26 October 2011

The Insiders on Polls during Elections #nlpoli

Just when you thought it was safe to stick your head up now that the poll-talk was gone, along comes three party insiders talking to Peter Mansbridge about polls.

Take the time to watch this video. The front segment on polls is short. You will learn a lot about how political strategists use their own polls to drive campaign decision-making.

And you’ll also hear a pretty frank and largely dismissive discussion about the polls you read in the media.  Most of that discussion will sound very familiar to you. 

The simple answer as to why the public polls are so spectacularly wrong is, as David Herle notes, that the public polls don’t look at voters.  They are actually looking at the population as a whole and with turn-outs dropping, those polls just don’t do a very good job at picking up on opinion in a progressively smaller bit of the electorate.

People lie to pollsters.  What a shock.  People lied to pollsters regularly and quite openly in the polls in this province during the recent general election. 

The pollsters won’t talk about that or their abysmal accuracy because the polls they release are marketing tools.  The media won’t talk about the wildly inaccurate polls because they are marketing tools for them as well.

Kathleen Monk adds a nice bit of colour on how the NDP used polling to determine the emphasis they placed on Jack during the last campaign compared to anything else.

And Jaime Watt adds the fine touch of noting that party people use a bunch of different information – he calls them data points – to figure out what is happening with the campaign.  Polling is one thing.  Canvassing is another and cash flow from donations is yet another of several points.

That last one will tell you why the spreadsheeters like threehundredeight.com go off in the trees.  Not only are they relying on inherently faulty data – those inaccurate public polls – but they rely on basically one type of data to try and forecast how seats were going.

Just to give you a sense of how inadequate that approach can be, realise that your humble e-scribbler chewed over with a colleague used a variation on the poll analysis approach. It turned up some curious things as the polls flowed in the last provincial election.  As the NDP numbers grew and the Tories dropped,  a bunch of seats showed as coming into play.

St. John’s Centre and East would look like they were swinging.

But so too did Virginia Waters.

And Bellevue.

And the Isles of Notre Dame.

The only way you’d cross those off the list of seats that might actually swing is by pulling in other sources of information.  The Straits never showed up on the chart and, frankly, without any signs of anything from that one seat, no one likely saw the change to the NDP coming.

For what it’s worth, your humble e–scribbler’s sister dazzled some of her townie Tory friends by naming seats in Sin Jawns that definitely flipped to the NDP. She never told them where she got the information but it came from an analysis of the polls and other tidbits.

You work with what you’ve got, even if it some of it is inaccurate, but with enough data points you can still build a pretty reliable picture of what’s going on.

The townies never saw  the changes coming largely because the media never reported any of the battle.  But after the votes were counted, her friends figured she was some sort of magician or witch.  A little information can make a lot of difference.

- srbp -