02/05/16

Election polls: How scientific are they?

ballot_boxes_flickr_Keith_Bacongco
Copyright: Flickr / Keith Bacongco

Send to a friend

The details you provide on this page will not be used to send unsolicited email, and will not be sold to a 3rd party. See privacy policy.

[MANILA] It’s been said often enough that statistics don’t lie, people do.

Nowhere is this perhaps more apparent than in the research field of election polls, where what we see as factual numbers may not be the facts we are led to believe.

In the Philippines, pollsters are as much in the centre of hotly contested presidential elections, with one candidate assailing what she terms as “voodoo surveys” that have “become the kingmakers and not the voters”.

For physicist Giovanni Tapang of the University of the Philippines Diliman, there’s little doubt that poll surveys — “if done properly” — can be a “useful instrument to arrive at a sampling of public opinion”.

“Although the number of respondents may seem small, typically 1,200 or so, statistical survey methodologies provide a powerful tool to gauge the general pulse from only a handful of people,” he notes.

There’s a branch of political science — or mathematics, depending on who, Tapang says, conducts this — devoted to studying elections. Called psephology, from the Greek word “psephos” or pebble used by ancient Greeks to cast votes, it uses statistical analysis and game theory to analyse elections and opinion polls.

Tapang says the problems manifest when there is lack of transparency in choosing the sample groups, which should be done in a random manner. Additionally, the way survey questions are framed — and sequenced — can create bias against certain candidates while favoring their opponents whose side in fact paid for the survey.

In this regard, it’s important to look at the progeny of surveys to determine their credibility. Who conducted them? Who commissioned them? Are the funders identified with political interests?

Another problem arises when polls tend to undersample block and regional voting, including the vote delivery of religious and political blocs. In semi-feudalistic settings like the Philippines, where patronage politics is the norm, undersampling can heavily skew poll estimates.

Because of the inexactitude of the models and methodologies of election polling, no wonder then it gets a bad rap for being fuzzy science or even pseudoscience.

For Tapang, the simple challenge to pollsters is to show how and why their surveys are non-biased. Put another way, it’s for the pollsters to disprove that “facts are stubborn things, but statistics are pliable”.

This piece was produced by SciDev.Net’s South-East Asia & Pacific desk.