Simple question: how does RCP decide which Rasmussen polls to include?

[DLnyc]

Of course, one could ask why RCP (Real Clear Politics) includes any Rasmussen polls at all. But another good question, it seems to me, is which of these (consistently Republican-leaning) polls they decide to include.

On their "General Election: Trump vs. Clinton vs. Johnson vs. Stein" poll collection (http://www.realclearpolitics.com/epolls/2016/president/us/general_election_trump_vs_clinton_vs_johnson_vs_stein-5952.html), Rasmussen appears, as far as I can tell, at completely random moments.

At the top of the page, it shows only a Rasmussen poll ending on 10/23.

Below the graph, in the longer list, Rasmussen shows up 10/23, 10/20, 10/13, 10/10, 10/4, 10/2, 9/29, 9/28, 9/21, 9/13, 9/7, 8/30, 8/24, 8/16, 8/10, 8/2.

(Oddly, the list at the top skips the 10/20 one that appears below.)

I don't really see what the pattern might be.

Do they just throw one in when they need to pull Clinton's average down a bit? Do they roll some dice? Do they wait for a call from the RNC? Do they throw in the Rasmussen to make the International Business Daily one seem less stupid? I don't really get it.

[jamese777]

1. I can't explain

RCP's rationale.
RasmussenReports did not do well in 2012, "The final Rasmussen Reports daily tracking poll showed Mitt Romney with 49 percent national support and President Obama with 48 percent national support. Obama won the election by close to 4 percentage points.

A Fordham University study by Dr. Costas Panagopoulos compared pre-election polling with the results from election day. The study ranked Rasmussen Reports 24th out of 28 polls in accuracy, one slot above Gallup.

An analysis by Nate Silver on FiveThirtyEight ranked Rasmussen 20th out of 23 pollsters for accuracy in the 2012 elections with an average error of 4.2 points."

But in 2004 and 2008:
In the 2004 presidential election, "Rasmussen...beat most of their human competitors in the battleground states, often by large margins," according to Slate magazine. Rasmussen projected the 2004 presidential results within one percentage point of the actual vote totals earned by both George W. Bush and John Kerry.

In 2004, Slate said they “publicly doubted and privately derided Rasmussen” polls because of the methodology. However, after the election, they concluded that Rasmussen’s polls were the most accurae.

2008
According to Politico, "Rasmussen's final poll of the 2008 general election—showing Obama defeating Arizona Sen. John McCain 52 percent to 46 percent—closely mirrored the election's outcome." In reference to the 2008 presidential election, a Talking Points Memo article said, "Rasmussen's final polls had Obama ahead 52%-46%, which was nearly identical to Obama's final margin of 53%–46%, and made him one of the most accurate pollsters out there." An analysis by Costas Panagopolous in 2009 ranked 23 survey research organizations on the accuracy of their final, national pre-election polls based upon Obama's 7.2% margin of victory; the analysis determined that Rasmussen Reports was tied for 9th most accurate. Democracy Corps, Foxnews/Opinion Dynamic, CNN/Opinion Research, and Ipsos/McClatchy all predicted an accurate seven point spread.--Wikipedia

[DLnyc]

2. Interesting statistics, thanks! But it is easy to move to being accurate in the last few days.

The question is are their numbers several weeks out, like now, accurate, when they are consistently significantly to the right of the rest of the polling.

Also, while they may have accurately predicted the number of votes counted in Ohio in 2004, that number may not have been the actual number of votes cast in Ohio.

[Statistical]

4. That actually isn't bad. Remember margin of error.

If a poll has a 4% margin of error and says 49% Romney, 48% Obama it is actually saying there is a 95% confidence the outcome will be Romney having 47 to 51%.

Romney got 47.1% of the popular vote. If a poll shows a "winner" within the margin of error it is essentially saying the poll can't say who will win but the results with be within a few %. But yeah Rasmussen was worse than the others still I think it is important to show that even the "wrong" poll was close the range being predicted. This is what freepers fail to understand when they say the "polls get it wrong". Yeah they might not be exact but nobody in the past has shown one candidate up 12% and then they ended up losing. Polls might be off but not that off.

[Denzil_DC]

3. I gave up paying any attention to RCP during the 2008 presidential election

precisely because they blatantly cherrypicked which polls to include in their aggregate. You can probably guess which party they favored.

I don't know whether they've been as blatant since, but it used to be pretty transparent.

[DLnyc]

5. Yes, that is my point, thank you!

Rasmussen comes out every weekday:

"Rasmussen Reports updates its White House Watch survey daily Monday through Friday at 8:30 am Eastern based on a three-day rolling average of 1,500 Likely U.S. Voters."

http://www.rasmussenreports.com/public_content/politics/elections/election_2016/white_house_watch_oct24

But RCP doesn't include rasmussen every day. Nor do they include it every week. Instead, they seem to include it "sporadically". The most likely explanation, in my humble opinion, is that RCP uses the obviously-biased Rasmussen poll when they want to pull Clinton's numbers down in their averagel

As far as the obvious bias in Rasmussen, it doesn't help to look at the results from 2012 the day before the election, since of course they move their numbers toward reality right before the election, to improve their credibility.

