Wednesday, May 17, 2006

[Technology, Polling]

The Shift to Cell Phones.

Among pollsters, the effect of cell phones on surveys has been the subject of some speculation, if little agreement. The folks at Pew decided to look into it a little bit, and the results are quite interesting. Do a substantial number of the population use only cell phones? Do they look different from landline users? Is this affecting poll numbers? Yes, yes, and no:
A new study of the issue finds that cell-only Americans – an estimated 7%-9% of the general public – are significantly different in many ways from those reachable on a landline. They are younger, less affluent, less likely to be married or to own their home, and more liberal on many political questions.

Yet despite these differences, the absence of this group from traditional telephone surveys has only a minimal impact on the results.
The differences between three populations--landline-only users, cell phone-only users, and dual users--is rather striking. In some cases, landline- and cell phone-only users are mirror opposites (only 10% of landline users are under thirty and 41% qualify for social security, while 48% of cell-only users are under 30 and only 4% are 65+) but in other cases, they're quite similar and contrast the dual users (about a quarter of single users have college degrees, compared to 40% of dual users). Fascinating stuff. You can see the results yourself--they're a little too detailed to discuss here.

In terms of polling, there are other interesting findings. Pew researchers found that it's possible to conduct cell-phone surveys, but it presents different challenges and is more expensive. They found:
  • The response rate was 30% in the landline frame but only 20% in the cell phone frame.
  • Half of the people reached in the landline sample (50%) cooperated with the interview, compared with roughly a quarter (28%) of those reached in the cell phone sample.
  • One consequence of this is that more people reached in the cell frame turned out to be ineligible because of their age than is typically the case in a household-based landline sample.
  • Data collection costs (apart from overall study design, programming, and analysis costs) were slightly more than twice as high for the cell phone sample as for the landline sample.
Because cell-phone users have to pay for incoming calls, Pew offered them $10 for their time--and this is the reason the expense was higher. It would be interesting to see whether cell phone users' participation differed without the inducement (once offered, 86% took the ten-spot).

My guess is that for the next several years, cell phone use will have a more pronounced effect on polling as it becomes a larger proportion of the sample. There will be a time when this begins to normalize, however, as cell phone use becomes commonplace. (And, given the age of landline-only users, it ain't far off.) Over the next few years, and particularly in the lead-up to 2008, we should start checking polling methodologies to see how they're weighting cell phone use. Two years from now, based on current trends, 15% of Americans will only use a cell phone, and that will be enough to skew findings.


Torrid said...

My old boss--Scott Keeter--headed up that analysis for Pew, I believe. He's an expert on what's called "nonresponse bias"--that is, how an inability to reach certain respondents, who are measurably different in some way from people who do respond, can skew polling.

Great guy, smart as a whip. I vouch for the accuracy of their findings, based on my experience with Scott.

iggi said...

i have no landline and have not had one in over 3 years...if that adds anything to the argument.

Jeff Alworth said...

Yeah, Pew is the gold standard in polling. Their methodology is impeccable, as is their survey design. I look at other polls, but I wait for Pew to find out what's really happening.

Cool about your boss. Do you do polling?

Jeff Alworth said...

Iggi, you're essentially a textbook cell-only user, down the line. It may be the only place where you're demographically average.

Torrid said...

I do statistical analysis these days, but I was part of a university polling house for 10 years, yeah. Really liked the work.