There is precedent for Rasmussen erring huge in favor of Republicans

by: Chris Bowers

Mon Apr 19, 2010 at 14:43


Rasmussen Reports has become a controversial polling outfit in this election cycle, for two reasons: a massive increase in the amount of polling it produces and its newfound "house effect" in favor of Republican candidates.

First, during 2009-2010, Rasmussen Reports has produced has drastically increased its quantity of polling.  For example, in addition to their daily tracking poll on President Obama's job performance and their weekly poll on the generic congressional ballot, 28% of all the polls conducted on the 2010 Senate elections, up from 18% in 2007-2008.

Second, and most importantly, Rasmussen has become controversial because its massive amount of polling has skewed decidedly in favor of Republican candidates.  Rasmussen surveys have favored Republicans by 5.5% compared to other polling firms so far in 2009-2010.  In 2007-2008, Rasmussen was in line with all other surveys, showing no House effect at all.

As Nate Silver showed over the weekend, this newfound skew in favor of Republicans is not simply the result of Rasmussen surveying likely voters while other polling firms are looking at registered voters or all adults.  Comparing apples to apples, Rasmussen's likely voter generic congressional ballot polls favor Republicans by 6% compared other likely voter generic congressional ballots.  Further, Rasmussen's numbers for partisan identification among all adults also favor Republicans by about 6% compared to other polls that measure partisan identification among all adults.  

To reiterate: Rasmussen's likely voter polls are skewing 6% in favor of Republicans compared to other likely voter polls, and Rasmussen's polls of all adults are skewing 6% in favor of Republicans compared to other polls of all adults.  So, this has nothing to do with an enthusiasm gap or a likely voter screen.  Rasmussen is simply flooding the zone with polls that skew towards Republicans by an average of 6%.

The most likely response defenders of Rasmussen polls will make to these numbers is to point at Rasmussen's strong track record of predicting election results from 2004-2008.  However, I am reminded of an earlier election when Rasmussen erred massively due to an enormous, pro-Republican "House effect."  Back on November 7th, 2000, when operating the "Portrait of America" poll, Rasmussen projected Bush to win the national popular vote by 9% over Gore, 49%-40%.  The average of the other 11 polls in the national trial heat only put Bush up by 1.5%, which means that Rasmussen has a "house effect" of 7.5% in favor of Republicans:


The difference between Rasmussen and all other polling firms in 2000 sure looks a lot like the difference between Rasmussen and other polling firms now.

Finally, even though I don't enjoy making these implications, it is worth noting that in the 2000 election, Rasmussen's primary media sponsor was World Net Daily.  While I have absolutely no proof at all that WND's sponsorship of Rasmussen played a role in Rasmussen's massive, outlying house effect in favor of Republicans, it certainly does raise an eyebrow or two.  Similarly, in 2009, Rasmussen scored significant new capital investments.  Again, even though there is no apparent connection between the new, pro-Republican "house effect" in Rasmussen and their new financial partnership, it raises eyebrows.

Calling veracity of individual polls into doubt has an increasingly long, but very checkered, history.  From 2004-2008, simply averaging the results from almost all polls over the final 15 days of a campaign, without any weighting at all, has proven to be the most reliable method of forecasting general election results.  Zogby Interactive, Strategic Vision, and Columbus Dispatch polling are the only current examples of polls that are so inherently flawed it is better to just ignore them altogether.  I seriously doubt that Rasmussen will end up on that scrap heap, but given their vastly increased quantity of polling, their newfound (or re-emerged) pro-Republican "house effect" can't simply be dismissed.  Something has happened to Rasmussen polling, and more answers are needed.  

Chris Bowers :: There is precedent for Rasmussen erring huge in favor of Republicans

Tags: , , (All Tags)
Print Friendly View Send As Email

how did Rasmussen do in forecasting in the 2009 races it polled? (0.00 / 0)
For example, Scott Brown, MA?

Strangely enough... (4.00 / 1)
I think Rasmussen is practically responsible for Scott Brown's win in MA.  Other polls were showing a 15 point or so lead for Coakley, then Rasmussen came out with a poll showing that cut in half and not only did that significantly spike the enthusiasm for him among conservatives, he basically received non-stop positive news coverage from that point on (even after a couple more polls around the same period still supposedly showing Coakley up double-digits, if I recall correctly).

Even though polling is trying to measure something, it actually helps create narratives in today's media environment.  The mysterious influx of capital along with their significant pro-GOP bias (and proven results in MA) should raise a lot more than eyebrows.


[ Parent ]
I disagree (4.00 / 1)
Rasmussen if anything was the only pollster that moved DEMS into action: Republican grassroots, bloggers were already energized from mid-December on. And I remember reading the NRSC commissioned a December poll showing a competitive race.

But Coakley was sleeping, as was the DSCC and DNC - and Rasmussen is the first thing that made any of these wake up.

Also remember that PPP polled within days of Rasmussen and found an even tighter race - and it was dismissed by everyone until all polls confirmed it.


[ Parent ]
Could be.. (0.00 / 0)
Though I don't think Dems were really spurred into action until about a week+ left in the campaign... pretty much ignoring the Ras poll, particularly after Coakley and... I think it was Quinnipiac or something showed something like a 14 and 10 point race in favor of Coakley in the same time period... I think PPP then polled a few days later because of the odd competitiveness shown by the Ras poll, at which point the conservative frenzy was already upon us (and thus pushing the race into highly competitive territory).

