May 10th 2013: Don't be fooled, new polls don't show a 50-50 race

With the two new polls released yesterday, in particular the Oracle Poll showing a 4-points gap only between NDP and Liberals, some might be tempted to think that we now have a wild race where everything can happen. News medias sell more if they write about a race, remember the last US election?

By averaging the two polls published yesterday (so the two most recent polls), we'd get the NDP around 41%, the BC Liberals at almost 36% and the Green and Conservatives at 12.8% and 8.8% respectively. This is still a 5-points lead for the NDP. This is actually not significant. Indeed, for these levels of supports and with a sample size of 1000, the lead needs to be at least 5.4 points to be significant at 95% confidence interval). But given that we are actually using two polls with total sample size of 1800, the NDP lead is indeed significant. But remember, not being significant only means that it's possible for the Liberals to actually be ahead. It doesn't mean it's likely to be the case! If you prefer, it means that while the Liberals could be ahead of the NDP, it's just as equally possible for the NDP to actually have a very important lead of almost 13 points (margins of error work both way, remember that!).

If I run my 1000 simulations using these numbers, I get that Adrian Dix would win the election 84% of the time! Remember, this accounts for the uncertainty (or randomness) due to the polls as well as the electoral system. So it shows clearly that while we do have a much closer race than at the beginning of this campaign, the odds are still in the BC NDP favor.

Even if I was to only use the best poll for the Liberals (the Oracle poll), where Christy Clark would only be 4-points behind, I'd still find that the NDP would have 76% chances of winning. But there, any over- or under-estimation by the polls is becoming critical as each point translates into many more seats and chances of winning.

In conclusion, yes polls aren't perfectly accurate and can be quite wrong at times, but you have to wonder how likely it is. You can mention the Alberta election, but you should also remember that polls are not that wrong in average. My model actually accounts for that by going way beyond the traditional margins of error. So yes it's possible for the BC Liberals to win next Tuesday, but it isn't likely. Still, for Christy Clark, the simple fact that a victory is possible is already a huge improvement over simply 2 weeks ago!