Friday, October 16, 2015

Political polling in the Canadian election - use with care

There was an interesting graph in yesterday's Globe and Mail, tracing support for the three main parties in the upcoming Canadian federal election, since the writ was dropped in early August up until today, just a few short days before election day, and mapping on it 14 major political events during that time which one might reasonably expect to affect support for one or more of the parties.
Unfortunately, the online version of the article does not include the actual graph, so I have actually laboriously scanned it (rather badly), and here it is:


The numbered political events are as follows:
  1. Aug 6: Trudeau holds his own in first debate
  2. Aug 12: Nigel Wright testifies at Duffy trial
  3. Aug 24: Chinese stock market rattles TSX
  4. Aug 27: Liberals promise deficits
  5. Sep 3: Alan Kurdi photo prompts calls for refugee action
  6. Sep 14: Conservatives trumpet $1.9-billion surplus
  7. Sep 16: NDP promises balanced budgets
  8. Sep 16: Conservatives vow appeal of niqab ruling
  9. Sep 24: Mulcair stands firm on niqab
  10. Sep 26: Terrorist stripped of citizenship
  11. Oct 2: Tories announce 'barbaric cultural practices' hotline
  12. Oct 5: TPP deal reached
  13. Oct 13: Harper appears with Fords
  14. Oct 14: Liberal campaign co-chair resigns
So, back on August 6th, the three parties were all bunched up, with the Conservatives (blue in the graph above) leading the polls at about 32%, followed by the NDP (orange) at 30%, and the Liberals (red) trailing at 28%. Two months later, the Liberals are showing at about 37%, followed by the Conservatives at just below 30%, and the NDP having sunk to 23%. In between these two dates, there have been various twists and turns, with, at one point, for example, the Conservatives way out of contention and the NDP in the lead, and another segment where all three parties are completely neck-and-neck-and-neck, swapping leads almost at random.
In fact, it is this apparent randomness that captured my attention. Looking at the graph, in relation to these events:
  • Yes, the Tories lost some support after complaints about their Syrian refugee policy in the wake of the Alan Kurdi photo, as predicted, but their fortunes had actually been sliding for some weeks before that, and indeed just a few days after that Conservative support shows a strong uptick, à propos of nothing in particular.
  • Justin Trudeau's out-on-a-limb promise of deficit spending at the end of August, which might have been expected to have a dramatic effect one way or another, actually seems to have had very minor consequences.
  • Events in early September, like the Conservative economic surplus news, the NDP's pledge of balanced budgets, and the much-ballyhooed Tory battle against the niqab in citizenship ceremonies, appears to have had next to no effect on relative party support (the complicated braiding in the middle section of the graph).
  • It was only when Thomas Mulcair came out strongly against the niqab ban on September 24th that the NDP's support begins to plummet to present levels, largely through perceptions of the issue in Quebec, and the Liberals appear to be the ones reaping the rewards of this, through no direct action of their own.
  • The resignation of the Liberal campaign co-chair, Dan Gagnier, is such a recent event that there is no way to predict how, if at all, Liberal support will be affected.
Of course, all of this is subject to many different interpretations, for example: people's political views are not strongly affected by day-to-day current affairs; people react unpredictably, even randomly, to political and economic events; polls are not good indicators of people's attitudes and voting intentions; etc.
The last of these may be especially important. Despite great advances in modelling and theory, the social science of political polling has taken a series of hits in recent years, right across the world, with several major elections yielding totally unexpected results, completely at odds with the predictions of respected polling organizations. A particularly good example of this is the "shy Tory effect", which I have described in more detail in a previous blog posting, whereby Conservatives are often too embarrassed to admit to their anti-social and selfish views until hidden behind the screen of a ballot box.
In fact, another recent article on the psychology of voting suggests that we are actually way less logical and reasoned in our voting habits than we might think ourselves. For example, however little sense it might make, studies have shown that voters tend to punish incumbent politicians for natural disasters (or other bad news over which the politicians clearly have no control) that occur just before an election. Also, however much we may feel that attack ads do not affect us, it is a proven fact that negative and ad hominem advertising does in fact lodge in our brains more than positive or purely factual advertising. Thankfully, though, studies have shown that attack ads are not actually all that effective in changing voter intentions, although - beware! - the later in the campaign the ads are shown, the more effective they are.
Even something as random as the gender of our children can affect our supposedly logical voting habits: parents ( and particularly fathers) with daughters are more likely to lean to the left than those with sons. What?! Paradoxically, perhaps, the more our views are directly challenged, the stronger and more entrenched our opinions become. And, finally, and perhaps most depressingly, we are more likely to believe information that is easily available (which, in this Internet age, could be of entirely unreliable provenance), and we are more likely to focus on aspects of an issue that are easier to understand (say, niqabs, for example) than on more complex aspects (say, the vagaries of taxation policy or carbon pricing), regardless of the relative importance of the issues. In fact, all in all, the less said about our emotional and inexplicable voting behaviour the better!
The other thing to be aware of in all this is the shortcomings of popular support polling in predicting seat allocations in Canada's first-past-the-post political system. Overall average support for a particular party does not necessarily translate to a similar level of representation in parliamentary seats, depending on how that support is distributed geographically. For example, the Conservatives won a healthy majority in 2011 with just 39% of the popular vote.
So, by all means make use the polls and predictions if you must (Three Hundred Eight and the CBC Poll Tracker are perhaps the most reliable). But - caveat emptor - use them with a healthy dose of skepticism, and don't expect the final result to bear much relation to even the most recent of polls.

UPDATE
For what it's worth, with the benefit of day-after-the-election 20-20 vision, we can conclude that maybe polling is not dead after all. The final count of the popular vote in the election itself compares pretty closely with the latest polls:
                                      Liberal           Conservative         NDP
Actual:                           39.5%                  31.9%            19.7%
Latest poll:                     37.2%                  30.0%            21.7%
In my own individual riding of Beaches-East York, the polls were also remarkably accurate (Liberals in poll 49.7%, actual 49.2%, NDP in poll 29.5%, actual 30.2%).

No comments: