Lack of wireless-only helps explain Rasmussen outliers

by: Chris Bowers

Mon May 24, 2010 at 11:57

A new study from the CDC shows that 24.5% of US households, and 22.9% of US adults, were wireless-only in the second half of 2009.  This is a sharp increase over the past two years.  In late 2007, 14.7% of US households, and 13.6% of US adults, were wireless-only.

This rapidly increasing level of wireless-only Americans will have an impact on political polling.  Specifically, excluding wireless-only adults from political surveys will have a statistically significant, negative impact on Democratic performance in political polling.  This was confirmed in a recent study by Pew, which compared the national generic ballot preference of a landline-only sample of 4,683 registered voters with a combined landline and cell-phone sample of 7,055 registered voters:

In the landline sample, Republican candidates have a 47%-to-41% margin over Democratic candidates on the 2010 generic horserace, but in the combined sample voters are evenly divided in their candidate preferences for this November (44% for each party).

There is still a margin of error in a poll with such a large sample size, but it is just barely over plus or minus 1%.  As such, with an overall six-point gap, the survey shows a statistically significant difference between polls that include cell-phone only adults and polls that do not.

Previous research has shown that Rasmussen's use of a likely voter screen is not the reason why their polls now differ strongly from the trendline of all other polls.  Rasmussen's likely voter polls are about six points more favorable to Republicans than other likely voter polls.  Also, Rasmussen polls of all adults are six points more favorable to Republicans than other polls of all adults.  This six-point pro-Republican tilt is exactly the gap found by Pew in their landline-only sample.

Since Rasmussen Reports excludes wireless-only adults from their surveys (possibly due to restrictions on automated phone calls to cell phones), it is likely that the wireless-only effect is one of the main reasons that Rasmussen surveys are now outlying compared to other polling outfits.  Further, the lack of wireless-only adults also plays a role in the recent, particularly eye-popping Rasmussen results in Kentucky (Rand Paul up by 25%) and Arkansas (Boozman up by 27% on Bill Halter).  In 2007, state-level data showed that cell-phone only adults were particularly prevalent in Arkansas and Kentucky.  While 13.6% of the nation as a whole was cell-phone only in 2007, in Arkansas that figure was 21.2% and in Kentucky it was 21.6% (PDF, page 5).  Given the trends nationwide, by May 2010 it is likely that roughly one in three adults in Arkansas and Kentucky are wireless-only.  This would make for an even more pronounced, localized gap than the national discrepancy found in Pew's 2010 study.

Additionally, given that New Jersey has one of the lowest rates of wireless-only adults in the country, this might also help explain why Rasmussen polls for the New Jersey Governor's campaign did not significantly differ from other polls in New Jersey.

The rapidly rise in wireless-only adults, along with the confirmation of the pro-Democratic tilt of those adults, helps explain some, and possibly all, of the difference between Rasmussen polls and other polling firms.  Americans are dumping landlines at a rapid rate, and those Americans doing do skew heavily Democratic.

Update: As an election forecaster, I honestly don't know what to do about the emerging gap between polls that include cell-phone only and those that do not.  About all I can say is that "it's a problem."  I strongly prefer to base my methodology on empirical studies of what has worked in the past when it comes to poll-based electoral forecasting, but the emergence of a very sizable, heavily-skewed cell-phone only population is a relatively new problem.  Polls from 2004-2008 will not offer much guidance here.

Chris Bowers :: Lack of wireless-only helps explain Rasmussen outliers

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

Well done, Chris. (4.00 / 2)
It makes a lot of sense, but I also do not trust Rasmussen since their "effect" tracks their owner's political views.  

Rasmussen is a powerful propaganga wing of GOP (4.00 / 1)
Rasmussen makes polls to shape the political landscape and create negatives narratives for their patrons' adversaries.  The biased phrasing of questions, the burying of any undesirable secondary numbers, the questions they chose to poll, the political races they choose to poll and those they choose NOT to poll, are all patent indications of political subservience.  Rasmussen polling of races well before election day is both more frequent, and more obviously skewed than in the weeks running up to election day where Rasmussen has considerable incentives to conform to electoral reality, which may unfortunately, have already been considerably shaped by their previous barrage of fabricated polls.  Some inside journalistic investigation into the "Rasmussen trade secret" would definately be beneficial.

Any thoughts on generic ballot polling? (0.00 / 0)
While this does explain Ras polling, it still doesn't explain congressional level polling.  Specifically, I'm thinking of PA-12.  The straight poll average of the 11 polls on wikipedia show 41.8-40.0 Critz - +1.8D.  Other than PPP and R2K, I haven't heard much about the other pollsters.  I'm assuming PPP and R2K include cell-phone users - don't know about the others.  For the three PPP/R2K polls the average was 46-42.7 Burns (R) - +3.3R.  Of course, the final result was +7.5D.  The polls were way off, favoring the R by over 10 points in the case of PPP/R2K.  Furthermore, I remember seeing an article about the NRCC saying their internal polling numbers were also way off - they truly thought they were investing money in a winnable race.  I know there is greater difficulty with polling special elections, but there still seems to be other factors at work here.  

About a year ago I remember there being a lot of chatter about tea-baggers telling pollsters they're "independent" when in fact they've never voted D in the their lives and should be more accurately counted as the R base.  With the PA-12 results, and comparing the number of people who voted in the KY D-Sen primary to the R primary, it seems like this is a factor that should be paid more attention.  It wouldn't be too difficult for pollsters to tease out this information.  In polls that establish voter party identification, poll participants should be asked who they voted for in the '00, '04 and '08 pres. races.  If those currently calling themselves independents voted more heavily R in those three pres. races than the polls in 00, '04 and '08 suggested, then we know there is a tea-bagger effect in play.  

In the case of Ras, the omission of cell-phone users means he is double biassing Ds - once in his political race polls, and again in his party identification polling.  Of course, he uses the latter to massage his raw political race polling numbers (party identification numbers are collected in separate polls than political race polls - right?) which compounds the error.  Beyond Ras, it is still worth considering there could be an effect from teabaggers lying about their party identification, even for pollsters that include cell-phone users.  Determining how this effects the current congressional ballot numbers may not be trivial, but there would be an effect.  

As a liberal D, maybe I'm just being too hopeful and trying to explain away poll results I don't like.  But it is hard to reconcile the PA-12 results and the KY D/R primary vote totals with the party identification numbers that pollsters currently use - and with the congressional ballot numbers that are currently being generated.  Whether there is a tea-bagger effect or not, there seems to be something in addition to cell phone-user omission negatively effecting D polling results.  I simply don't believe that if the election were held today the Dems would be crushed like all the polls predict.  Again, maybe I'm blinded by allegiances.  

This explanation doesn't (0.00 / 0)

Take the comparison of the two Florida Polls: one has Rubio up 8, the other Crist up 3.  

But the Job Approval numbers in the Rasmussen and the ISPSOS poll are nearly identical: 48 in Ipsos and 50 in Rasmussen.  

If this explanation was accurate you should see divergence in the Job Approval numbers, but you don't in the instance of the two Florida numbers.

Something else is at work in Florida.

That post should read (0.00 / 0)
Doesn't work for Florida.

[ Parent ]
Random Noise (0.00 / 0)
There's likely to be more than just one thing, but given how much noise there is in any small sample of polls, it's going to be quite a challenge to tease out, don't you think?

That's why I picked on the congressional ballot.  Ooooodles of polls.  A nice big fat curve over more than a year with almost no relationship to what everyone else was seeing.

Of course I'm interested in what you're onto (I once lived in Florida, on top of everything else).  But how would you go about getting a better picture?

"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 ]
I think the answer (0.00 / 0)
is look at state polls, which are more frequent, and which have far greater diversity in pollsters.

I will try and do a study over the next couple of days.

Something clearly stinks about Rasmussen.  

[ Parent ]
Sounds Good (0.00 / 0)
Though it's certainly challenging, since factors like % with landlines vs. cell phones only vary widely from state to state.

"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 ]
Excellent Summary/Analysis (0.00 / 0)
My only niggle is when you write

The rapidly rise in wireless-only adults, along with the confirmation of the pro-Democratic tilt of those adults, helps explain some, and possibly all, of the difference between Rasmussen polls and other polling firms.

I think it probably explains most of the difference, but as my diary yesterday showed, the differences with generic congressional ballot polling are different in kind.  They don't parallel other polls with a more-or-less constant bias, they tell a totally different story over time.

"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

Also (4.00 / 1)
I've noticed that Ras's Obama approval in state polling is much better than his national polling would suggest.

Fladem says his approval in Florida last week in a Ras poll was 50%. That doesn't possibly square with a national poll number below 50. Likewise, I saw a Cali poll where Os rating was his rating was above 60%. Again, doesn't square with his national numbers.

The one thing, however, is that maybe because of the particular demographics of those states, his approval numbers have held up very well. I think this is definetly true of California - there aren't many blue collar whites in the state. But I don't think this is true of Florida. More grist . . .

I have noticed the same thing (0.00 / 0)

[ Parent ]

Open Left Campaigns



Advanced Search

Powered by: SoapBlox