecasts

Seems vague to me. Any time the researcher or someone running a simulation won't share their methodology, it's reason to be suspect. I'd like to see this model and what assumptions it's making.

15 hours ago, Actinguy said:

I mean, you could just look at their polling data.  It’s on their site.  

If it's based mostly on polling, we already know that's suspect at best. Besides, poll data is one thing. The key is, how do they translate polling data into probabilities? In their website programming, there has to be some sort of base probability or randomized formula being used to make each state's prediction. I'd like to know what that is, and when I clicked on the link above, I got a blank page saying they were "adjusting" it. Is it possible Trump was "winning" too often? Bottom line, the way it looks right now we have an unknown algorithm (maybe) on top of an imperfect polling system. And after 538 predicted a 97% chance of Hillary winning 4 years ago, one must look at all these simulations with a jaundiced eye.

I just ran another 100 trials: 10% overall electoral college wins. 27% chance of winning Florida? 25% chance of Wisconsin? 17% chance of Michigan? Only 58% of Texas? 3% of Colorado, a state he came within 5 points in 2016? There is no way I'm buying that. You can call it wishful subjective thinking if you want, but right now, so is everyone else's opinion. I know Trump's somewhat down in the polls right now (though that's tightening), but this comes off as a DNC wet dream.

Link to post
Share on other sites
34 minutes ago, servo75 said:

Seems deliberately vague to me. I'd like to see this model and what assumptions it's making. If it's based mostly on polling, we already know that's suspect at best. Maybe it's not based mostly on polling. Who knows? And that's the point.

 

Well, we know how well polling did in 2016. Besides, poll data is one thing. The key is, how do they translate polling data into probabilities? In their website programming, there has to be some sort of base probability or randomized formula being used to make each state's prediction. I'd like to know what that is, and when I clicked on the link above, I got a blank page saying they were "adjusting" it. I guess Trump was "winning" too often. Bottom line, the way it looks right now we have an unknown algorithm (maybe) on top of an imperfect polling system.

Ok...you want their data, but you also don't believe their data without even looking at it.  This simulator may not be for you.  

Link to post
Share on other sites
5 hours ago, Actinguy said:

Ok...you want their data, but you also don't believe their data without even looking at it.  This simulator may not be for you.  

Well, how can I believe (or not believe) their data without seeing it? Or at least their methodologies. (Incidentally, I have a graduate degree in Statistics so I know what I'm talking about when it comes to polling and simulations).

I'm just looking for where they get the numbers from.  Yes, I get that their polling numbers are on their website. I won't go into again how unreliable polling can be, so at best it's a methodology built on stilts. But I digress... #1 How does one translate a Biden +3 poll average into a probability? Is there some regression modeling going on? Yes I get it, they run 25,000 simulations per day, and let's say that their "base probabilities" come from those simulations. In fact, if you run their simulator 400 times, one can probably assume that the experimental percentages are roughly equal to the base probabilities. A little something we call regression to the mean. But then that begs the next question - 25,000 simulations of what? At some point up the chain there has to be a methodology for calculating the likelihood of winning any given state in any given simulation. One way to run such a simulation is to perhaps generate a random number from 1 to 100. If Candidate A has a 60% chance of winning a state, then any number 41 or above will be marked as a win, 40 or below a loss. So chicken or the egg, where does the 60/40 come from? Although I'm simplifying things a bit here, and that's not the only way to run a simulation, but under the hood, there has to be some way of making that calculation. That's what I'm looking for.

By the way, I ran 100 simulations of my own. Statistics: 18 Trump wins, 81 Biden wins, 1 tie. I will reserve judgement until I hear back from them (I emailed them to ask what their specific methodology is) but this does seem like it's based on skewed media polls that were wrong 4 years ago.

Electoral votes mean: 213.9 to 324.1. For Trump EV, median 211 (so pretty bell-shaped) and standard deviation of 59.4. That's pretty big - a 28% coefficient of variation which is pushing the border for acceptable values. I do apologize if I'm bringing back nightmares of College Statistics classes, but I've taught that class many times, so... anyway... My point is that the results have too much variation to be any type of reliable prediction.

 

 

 

Link to post
Share on other sites
11 hours ago, servo75 said:

Well, how can I believe (or not believe) their data without seeing it? Or at least their methodologies. (Incidentally, I have a graduate degree in Statistics so I know what I'm talking about when it comes to polling and simulations).

