There is an interesting analysis of polling methodology by the Pew organization in America. Yeah, I know, if the reader of our blog is not deeply into American presidential politics, then polling (statistics) can be very boring. Even more boring is to discuss the market research methodology for all who are not stats-nuts like me, ha-ha.. But this is actually quite relevant and illustrates a growing discrepancy between the old way and the new way, and does apply to most media markets etc.
So, the basic premise of Pew's article by Keeter, Dimock and Christian at Pew, yesterday 23 Sept and entitled Cellphones and the 2008 Vote, is that if a telephone-interview based presidential poll has been conducted without cellphone polling (ie done only on fixed landline phones), it distorts the results. Their analysis of their own data, over several polls, finds that a balanced fixed landline and cellphone based poll, will add approximately 2 percentage points to Obama's advantage than a poll that is done using only landline phones. If this finding is true, it would help explain why some polls rather consistently in this year, have had an approx 2 point variance in their findings, when Obama had his bounce after the Democratic convention, or when McCain gained his two bounces, one with the announcement of Gov Palin as his VP, and then the Republican convention bounce; in various repeating national surveys, there has been a rather consistent 2 point difference.
ho
So, Mr Two Mobile Phones, what's going on? Good question. Pew is right and their statistical analysis of this "anomaly" has very sound sociological reasons. And they are reflective of an increasing bias between these two methodologies in conducting polls.
So in 2004, at the last Presidential election in America, the American cellphone penetration had just passed 50% per capita, and was still well below the fixed landline phone penetration. Essentially almost every cellphone owner would also have a fixed landline. Cellphones were just shifting from being executive "toys" to being mass market telecoms tools. There was an age (and sex) bias in cellphones, with employed adults having the majority of cellphones, and it was still a rich white man's tool, spreading beyond that user base.
Fast forward to 2008. Today American cellphone penetration rate is nearing 90%. There are more cellphones than fixed landlines. And most importantly, over 15% of Americans have "cut the cord" ie their households have no fixed landline, but have only a cellphone. This is not a USA-specific trend, all of these trends are global trends, reported in my books and on this blog for years, and were first observed in Finland and Scandinavia ten years ago.
This trend - of a cellphone-only household, is the first significant factor that causes bias in this year's election polling. It is not spread universally. Older people tend to have steady homes, often owning their homes and have had fixed landlines installed decades ago and keep their phones. Why not.
Younger people, especially first time employed, are often moving from one address to another, in search of jobs, and often move homes as they meet up with partners, and move out when then split up, then move in again when they get married, start families, move to larger homes for their kids, etc. This nomadic behaviour typically starts with American youth when they go to college, and will continue for years, until they perhaps settle down when they start to raise kids. Every time a young adult moves from one address to the next, there is the issue of deciding, do they still want to install a fixed landline phone at the next home. And since every American has a cellphone (90% penetration rate means its mostly pre-teen children and some very elderly, who are no longer without a cellphone), when you move to a new address, the cellphone will work fine (in most cases, most locations, in cities, with the caveats that American cellular networks do not all cover the full country and their network coverage is incredibly bad and patchy, compared to more advanced telecoms markets in Europe, Asia and Australia)
So if you are a pollster, and your polling system targets only fixed landline phones, you will miss out on all of these who are very economically viable, have a cellphone, but have decided no longer to maintain a fixed landline phone connection to the home.
Is that relevant? In some elections there is not a significant age split in support. It would then not matter. But this time, Republican Senator McCain, who is 72 years old, has his base of support with those over the age of 65, and Democratic Senator Obama, who is 47 years old, has his base of support with those under the age of 40.
Just on this factor, any pollster who does not sample the cellphone owning population, will be avoiding a significant part of Obama's main support base.
Lets dig a bit deeper. Perhaps a young voting-aged US adult who does have a landline phone, would behave and hold similar values to the similarly aged adult who only uses a cellphone. Maybe while there is an age based difference, perhaps there is no significant impact to presidential preference this year?
One of Obama's other key constituents is the university student age population. He had a consistent advantage in the primaries against Hillary Clinton in all college towns in America. Remember the thought, that Americans often leave home when they go to study at university. But not all go to university. Who doesn't? There are some who cannot afford to study. Then there are those with very poor academic results, and can't get into college. And finally, there are those who have found their career or life choice which does not require college - whether perhaps being a farmer or following into a parent's job say at a small factory etc; or perhaps especially for some young women, finding the husband and getting married and not wanting to go to college.
Many of these, who do not go to college, would then be likely to stay at home in the home of their parents.
So if the pollster is looking for young adults (to fit a nationally representative sample) and uses only fixed landline phones, that pollster will hit particularly heavily those young adults who still live with their parents.
They would severely under-represent college aged (not living with parents) young adults.
This I think is a very solid reason why not only age matters this year, but the college-educated sector of the youth (about half of American high school graduates do continue onto college, so this is a large sector of American voting age young adults)
But there is also a more profound impact this year. It is social networking (can you say "Communities Dominate" ha-ha). Senator Obama has used all forms of social networking and digital communication methods, from MySpace and Facebook to Twitter and SMS - remember his announcement of VP choice by SMS text message last month.
Among those who are active in digital social networks, Obama has a huge advantage in support. He has consistently led in building social networks and various methods on them for his campaign and those young Americans who are active in the social networking space, are particularly attracted to Obama, not only for his methods of speaking to them (and for being young and hip and different); but also for promoting consistently issues that appeal to young voters, college loans and programmes, green values, new jobs, etc etc etc.
Now how do these digitally hyperactive young adults correspond with cutting the cord in the home? Of course there is very strong overlap. There are plenty of heavily social networked young adults who have landlines - but more often than not, they use those only for broadband internet access, and if they need a landline type of phone service, will then use Skype for example.
So now, in this year's election, the most active of Senator Obama's supporters will fit the "cellphone only" phone population very strongly, and any pollster interviewing only fixed landline telephone based voters, will miss out on a disproportionate part of these hard-core Obama supporters.
Remember the big picture. At the end of last year, already 15% of all American households with telephone connection, had no fixed landline at all. But very reasonably, these are predominantly young and employed or young and in college, and thus very strongly leaning to support Obama. And the trend to more Americans cutting the cord has been growing at 2% per year, according to the CTIA. So its very likely 17% today.
This means that it is very likely, that those pollsters who use only fixed landline based interviews, will be creating a systematic bias in their finding in favour of Senator McCain and against Senator Obama. The Pew article suggested it would be in the order of 2%. My quick analysis on the USA age pyramid and a very crude model of likely voters (older people vote more reliably than young people), and I find the same order of magnitude.
Now, what this means is two things. Those who report on polls, should ask for a clarification on cellphone interviews. And those who conduct polls, please consider very carefully your methodology. 2% is a giant error when the national polls suggest the difference is about that, 2%, nationally. Using the wrong method would easily result in the pollster going from being totally right, to totally wrong, in predicting the actual election.
Recent Comments