RCP has a better track record than 538 in every election since 2016

[Redban]

This site puts 538 on a pedestal. Some of the most partisan and out-there users say that we shouldn't even post any poll unless 538 posts them too

But looking at actual results, 538 has not been meaningfully superior to RCP in terms of actual results

When 538 ever misses compared to RCP, they usually miss in favor of Democrats (i.e. 538 overestimate the Dems). I believe this tendency to show more favorable results to Dems is the sole reason 538 is preferred on this site

It is fair to prefer 538 because you think they include more polls or have a nicer layout. You cannot, however, prefer 538 because they are more reliable or accurate; they are not


2022

Actual house national result =  R+2.8
538 house generic national = R+1.2
RCP house generic national = R+2.5  (RCP 1.3% more accurate than 538)

538 was more accurate in the AZ  and NV senate race and MI Governor race, but RCP was more accurate in the NY governor and NY senate race.  538 and RCP were roughly equal in the senate races in GA, PA, WI and governor races in GA, PA,  NV, AZ

2020

Actual presidential result = D+4.5
538 national = D+8.4
RCP national = D+7.2  (RCP 1.2% more accurate than 538)

Actual florida result = R+3.3
538 florida average = D+2.5
RCP florida average = D+0.9 (RCP 1.6% more accurate than 538)

Actual AZ result = D+0.3
538 AZ average = D+2.6
RCP AZ average = D+0.9 (RCP 1.7% more accurate than 538)

Actual MI result = D+2.8
538 MI average = D+7.9
RCP MI average = D+4.2 (RCP 3.7% more accurate than 538)

Actual NV result = D+2.7
538 NV average = D+5.3
RCP NV average = D+2.4 (RCP 2.3% more accurate than 538)

Actual PA result = D+1.2
538 PA average  = D+4.7
RCP PA average = D+1.2 (RCP 3.5% more accurate than 538)

Actual WI result = D+0.7
538 WI average = D+8.4
RCP WI average  = D+6.7 (RCP 1.7% more accurate than 538)

538 and RCP were both equally off  in GA, MN, NC, OH, TX. In each instance, however, 538 overestimated Dems more than RCP did

2018

Actual house national result =  D+8.4
538 house generic national = D+8.6
RCP house generic national = D+7.3 (RCP 0.7% less accurate than 538, so they were about the same)

I don't see state senate / governor averages on 538 site, so I'll do no comparisons there

2016

Actual presidential result = D+2.1
538 national = D+3.9
RCP national = D+2.8  (RCP 1.1% more accurate than 538)

Actual florida result = R+1.2
538 florida average = D+0.6
RCP florida average = R+0.4 (RCP 1.0% more accurate than 538)

Actual NH result = D+0.3
538 NH average = D+3.5
RCP NH average = D+0.6 (RCP 3.1% more accurate than 538)

Actual MN result = D+1.5
538 MN average = D+6.7
RCP MN average = D+0.6 (RCP 3.1% more accurate than 538)

Actual MI result = R+0.3
538 NV average = D+4.3
RCP NV average = D+3.4 (RCP 1.0% more accurate than 538)

Actual NC result = R+3.7
538 NC average  = D+0.6
RCP NC average = R+1.0 (RCP 1.6% more accurate than 538)

Actual PA result = R+0.7
538 PA average = D+3.7
RCP PA average  = D+1.9 (RCP 1.8% more accurate than 538)

538 and RCP were both equally off  in IA, MI, OH, and WI.  In each instance, however, 538 overestimated Dems more than RCP did

[Randy Marsh]

They’re more transparent with what they use RCP includes arbitrary stuff sometimes

Also weighting by poll quality and house effects does seem to me a better model than just throwing everything in the average

But at the end of the day any polling average will be reasonably accurate though pretty off sometimes

[BlueSwan]

Most people here prefer 538 because their methodology is better and they are more transparent. But no  model is better than the underlying polls and in the last two presidential elections polling in general has underestimated Trump for various reasons. Since RCP often cherrypicks polls that are more favourable to the GOP it is not surprising that they "did better" than 538 in two presidential elections where reputable polls in general underestimated Trump. But the thing is that this is just RCP getting lucky - it is not RCP being "better".

[Amenhotep Bakari-Sellers]

Lol they Pred 53 RSand 230RH last just stop Redban gets nothing if Trump gets in office, Redban threaf

[Vern]

The best way to look at the state of the race is look at all the models and try to get a trend of what’s going on. No model is not without its flaws.

[Fmr. Gov. NickG]

So they weren’t more accurate in 2018, so the thread titlte is wrong?

In any case, I don’t think many people expect much of 538 any more since Nate Silver left.

[7,052,770]

This site puts 538 on a pedestal.

What? No we don't. It's not 2012 anymore. (And a lot of people didn't like 538 even back then.) You've got to keep up.

[Minnesota Mike]

RCP is just poll aggregator and it's fine. I use it alot myself because it's simple and goes back over decade.  The problem some people have is not the final product in November, it's just an average of the final polls after all, it's the spin and the uneven way they treat polls before the final averages. Polls with good Republican results just seem to remain in the averages longer than polls with good Democratic results. It's really NBD but it's real.

[Bush did 311]

An aggregator can be closer to national results and yet call more state by state races wrong. An anomalous result in Michigan or Pennsylvania is a drop in the bucket for the national average.

I think the national polling result from RCP has value but their state by state can be a mess. This is an issue everyone is having. National polls were not far off in 2016 or 2020. State and district polls have been getting worse for a variety of reasons.

[emailking]

by chance

[Progressive Pessimist]

Weren't they a disaster in 2022?

[Redban]

Weren't they a disaster in 2022?

If their performance in 2022 is considered a disaster, then 538 was a disaster too

[quesaisje]

I just want timely links to polls and a summary of topline results on a site that barely changes for decades.

Does 538 even have a trend line for head-to-head presidential polling yet?