So, 538 published their model for 2024 for the election and currently has Biden at a 52% chance of winning, Trump at a 47% chance of winning, and of course a 1% chance of a tie.
I think this is WAY too optimistic. I still have a 72% shot for Trump and a 28% shot for Biden. I admit my model is more simplistic than theirs, but if I were to simulate outcomes, given Biden only has one realistic path to 270, I bet he wouldn't do well.
I'm not gonna focus too much on the forecast itself here as skimming it over it seems to reflect much of what my own forecast does roughly, but I do wanna see how the fudge they have the forecast so optimistic for Biden. For that, I must turn to their methodology.
First they measure polls. They have a complex methodology dedicated to this and give reasons why. I skimmed it, but I generally stick with raw averages. I tried excluding some biased polls last time and look how that worked out. I actually am a believer that the polls are generally indicative of the final result, and that looking at the previous 5 election cycles recently, I can expect to get 47-48 states correct from polling averages alone. I also think polling averages do give a reasonable range of potential outcomes and that I really don't see a good reason to weigh them. They do. And I'm not necessarily going to say that they're wrong, but they are adding complication into the model that isn't really necessary, nor does it necessarily greatly increase the accuracy of it IMO.
I mean, 538's models tend to always strike me as being overly complicated and adding in a lot of stuff that doesn't need to exist while not really producing a better outcome than my simplistic ones do. Keep in mind in 2016 for example I had it more right than they even did, stating that Clinton had a 56% chance, while they gave her a 70% chance. While in both there was uncertainty allowing for a different outcome to occur, idk, I feel like just going by the averages, I kind of had it MORE right. Any time I try anything else i seem to get it "less right" in my opinion.
Next they make a "fundamentals" prediction. They kind of do something similar to the Allen Lichtman's "keys" model where they start assuming the environment will lean certain ways based on how certain conditions lean with the election. They mention stuff like economic conditions, political fundamentals like incumbency, etc. Again, I feel like this is overanalyzing things, and I think this is where things are going wrong. Polls seem to indicate people are unhappy with Biden, I get a feeling of economic malaise similar to the 1970s when I look at how people feel about Biden. But then you got people going by raw indicators and think he's doing great and should coast to reelection. This is where I think Lichtman goes wrong, and I think 538 is making a similar mistake.
From there, they tend to combine the two above factors and combine for uncertainty. They recognize they're trying to predict the outcome months in the future and peoples' minds can change, so they try to predict which direction they're gonna go in, I'm gonna assume they're going in a direction where things are gonna even out for Biden. And this CAN be the case. I'm becoming increasingly worried we're NOT gonna see this, and I admit, this is why I regularly update my model. I fully recognize a flaw of my model is that the polls CAN change, and I tend to just look at how things are NOW and how things will turn out by election day. I don't try to predict 5 months into the future but merely talk about the current state of affairs. What will happen in August, September, October? I don't know. I don't claim to know. I hope things even out, but I'm not really sure.
Do I necessarily agree with what 538 is doing? I mean this is where they differ from me. They ARE more sophisticated than me. Whether that's a good thing remains to be seen. I mean anyone can come up with some complex formula for predicting the outcome but whether the rubber meets the road is another matter.
They once again, try to account for polling bias from the industry itself, and again, after last time, I don't really wanna do this with mine. I tried, it made me MORE inaccurate. Just see my actual 2020 prediction vs the more revamped one I made more recently after the fact. I was WRONG in 2020. Even if I got the outcome right, I was nowhere even close on the margin. Normally polls alone gets you about 47-48 states, and honestly? Given we're aiming for around 95% accuracy in forecasts, that seems to track. I'm not sure these models actually really make things more clear or definitive, I've seen 538's forecasts in 2016 and 2020 and honestly? I don't really see them as meaningfully better than mine. Just saying.
And yeah that's generally what they do. I kinda see where they're going with this, but I still prefer to stick to my own analyses, just using 538 as a comparison to see how we differ, and yeah. Right now, our models differ. My model looks purely at polling. They tend to assume a lot of things like economic and political fundamentals and trim the polling averages. They seem to think based on this that it's anyone's game and it's roughly 50-50. I would disagree. I am currently leaning in favor of Trump, and if change happens in the next few months I'll be pleasantly surprised and report it, but right now I'm kinda crapping bricks at the polling data.
I think the "fundamentals" are actually really bad for Biden right now. The country seems unhappy with his leadership in general and nothing I'm seeing seems to indicate that he's gonna win. I'm just seeing a lot of negativity toward him nationally, with people wanting to bring the last guy back. I dont agree with that assessment with the voter and believe that would be disastrous for the country, but voters aren't particularly smart, I admit it, and yeah. That's where we're at.
We'll see if their model pans out but much like 2020, I'm fine sticking to my own.
For their record, their certainty Biden would win mirrored my own overconfidence, and that was fundamentally wrong. 62% was in retrospect a much better prediction, especially given the narrow margins, than around 90% was. And a 1.2% victory margin in the electoral college actually seems to track with that.
No comments:
Post a Comment