Not that any of this really matters in the end as Brown ended up legitimately winning the election, and that's what actually matters.  I just wonder whether he would've been able to if no poll had ever shown the race within 10 points.


[ Parent ]
Didn't poll in the last week (0.00 / 0)
In MA, it didn't produce a poll during the last week, so its hard to say.

In New Jersey, it only had a 2-3% "house effect," and was much more in line with other polls. Further, this effect was generated almost entirely by accurately finding less support for the third-party candidate than other polls.

In Virginia, it also showed no house effect. Was in line with other pollsters.

So really, it is odd. They didn't stick out from other polling firms from 2004-2008, or in the 2009 elections. Strange that their polls are sticking out otherwise.


[ Parent ]
Didn't Silver suggest an answer to this? (0.00 / 0)
Any pollster wants to be close when it comes to final results, so there's really no room for bias in predictive polls close to election day. When it comes to approval polls, or to election polls far ahead of election day, there's no real check: the only standard to judge their accuracy by is what other polls say, so all you get is a "he said she said" dead end. Therefore you're free to inject bias into these polls without consequences.

[ Parent ]
he said that (0.00 / 0)
but that's not enough to cover a fraudulent pollster. If your 1-week from election polls suddenly moved away from your previous polling (in several races) it would need explanation. You couldn't always explain it as a last-minute shift in the race if other pollsters don't show those shifts.

[ Parent ]
Ideally (0.00 / 0)
Pollsters should strive to use the same screens and weights in polls far in advance and those close to election day so that there is solid comparison over time.

Things You Don't Talk About in Polite Company: Religion, Politics, the Occasional Intersection of Both

[ Parent ]
Since polling is now a propaganda tool (0.00 / 0)
as much as one for analysis, it may be time to consider regulating it. If polling is really scientific as claimed, there is no excuse for allowing pollsters from withholding data like sampling methods and voter screens. Everything should be open enough for everybody to test their validity for themselves.  

Polling creates its own environment (4.00 / 3)
Polling can drive not just the narrative, but the voters themselves.

I think them flooding the zone withlarge numbers of polls is meant to do that.


"Incrementalism isn't a different path to the same place, it could be a different path to a different place"
Stoller


So is the wisest response (given a world (0.00 / 0)
where we had bottomless wallet) just paying ARG or Kaiser/Whatever, or another pollster with the reverse house effect, to flood the zone even more?

What would that cost, I wonder.


[ Parent ]
yes good idea (0.00 / 0)
though there are ways to ask questions and to sample populatins that can effect the results.

I assume our folks are not going to be engaging in that so it won't counter all of their flooding the zone

"Incrementalism isn't a different path to the same place, it could be a different path to a different place"
Stoller


[ Parent ]
Someone (0.00 / 0)
actually remembers the Portrait of America disaster other than me....

Rassmussen since then has been ok - though their Bush apporval ratings are suspect.

What is really important at this stage for people to remember when looking at polls is that well over 50% of the state polls are from four pollsters: Rasmussen, PPP, R3K and Quinnipiac.

If one of those are wrong, the average is going to be wrong.    


Should we be at all concerned about the house effect of Research 2000? (0.00 / 0)
According to Silver's graph, Rasmussen has a +5.5 pro-Republican house effect, but R2K has a +4.4 pro-Democratic house effect.

Things You Don't Talk About in Polite Company: Religion, Politics, the Occasional Intersection of Both

I Haven't Looked At Silver's Diary (0.00 / 0)
but as noted below, back on March 22, the collection of polls that Chris used showed a much more modest house effect for R2000.  The Economist had a much larger pro-Dem house effect.

"You know what they say -- those of us who fail history... doomed to repeat it in summer school." -- Buffy The Vampire Slayer, Season 6, Episode 3

[ Parent ]
Well, I'm Glad To See Some Followup To My Earlier Growsing About Rasmussen (4.00 / 1)
In responses to Chris's March 22nd diary, "National House Ballot, post-hcr baseline", I responded with this:

Rasmussen Distorts This Picture

Unsurprisingly, Rassmussen is skewing these results.

The 4 Rasmussen polls average Dem: 35.75, Rep: 44.25, for a -8.5 margin, compared to all the rest.

Removing Rasmussen for the calculations produces an average of Dem: 43.16, Rep: 41.11, for a 2.05 margin.

You might think the 4 DKos polls would be big outlier the other way.  But you'd wrong: DKos polls average Dem: 43.75, Rep: 41.25, for a 2.5 margin, just modestly higher than the non-Rasmussen average.

You'd be better off looking to the Economist's 4 polls.  Their average: Dem: 45, Rep: 39.75, for a 5.25 margin.  Damn socialist capitalist running dogs!

If you simply replaced the 4 Rasmussen polls with a single poll entry averaging all four, this would result in Dem: 42.79, Rep: 41.26, for a 1.53 margin, which is my hunch for what's probably the most accurate.

If you replace ALL multiple polls with a single poll entry averaging them, the result would be Dem: 42.13, Rep: 41.65, for a 0.48 margin.

Which, IMHO, is probably the outer bound for how well the Reps are doing right now.

Bottom line: Trust, but verify.  Only without the whole "trust" part.

"You know what they say -- those of us who fail history... doomed to repeat it in summer school." -- Buffy The Vampire Slayer, Season 6, Episode 3


I hate to be critical of you, (0.00 / 0)
but I can't actually read this article.  It's too garbled.  You need to proofread.

USER MENU

Open Left Campaigns

SEARCH

   

Advanced Search

QUICK HITS
STATE BLOGS
Powered by: SoapBlox