Federal Election Polls

ScottM · October 2, 2015

Fair point. Would you happen to have the article from Grenier? I can't seem to find it, and it does very much interest me! (I'm a huge political junkie )

I found it much more quickly than I expected. And I had the Liberals and NDP reversed. Anyway, here you go: http://www.threehundredeight.com/2014/12/whats-threshold-for-minority-or.html

angrypenguin · October 2, 2015

I found it much more quickly than I expected. And I had the Liberals and NDP reversed. Anyway, here you go: http://www.threehundredeight.com/2014/12/whats-threshold-for-minority-or.html

Thank you!

angrypenguin · October 2, 2015

Does anyone know what's going on with the EKOS poll? I would have expected them to public their results yesterday, but still nothing.

cybercoma · October 2, 2015

I still think Nanos is probably closest. That's where I get the idea that we have no idea.

Nanos will show a Conservative lead in a few days if the other pollsters are correct.

cybercoma · October 2, 2015

Something is messed up with polling in the last decade or so. That they're wrong is one thing, but they they are usually wrong in the same fashion (under-reporting conservative support) is quite another. Are Conservatives less likely to answer polls?

This is actually not true. Can you cite anything that backs this up? Most polls are within the margin of error, save for the couple provincial ones that were out. One of which got the popular support correct within the margin of error, but made the wrong prediction. The other was just way off.

cybercoma · October 2, 2015

Hey guys. A lot of pro poll readers on here. What do you all make of this?

http://www.thestar.com/news/federal-election/2015/10/01/what-the-poll-of-polls-says-about-the-federal-election.html

"The Signal" seems interesting.

Thanks for sharing. I've had an insanely busy week, so I can't take a look at the methodology now. However, this looks like a good source for an accurate prediction. I'll take a look through their methodology when I get chance, so I can better understand what they're doing differently. I notice that they accurately have the seat flip in Fredericton predicted, but they appear to be lowballing the CPC candidate. Last I heard the race was quite a bit closer.

Edited October 2, 2015 by cybercoma

angrypenguin · October 2, 2015

Thanks for sharing. I've had an insanely busy week, so I can't take a look at the methodology now. However, this looks like a good source for an accurate prediction. I'll take a look through their methodology when I get chance, so I can better understand what they're doing differently.

I'd love to get yours as well as others feedback. I took an extensive look and am quite fairly certain that's one of the more accurate polls. If you see on the top right, they also did update today their #'s. I'm fairly certain it wasn't around in 2011, so that's a downer. (for comparison sake)

Edited October 2, 2015 by angrypenguin

cybercoma · October 2, 2015

Nanos most recent poll (overnight) shows Liberals on the top again likely at the expense of NDP. I am both confused and dizzy by this. Nice surprise but are they playing Yo Yo with us!!.

Just to be clear, once again, the Liberals are not leading here. They're statistically tied with the CPC.

Hydraboss · October 2, 2015

Even though they weren't around in 2011, their model would allow them to retroactively rebuild their charts to see if they would have predicted correctly. Now THAT would be interesting.

CITIZEN_2015 · October 2, 2015

Just to be clear, once again, the Liberals are not leading here. They're statistically tied with the CPC.

The liberals are indeed leading HERE as 33.5 is greater than 31.9, but yes statistically speaking it is a tie due to 2.8% margin of error.

Edited October 2, 2015 by CITIZEN_2015

cybercoma · October 2, 2015

Leger falls in line with Nanos, while Forum and Abacus stand together in a stark contrast. The only thing they agree on is that the NDP is in trouble, though Leger takes that to an entirely new level with their Quebec numbers: 28% with the other three parties all within 7%. If that's true, they're in big trouble. After all, over half of their caucus was from Quebec following the 2011 election.

It gets even more bizarre when you realize the methods are similar. Both Léger and Angus use online panelists, while Forum uses Interactive Voice Response (automated system that allows you to speak in response or enter keypad presses to respond) and Nanos uses telephone (landline and cellular). Leger and Angus are showing opposite results, as are Nanos and Forum. It's crazy.

cybercoma · October 2, 2015

The liberals are indeed leading HERE as 33.5 is greater than 31.9, but yes statistically speaking it is a tie due to 2.8% margin of error.

Who cares about the polling sample. The only reason we poll is to make population inferences. In that sense, they're not leading. They're tied. It's only when you're discussing the specific sample drawn are they leading.

angrypenguin · October 2, 2015

And if you throw in Ekos in the mix it gets even more nuts.

Edited October 2, 2015 by angrypenguin

cybercoma · October 2, 2015

I'd love to get yours as well as others feedback. I took an extensive look and am quite fairly certain that's one of the more accurate polls. If you see on the top right, they also did update today their #'s. I'm fairly certain it wasn't around in 2011, so that's a downer. (for comparison sake)

At a glance, I don't see enough information on how they determine their dynamic bias modifiers. It's tough to say how accurate this will be as a result. Someone in the comments mentioned they don't control for methods (online vs telephone vs ivr, etc); however, I think that's safely accounted for in these "house bias" modifiers.

They also correctly mention that "even small differences in national vote share can have relatively large effects on seat share.” As a result, these modifiers can introduce large changes in the number of predicted seats. They’re meant to account for bias, but could be adding bias depending on how they're calculating them. We'll never know how they do that because this is probably considered a "trade secret" just as the various weightings that the firms use are considered a "trade secret."

I guess my biggest concern at the end of the day is how they're determining these biases, both in house bias as well as national polling bias. As far as I can tell, most of the bias that we see falls within the margin of error for the sample sizes that these firms use. It's good to have a polling aggregate like they use in the presidential election in the United States, but FPTP makes predictions a lot tougher because our system is based solely on riding-level results.

The absolute best poll you could do (aside from the election itself) would be to have roughly 300 respondents in every single riding. That's 101400 respondents across the country. However, you would need to contact far more than that to get sufficient responses. The response rates for Nanos telephone polling is roughly 9%, which is sad. That means to get 300 respondents in every riding, you would need to call over 3000 people. On a national level that's just over 1 million people that you would need to contact in order to get your 101400 respondents. The resources you would need to do that week-to-week is insane, but then you would have a riding-by-riding level analysis that's very accurate.

Edited October 2, 2015 by cybercoma

Hydraboss · October 2, 2015

Cyber, that's all true but even then your results would only be as accurate as the answers given, which may or may not match the end result. People lie to pollsters (although I have no idea why anyone would bother - either answer truthfully or decline).

angrypenguin · October 2, 2015

Hydra

They lie because it's stupid and they get paid to fill out surveys. My wife sadly does this.

Hydraboss · October 2, 2015

What?

I've never heard of anyone getting paid to participate in a poll.

edit -> I really need to participate in more polls!! How do I get on THAT list?

Edited October 2, 2015 by Hydraboss

cybercoma · October 2, 2015

Cyber, that's all true but even then your results would only be as accurate as the answers given, which may or may not match the end result. People lie to pollsters (although I have no idea why anyone would bother - either answer truthfully or decline).

This is usually accounted for in a variety of ways, but there's always that chance. What's clear is that although there are some liars, most of the time this isn't the case. The vast majority of people who do take the time to respond do so truthfully.

cybercoma · October 2, 2015

What?

I've never heard of anyone getting paid to participate in a poll.

edit -> I really need to participate in more polls!! How do I get on THAT list?

Join one of the pollsters' online panels. Unfortunately, I can't because I'm in what they call "a related industry." The thing is you have to answer a crapload of polls, including market research for businesses.

Edited October 2, 2015 by cybercoma

Hydraboss · October 2, 2015

Okay, that I knew about. I thought penguin meant people were getting paid by pollsters for "random phone" polls.

Smallc · October 2, 2015

Nanos will show a Conservative lead in a few days if the other pollsters are correct.

I don't believe that they are. I don't understand it if they are.

Hydraboss · October 2, 2015

This is usually accounted for in a variety of ways, but there's always that chance. What's clear is that although there are some liars, most of the time this isn't the case. The vast majority of people who do take the time to respond do so truthfully.

So my question is: How can they account for that unless they know who's lying? Look at MLW for instance; how many people have stated they lie to pollsters and how do you know that they're telling the truth about lying....

It just seems to me that the margin of error would have to include these suspected "liars" and their answers in the final result. Not sure how anyone could statistically quantify it past a "feeling".

cybercoma · October 2, 2015

I don't believe that they are. I don't understand it if they are.

Base emotions, pure and simple. Fear of the unknown. Xenophobia. Moral panic. Who's got time to understand the issues when you can just make decisions based on gut instincts? Never mind that the special RCMP task force is utterly irrelevant when all of the things they're targeting are already illegal. Never mind that stripping people of their citizenship and sending them to countries that are hotbeds of terrorism will actually contribute more to the problem. Never mind that not a single party has ever said we should let all the refugees in without any security checks whatsoever. The other parties side with the terrorists (you can tell because they support Islam and all Muslims are terrorists, at least that's what's heavily implied) and only the Tories are interested in fighting terrorism and keeping Canadians safe. If you elect anyone else, we'll have Sharia Law and ISIS in Canada by the end of the year.

To anyone with two braincells to rub together, this kind of hyperbole is mouth-breathing, knuckle-dragging garbage that does a disservice to intelligent debate on the real issue of world terrorism and Islamic extremism. But ain't nobody got time for that. F*** the Muslims. They're not "one of us." Keep Canada pure! Base reactions. That's all most people have time for. And when you consider 50% of the population has an IQ of 100 or less, do you think they're actually going to rationally consider these issues? Do you think they're going to engage in critical thought and question their assumptions? Hell no. And that's what you're seeing here. A hot button issue tapping into a moral panic since 9/11. People don't take the time to see grey on these things. It's black and white with nothing in between. Too bad the real world isn't that simple and these problems are infinitely complex.

jacee · October 2, 2015

So my question is: How can they account for that unless they know who's lying? Look at MLW for instance; how many people have stated they lie to pollsters and how do you know that they're telling the truth about lying....

It just seems to me that the margin of error would have to include these suspected "liars" and their answers in the final result. Not sure how anyone could statistically quantify it past a "feeling".

The margin of error takes into account variability in results that is not accounted for in other ways, so lying would be in there too.

cybercoma · October 2, 2015

So my question is: How can they account for that unless they know who's lying?

Considering multiple polls are conducted with roughly similar results, it's unlikely that lying has much of an effect. Honestly, people generally don't lie just for the sake of lying. It happens, but it's not going to be a substantial number of people out of the 2500 respondents in these polls. There are a number of response biases that researchers try to consider when they're polling, which is why they'll do things like randomly order the parties when giving them as possible answers. This controls for people just picking the first available response. Stuff like that.