Saturday, October 19, 2024

What my forecast would look like using different polling aggregators and my thoughts on it

 So, I decided to use some other polling aggregators to see how it would impact my forecast. This is kind of nuts, so first, let me repost my normal weekly election update from earlier:

This is what I get when I use RCP with just a tiny bit of 538 for Alaska data. Generally speaking, while the race is a tossup, it's very much leaning toward Trump. All 7 swing states are going in his direction, although for many it's by a point or less. Not a really solid prediction for Trump, still anyone's game, but concerning.

But let's say I switch tio 538, what would my election update look like?

This is a lot more like a forecast I had a couple weeks ago. It still has Trump winning but barely, the race is basically 49-51 instead of 41-59. While both are tossups, the one is a lot more...50-50. Not a huge statistically significant difference, but it is a difference. Their forecast has been about a point more blue than RCP's the entire time. So the trned is available there too. When I had the race 50-50, they would've had like 60-40 for Harris. When I had almost 60-40 for Harris, they might've been like 69-31 like you'd get with a 2 point lead, etc. So they have been more blue leaning, but with the race swinging toward trump, they've gone more into tossup territory, whereas im being pushed to the trump end of the tossup territory about to go lean trump.

Also, I must say, I do think 538 is a fairer aggregator in some ways. RCP used to be better. I do think they are being a bit more partisan. Like there was a morning consult poll that made it on 538, but not RCP. RCP USED to track morning consult, but they just removed them, especially as trumpers scream it's D biased the way we scream about the R pollsters flooding things. Curious how they'd not have that. They also removed PPP which is admittedly D leaning and that impacted results. So RCP isn't really bad for allowing republican polling I'd say, but they ARE sneakily removing D leaning polls from their average it seems. And I find that kind of problematic.

But wait, there's more. Let's look for a couple more aggregators.

I admit, after this, the aggregators kinda suck. RCP and 538 are the two giants of the industry. There are smaller ones but many are more limited, still, I found a third one that's decent enough called race to the white house. They actually do really good work in some ways and their methodologies seem very similar to mine. But at the same time i think they have the opposite problem of RCP. Their forecast is very blue leaning and as much an outlier in the D direction as RCP is in the R direction. Let's look at THEIR map.

Now we're solidly winning the rust belt again (although in tossup margins) and nevada. The sun belt is still pink, heck, both 538 and RTTWH have the sun belt being redder than I have it with RCP, but yeah on the whole their forecast seems quite...D leaning. Like...too D leaning. Still, they do some things I think are right. With them we finally got ME2 being likely R instead of a tossup (based on 2 polls with 1 poll being an extreme outlier). We also get oregon not being fricking lean D instead of safe D. I really think RCP drops the ball sometimes on these less polled states. Still...not as much as practically every other aggregator out there. Beyond that. Most of them just cover swing states. They dont cover the lean D/R ones as much. They just have...limited data. I did look at one here, and this is the best one of the others (the hill).

As you can see with half the states there isnt even a difference. They ended up with a forecast for swing states that's a lot like RCP's, with the GOP winning most states. it also had a weird new hampshire D+2 result which was just WTF. But beyond that they just lacked data.

There were others similar to this. Votehub is another up and coming aggregator but currently seems to only focus on swing states. And yeah...that's the aggregators for you. 

If I average the 4 maps together, I get this:


It seems to be just slightly redder than the 538 map. RCP and the Hill are more conservative leaning, 538 seems in the middle, and Race to the White House is very left leaning. 

I guess, if I were new to the polling game and not stuck in my ways, I'd switch to 538. They have the best aggregator IMO. And RCP is kind of dropping off hard. However, I dont really wanna switch my methods this late, so I'll continue using RCP for the rest of this election cycle. And if the race leans to the right, I might continue to use it. I think it's the best of the 4 currently. RCP is kinda going in the trumpy direction of removing D leaning polls but allowing their site to be flooded with R aggregated ones. And while it shifts things less than a point in itself to allow those polls in the aggregator, removing the D leaning ones has a larger effect vs 538/RTTWH.

Still, at the same time, if I took all of the deviations in the notes column mentioned above and averaged them out, I get almost the exact partisan lean as RCP has. It's just that it's distributed among the states differently. RCP seems to have more moderate results for the more D and R leaning states. meaning other aggregators will have D states that are 4-7 lean more like 6-10 on other sites. On the flip side, the same is true among R leaning swing states like Iowa and ME2. And I actually see their results strange and dont trust them much. I do think they ARE more R leaning. 

The big difference really comes down to the swing states. The sun belt is less pink, and the rust belt is more pink in their aggregator. Other sites seem to have the pattern we've been seeing where the rust belt is a bit more on the blue side of things, while the sun belt is more pink. Right now, RCP has them all over the place and the ordering is all messed up and chaotic. Normally I'd expect michigan and wisconsin to be more blue, north carolina, georgia, and arizona to be more red, PA to be in the middle, and nevada to be all over the place (it really is the wild card of the 7 and polls the most erratically). But instead RCP has them all mixed up with Michigan in particularly mingling with the sun belt states more than it normally does. This could be them culling the averages, or just a weird quirk of current data. But yeah, it's weird. 

Either way, I'm kind of losing respect for RCP somewhat as an aggregator. I've used them for literally 16 years now. They were THE original aggregator. They were the power players. I saw on wikipedia they were called the dow jones of political polling. But...it seems like they've been falling off the wagon in the age of trump and kind of shifting more and more right wing while not enough to lose all credibility, and at this point, 538 seems better.

At the same time, maybe I shouldnt talk crap about RCP until AFTER the election. After all, 538 IS overated. RCP got 2016 more right than anyone. They also got 2020 more right than anyone despite coming under fire for republican bias that election too. So...maybe they actually aren't bad, they actually got it right, and  538 is simply wrong. Wouldn't be the first time.

We'll see on election day. Anyway, I'll use RCP for the rest of this cycle for consistency's sake and think about where i go from here in future election cycles. if Trump wins and overperforms, RCP gets my support for another cycle. If Harris wins and overperforms and the polls are flat out wrong...well...I might be switching to 538 or another aggregator. We'll see *puts sword in ground like an anime character watching other people fight.*

EDIT: Also, let me just say that I still think, of the four covered, RCP is second best. I just trust 538 a little more at this point honestly. RTTWH is WAY too D leaning and way too new to the block. And the others aren't even...complete. 

So yeah, 538 > RCP > RTTWH > everything else IMO. Still using the second best site out there. I just think 538 might be better as an aggregator at this point as RCP makes some sketchy calls about what polls to include and not include sometimes.


No comments:

Post a Comment