Instead, let's look at Rasmussen next to the previous and following 3 polls, in the order shown in RCP:

CNN/ORC 10/20 - 10/23 779 LV 3.5 49 44 3 2 Clinton +5
Rasmussen Reports 10/19 - 10/23 1500 LV 2.5 41 43 5 3 Trump +2 <<=======
IBD/TIPP Tracking 10/18 - 10/23 815 LV 3.6 41 41 8 4 Tie
ABC News Tracking 10/20 - 10/22 874 LV 3.5 50 38 5 2 Clinton +12
Rasmussen Reports 10/18 - 10/20 1500 LV 2.5 41 43 5 3 Trump +2 <<=======
IBD/TIPP Tracking 10/16 - 10/21 791 LV 3.6 40 42 7 4 Trump +2
Quinnipiac 10/17 - 10/18 1007 LV 3.1 47 40 7 1 Clinton +7
Economist/YouGov 10/15 - 10/18 925 RV 3.9 42 38 6 1 Clinton +4
Average of other polls: C + 4.3% but Ras T + 2% (even including 2 IBD polls!)

CBS News 10/12 - 10/16 1189 LV 3.0 47 38 8 3 Clinton +9
NBC News/SM 10/10 - 10/16 24804 LV 1.0 46 40 8 4 Clinton +6
Boston Globe 10/11 - 10/14 845 LV 3.4 46 36 5 2 Clinton +10
Rasmussen Reports 10/11 - 10/13 1500 LV 2.5 41 43 6 2 Trump +2 <<=======
ABC News/Wash Post 10/10 - 10/13 740 LV 4.0 47 43 5 2 Clinton +4
NBC News/Wall St. Jrnl 10/10 - 10/13 905 LV 3.3 48 37 7 2 Clinton +11
FOX News 10/10 - 10/12 917 LV 3.0 45 38 7 3 Clinton +7
Average of other polls: C + 7.8% but Ras T + 2%

NBC News/Wall St. Jrnl 10/8 - 10/10 806 LV 3.5 46 37 8 2 Clinton +9
Reuters/Ipsos 10/6 - 10/10 2363 LV 2.2 44 37 6 2 Clinton +7
NBC News/Wall St. Jrnl 10/8 - 10/9 447 LV 4.6 46 35 9 2 Clinton +11
Rasmussen Reports 10/6 - 10/10 1500 LV 2.5 44 39 7 2 Clinton +5 <<=======
Economist/YouGov 10/7 - 10/8 971 RV 4.2 44 38 5 1 Clinton +6
Quinnipiac 10/5 - 10/6 1064 LV 3.0 45 40 6 3 Clinton +5
NBC News/SM 10/3 - 10/9 23329 LV 1.0 46 41 8 3 Clinton +5
Average of other polls: C + 7.2% but Ras C + 5%

Quinnipiac 10/5 - 10/6 1064 LV 3.0 45 40 6 3 Clinton +5
NBC News/SM 10/3 - 10/9 23329 LV 1.0 46 41 8 3 Clinton +5
FOX News 10/3 - 10/6 896 LV 3.0 44 42 6 2 Clinton +2
Rasmussen Reports 10/2 - 10/4 1500 LV 2.5 41 42 8 2 Trump +1 C <<=======
Gravis 10/3 - 10/3 1690 RV 2.4 44 44 5 1 Tie
Economist/YouGov 10/1 - 10/3 911 RV 3.9 43 40 5 3 Clinton +3
Reuters/Ipsos 9/29 - 10/3 1239 LV 3.2 42 36 8 2 Clinton +6
Average of other polls: C + 3.5% but Ras T + 1%

Economist/YouGov 10/1 - 10/3 911 RV 3.9 43 40 5 3 Clinton +3
Reuters/Ipsos 9/29 - 10/3 1239 LV 3.2 42 36 8 2 Clinton +6
CNN/ORC 9/28 - 10/2 1213 LV 3.0 47 42 7 2 Clinton +5
Rasmussen Reports 9/28 - 10/2 1500 LV 2.5 43 40 8 2 Clinton +3 <<=======
CBS News 9/28 - 10/2 1217 LV 4.0 45 41 8 3 Clinton +4
FOX News 9/27 - 9/29 911 LV 3.0 43 40 8 4 Clinton +3
NBC News/SM 9/26 - 10/2 26925 LV 1.0 46 40 9 3 Clinton +6
Average of other polls: C + 4.5% but Ras C + 3%

CBS News 9/28 - 10/2 1217 LV 4.0 45 41 8 3 Clinton +4
FOX News 9/27 - 9/29 911 LV 3.0 43 40 8 4 Clinton +3
NBC News/SM 9/26 - 10/2 26925 LV 1.0 46 40 9 3 Clinton +6
Rasmussen Reports 9/27 - 9/29 1500 LV 2.5 43 42 6 2 Clinton +1 <<=======
PPP (D) 9/27 - 9/28 933 LV 3.2 44 40 6 1 Clinton +4
Rasmussen Reports 9/26 - 9/28 1500 LV 2.5 42 41 7 2 Clinton +1 <<=======
Reuters/Ipsos 9/22 - 9/26 1041 LV 3.5 42 38 7 2 Clinton +4
Quinnipiac 9/22 - 9/25 1115 LV 2.9 44 43 8 2 Clinton +1
Average of other polls: C + 3.7% but Ras C + 1%

I could go on, but I think the pattern is pretty obvious: Rasmussen runs well to the right of the average of other polls! (Also, they seem to be getting more ridiculous in the last week or two.)