Monday, October 27, 2014

missed call: the influence of cell phone culture on political polls


Politics and prophecy have ancient mutual origins in military tradition. It is obvious why knowledge of the future confers strategic advantage. Once a tradition of mysticism and ritual, prophecy now involves the application of algorithmic calculation to large data sets for the production of useful extrapolations. This is how finance capitalism evaluates companies, how Target uses sales data to know about a woman’s pregnancy before she does, and how campaigning politicians know which doors to knock on or avoid. In the era of big data, we should not be surprised that big money remains the dominant influence.

If it seems as though new, contradictory polls are produced daily, then we can thank the news media for increasingly relying on polling data to provide inexpensive programming. Commercial news is an entertainment product, a consequence of media conglomeration by large multinationals. In this context, polls quantify the drama of the electoral road and turn the relative boredom of electioneering into an adult videogame formatted for inexpensive mass consumption. Of course, without editorial discretion on the part of media agencies, this process often results in the publication of polls bearing dubious statistical legitimacy.

Gauging public opinion requires time to properly accomplish. Survey length and complexity dictates cost, and media organizations need to produce other content while waiting for the survey to be completed. As a result, new survey techniques which greatly simplify survey questions while reducing the time and budget required for data collection have come to the fore in the prediction industry, with the resultant products ready for media consumption. Some polling companies such as Angus Reid and Abacus Data have transitioned to online polls of dubious legitimacy. Most companies, such as MainstreetTechnologies and Forum Research – often cited in Toronto media – use interactive voice response (IVR) technology, a self-aggrandizing term for computerised phone surveys.

So what exactly is the problem with telephone polling in the 21st century? Telephone collection of public opinion data from a random selection of Canadians has long been the gold standard for the polling industry, as landlines existed in virtually every residence in the country and data could be collected in a cost-effective manner. However, academic and industry studies have noted that the recent decline in the response rate to telephone surveys has greatly impacted the validity of data produced. Reasons for declining response rates are numerous, but often involve technological developments such as line screening and the adoption of mobile phones. Unlike the phone books which graced every home when landlines were common, wireless carriers have not coordinated their databases to produce a national cellphone directory. Furthermore, due to built-in caller ID and pay-by-the-minute billing, cell phone users are more prone to ignore calls from unknown numbers. As a result of these issues, many telephone surveys omit cellphones from their sample sets, as it is difficult and expensive to correlate demographic information with individual numbers.

Youth, urban professionals under the age of 40, renters, and low-income voters in particular are not being captured by polls relying on landline survey data. Governmental research suggests that mobile-exclusive residences currently represent nearly 19% of Canadian households, a number that is sure to rise as nearly 65% of people under 35 report using mobile phones exclusively. As a result, poll data is skewed toward older, wealthier voters in rural and suburban communities, reflecting a bias for conservative candidates. This bias evidences in polls as reported by the news media, but often vanishes once votes are actually counted on election day: witness the last Ontario election, in which poll data almost universally predicted a Conservative victory, while the actual election granted a majority win for the Liberal party. In a similar manner, Olivia Chow’s popularity lies with demographic groups not captured by landline surveys and so may not be reflected in poll results indicating a race between John Tory and Doug Ford.

According to polling companies, the use of IVR along with advanced statistical analysis results in a rate of predictive accuracy comparable to landline telephone surveys and other established methods for gauging public opinion. However, more often than not, polling companies simply do not perform the requisite statistical calibration to legitimate their results, suggesting that their data acquisition methodologies emphasize turnaround time and affordability rather than statistical viability. My own calculations indicate that IVR is only accurate when the results of numerous polls are averaged over a much longer term than the daily surveys being reported in the news media. Importantly, the long term trend is not reflected by individual studies, which vary wildly from the long-term median.

As a result of focusing on short-term results skewed by unrepresentative population samples, the news media often misrepresents public opinion to the voting public. With an increasing number of miscalled elections, hopefully the public learns the sense of editorial mistrust and critical evaluation which the news media, in thrall to the temporal acceleration of market forces, have relinquished. 


Published for rabble.ca 

No comments: