Redoing my 2020 predictions and using what they should have been in order to simulate 2020

 So, the original exercise I did with simulations I did in 2020. And...it didnt go well for Trump at all. Biden won all 20 simulations I did back then. I did get some end results that kinda sorta looked like what actually happened, and some were narrower electorally than the actual election results. The real results were well within the probability of my model. 

But...after that, I ended up removing all of the weirdo right wing polls that seemed to be poll bombing at the time, and ended up with an EXTREMELY bullish prediction that ended up not being even remotely close. Again, it was within the statistical realm of probability, but let's face it, given how close Biden actually came to losing, I feel like I would've done better had I actually predicted things without using the methods I used to remove suspicious polls. because the end results were more accurate with those polls than without them.

That and if I ran 2020 again with my final numbers there's like zero way trump would probably win according to this new model. Maybe not literally zero, but given I'm only gonna be doing 20-40 goes here, we probably won't even get one trump model.

So I'm gonna take my 2020 results and just go by the polling averages on RCP without me changing them to reflect what I thought the reality would be.

States	Winner/Margin	Z score	% D Win	% R Win	EV D win	EV R win
Wisconsin	Biden + 6.7%	-1.68	95.3%	4.7%	226	322
Nebraska CD2	Biden +5.0%	-1.25	89.4%	10.6%	227	312
Minnesota	Biden +4.3%	-1.08	86.0%	14.0%	237	311
Maine CD2	Biden +3.7%	-0.93	82.4%	17.6%	238	301
Michigan	Biden + 2.8%	-0.70	76.0%	24.0%	254	300
Nevada	Biden +2.4%	-0.60	72.7%	27.3%	260	284
Pennsylvania	Biden +1.2%	-0.30	61.8%	38.2%	280	278
Florida	Biden +0.9%	-0.23	59.1%	40.9%	309	258
Arizona	Biden +0.9%	-0.23	59.1%	40.9%	320	229
North Carolina	Trump +0.2%	+0.05	48.0%	52.0%	335	218
Georgia	Trump +1.0%	+0.25	40.1%	59.9%	351	203
Ohio	Trump +1.0%	+0.25	40.1%	59.9%	369	187
Texas	Trump +1.3%	+0.33	37.1%	62.9%	407	169
Iowa	Trump +2.0%	+0.50	30.8%	69.2%	413	131
Alaska	Trump +6.0%	+1.50	6.7%	93.4%	416	125
Montana	Trump +6.8%	+1.70	4.5%	95.5%	419	122
Missouri	Trump +7.0%	+1.75	4.0%	96.0%	429	119
South Carolina	Trump +7.5%	+1.88	3.0%	97.0%	438	109

That's....actually surprisingly more accurate than what I actually predicted. So I messed up in how I did 2020. My bad. I won't be adjusting poll averages again in 2024 unless I see a good reason to do so. 

Anyway, that said, Biden actually only had a 62% chance to win 2020, and Trump had a 38% chance. This is what my actual prediction map should've been. That feels far more accurate given how close the results were. Biden won by the skin of his teeth, and the deciding state, PA, got 1.2% exactly. Michigan and Nevada were also far more accurate. Only Maine CD2 and Wisconsin were crazy amounts off, as were the +6 trump states which shouldnt even be counted but I'll count them for the sake of that being what the polling averages were. 

Actually, I'm going to go further. After looking at this, I don't think 2020 polls were off that much. The results were actually very predictable, with the only deviations being Georgia went blue, Florida went red, and Maine CD2 going red. And honestly, outside of Maine CD2, which lacked extensive polling, the polls were only off by like a point. So the polls actually WERE accurate in 2020, and I was wrong for messing with them. 

Anyway, at this point, I have a chart with which to do simulations. And of course, here's a blank map to follow along. You can do these simulations yourself just using a random number generation. Just take each of the states, and be like, okay, if it's below (R chance of win) it's trump, if it's above, it's Biden. Like Wisconsin you would say anything 5 or under is a Trump win, whereas 6-100 is Biden. Just do that with all of the states. 

Without further ado, let's do some trials.

Trial 1: 282-256 Biden

This first result is already much narrower than 2020 actually was, with a lot of states breaking completely differently than they actually did. Still, this was very possible.

Trial 2: 306-232 Biden

Basically the actual result with Michigan and Georgia flipped. Shows how accurate the polls actually were.

Trial 3: 308-230 Biden

Trump doesn't seem to be doing that good on this map, despite having a 38% chance on paper. Probably because of all of the states he needs to win while dems only need a handful to stay relevant.

Trial 4: 361-177 Biden

Yeah I can see how this is actually a really hard map for Trump to win. Mainly because he only starts out with 100 electoral votes and needs to run the board. Given real elections are decided at least partially by waves Trump's chances are better than would be indicated with the individual trial model. But yeah, in individual trials this map is very punitive to Trump. The democrats start out twice as far ahead. And unless trump can score very good roles, it's gonna be over by the time we even get to the states trump has an actual decent shot at winning.

Trial 5: 312-226 Biden

Another one where it was over before it started. Another problem is that the democrats still stand a decent chance at taking a lot of significant states Trump won. Florida had a 59% chance of going D, the same as Arizona. North Carolina had a 48% chance. Ohio had a 40% chance, the same as Georgia, which went D. Texas wasn't much better at 38%. Iowa was at 31% chance for Biden.

This really is evidence of how "wave" models are more accurate than whatever this is. But whatever this is is fun, so I'm doing these simulations just to see what happens. 

Trial 6: 342-196 Biden

Another one. It was just too hard for Trump to penetrate into Biden's territory and significantly easier for Biden to take territory from Trump via this model. Wave model is actually more accurate. Still, as we can see by trial 2, what happened was well within the realm of possibility.

Trial 7: 385-153 Biden

Another D wipeout

Trial 8: 351-187 Biden

Texas flipping in 3/8 simulations really hurts the republicans hard.

Trial 9: 274-264 Biden

A narrow win but a win nonetheless.

Trial 10: 286-252 Biden

Trial 11: 334-204 Biden

Trial 12: 305-233 Biden

Trial 13: 351-187 Biden

Yeah Im seeing the obvious issues with this simulation. Wisconsin, Nebraska CD2, Minnesota, and Maine CD2 are difficult to flip in this simulation and don't reward many electoral votes anyway. Then Michigan only has a 24% chance of going red. Nevada a 27% chance. And you don't see a state reliably switch here and there until Pennsylvania. By then the democrat is up to 260. If the democrat wins PA, unless they either derailed Michigan or two other states up to that point. And then the republican is just one state away from losing. Florida? It's over. I think the furthest I got without the dem winning outright is North Carolina. And given Texas has almost the same odds of flipping blue as Pennsylvania red, yeah, the republican is just screwed.

Anyway I'm gonna end at 20 trials. I don't see a point in going on beyond that. It's very clear that this is a statistically lost cause for Trump.

However, let's be real, real elections aren't like this. Trump almost did win, and despite the 306-232 vote (which did come up once in this simulation, although with two states flipped), it was very close.

Trial 14: 355-183 Biden

Stop stop! He's already dead!

Trial 15: 326-212 Biden

How many times do we have to teach you this lesson old man?

Trial 16: 351-187 Biden

This is the hell I wish the orange one, that he sit here, like sisyphus, doing random 2020 election predictions until he realizes that yes, he actually lost legitimately. 

Trial 17: 334-204 Biden

Trial 18: 325-213 Biden

A really oddball one. I was actually wondering if Trump might win this one. But then he didn't. Because statistics. 

Trial 19: 281-257 Biden

Trial 20: 337-201 Biden

So yeah, even with adjusting the odds and including those darned Trafalgar polls Trump STILL cant win under this model.

Well, maybe he can, but it didn't happen in this sample and it seems to happen less than 5% of the time. 

As I discussed as I was doing it, this is a brutal simulation for Trump. He needs to run the board, while Biden just started out needing a handful of states. Even weighted for probability, Trump just doesnt have a decent chance. 

Still, again, I think the lesson here is that while these predictions do a decent job simulating what could have been, the trend/wave model is largely far more accurate. Which means Trump really did have close to a 38% chance, rather than a <5% one. 

Ugh. Kind of embarrassed of my 2020 predictions now....oh well at least they explains how the Donald randomly did so good...