I'm just looking for where they get the numbers from.  Yes, I get that their polling numbers are on their website. I won't go into again how unreliable polling can be, so at best it's a methodology built on stilts. But I digress... #1 How does one translate a Biden +3 poll average into a probability? Is there some regression modeling going on? Yes I get it, they run 25,000 simulations per day, and let's say that their "base probabilities" come from those simulations. In fact, if you run their simulator 400 times, one can probably assume that the experimental percentages are roughly equal to the base probabilities. A little something we call regression to the mean. But then that begs the next question - 25,000 simulations of what? At some point up the chain there has to be a methodology for calculating the likelihood of winning any given state in any given simulation. One way to run such a simulation is to perhaps generate a random number from 1 to 100. If Candidate A has a 60% chance of winning a state, then any number 41 or above will be marked as a win, 40 or below a loss. So chicken or the egg, where does the 60/40 come from? Although I'm simplifying things a bit here, and that's not the only way to run a simulation, but under the hood, there has to be some way of making that calculation. That's what I'm looking for.

By the way, I ran 100 simulations of my own. Statistics: 18 Trump wins, 81 Biden wins, 1 tie. I will reserve judgement until I hear back from them (I emailed them to ask what their specific methodology is) but this does seem like it's based on skewed media polls that were wrong 4 years ago.

Electoral votes mean: 213.9 to 324.1. For Trump EV, median 211 (so pretty bell-shaped) and standard deviation of 59.4. That's pretty big - a 28% coefficient of variation which is pushing the border for acceptable values. I do apologize if I'm bringing back nightmares of College Statistics classes, but I've taught that class many times, so... anyway... My point is that the results have too much variation to be any type of reliable prediction.

 

 

 

What data are you using to decide that the simulation is based on skewed media polls?  I want to see it.  Also, it's wrong.

;c)

Link to post
Share on other sites
Just now, Actinguy said:

What data are you using to decide that the simulation is based on skewed media polls?  I want to see it.  Also, it's wrong.

;c)

I think he's using his "authority" of having taught statistics, even though he may be ignoring the fact that these polls and simulators are made by more eminent statisticians than @servo75. The thing with Servo is that he has week claims of authority on about everything he talks about, whether it's about what the "left" is or that the Democrats "only have one wing," or that @Reagan04 isn't a Republican and/or Conservative, and that the fact that one must have money to make money is a good thing. Overall, he basically loses all credibilty for anything he says because of false statements. To his credit, he's a little more trustworthy than Trump at a press conference. 

Link to post
Share on other sites
12 minutes ago, vcczar said:

I think he's using his "authority" of having taught statistics, even though he may be ignoring the fact that these polls and simulators are made by more eminent statisticians than @servo75. The thing with Servo is that he has week claims of authority on about everything he talks about, whether it's about what the "left" is or that the Democrats "only have one wing," or that @Reagan04 isn't a Republican and/or Conservative, and that the fact that one must have money to make money is a good thing. Overall, he basically loses all credibilty for anything he says because of false statements. To his credit, he's a little more trustworthy than Trump at a press conference. 

I remember the time he said Obama was an outright, literal, died-in-the-wool, Communist, and, after I pushed the issue, he said it was no exaggeration or hyperbole and he knew EXACTLY what a Communist is and what they stand for (though he shows no evidence of having a real clue of this - quoting ridiculous, uneducated Glenn Beck and Ann Coulter myths on the issue instead), and said that Obama perfectly lived up to "Communist ideology," and was the first American Communist President.

  • Upvote 1
Link to post
Share on other sites
Just now, Patine said:

I remember the time he said was an outright, literal, died-in-the-wool, Communist, and, after I pushed the issue, he said it was no exaggeration or hyperbole and he knew EXACTLY what a Communist is and what they stand for (he shows no evidence of having a real clue of this - quoting ridiculous, uneducated Glenn Beck and Ann Coulter myths on the issue instead), and said that Obama perfectly lived up to "Communist ideology," and was the first American Communist President.

Yeah, that's proof of delusion and intolerance. 

Link to post
Share on other sites
10 hours ago, Patine said:

I remember the time he said Obama was an outright, literal, died-in-the-wool, Communist, and, after I pushed the issue, he said it was no exaggeration or hyperbole and he knew EXACTLY what a Communist is and what they stand for (though he shows no evidence of having a real clue of this - quoting ridiculous, uneducated Glenn Beck and Ann Coulter myths on the issue instead), and said that Obama perfectly lived up to "Communist ideology," and was the first American Communist President.

Well so much for truce. Goodbye.

Link to post
Share on other sites
10 minutes ago, Patine said:

So, you stand by your belief that Obama is a Communist?

He thinks anyone left of himself is a Communist. 

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
×
×
  • Create New...