Green support growing nationally
Hi. I have read a number of posts that suggest that the growth in support for the Green Party is somehow soft. The following graph shows the average of all the public polls I can get my hands on, averaged by quarter, showing the change in popular support versus the 2006 election. Most of these points have 10,000 or more respondents in them, and have a number of different polling companies represented. Margin of error is about 1% to 1.5% 19 times out of 20.
I don't think the graph could be clearer. It shows steady and consistent growth in support for the Green Party at the expense of all the others. Two years worth of consistent growth.
I would also like to address a common myth. Green support does not evaporate at election time. Only a small amount (approx 0.5% to 1%) has been torn away in past elections, likely by strategic voting, once the voters go to the ballot box. This happens for many parties, some win a little, some lose a little, depending on the news of the day. The magnitude of this change is almost insignificant.
The support we see on this graph is the sort of strength you see in a healthy and growing organism. Consistent and relentless.
Jim Johnston,
Lambton-Kent-Middlesex
Opinions expressed are my own.
- Jim Johnston's blog
- Login or register to post comments
Comments
Pooling polling
The general point about "healthy" growth is well-taken. I think there just has been reticence about interpretation of single polls. Also, since single polls must be treated as singular, it is disputable whether it makes fuller statistical sense to diminish error margins by lumping variegated methods. If one poll is "off" by 5% or so from an "actual" support of <10%, that is an enormous comparative factor, although it might make little effective difference to capture of seats. Lumping such a poll with other methodologies yielding closer results skews, and it is not legitimate to add 'em all up cross-methodologically. Now one might expect that methodologies do not differ all that much, so it is somewhat instructive to examine them all together. But if, say, one polling firm with a client or connexion eager to diminish the perception of a viable Green alternative, if such a firm asks which party the respondent would support without listing the party names, perhaps the Greens would show more poorly. That kind of poll cannot at all have its "error margin" lumped with others that, say, openly list the parties for respondents. The latter can be tallied with other polls that vary the order of the parties listed, and arguably that varied presentation for respondents actually serves to diminish the "error margin". But these methodologies are proprietary, are they not? Until enough information is had, pooling polling to diminish error factors is not really all that useful, or even legitimate. There is no doubt, however, let me repeat, that the trend is up, healthy, and the draw is away from all parties.
Good observations
Thanks for the observations, Daryl. I worked in the research industry for several years, and the methodology is basically the same in all companies - you have a survey instrument which has a specific order of questions, including random rotation of prompted questions. Then you select a random sample, typically through either random digit dialing, or from a database of known active phone numbers. Then you qualify whoever answers the phone, and collect the data. Afterward, the data is weighted to correct for sampling errors in, say, the gender of respondents, age, vocation or geography.
Bias can creep into the results during any or all of these processes. Since we are looking at a number of different polling companies combined, the net effect is that any systematic bias by one company is offset by the inclusion of more surveys. This takes advantage of the statistical principle of regression towards the mean .. that is, the more sample points, the more likely you are to be close to the theoretical "right" answer (which, of course, nobody knows). This is the same principle that allows us to predict the margin of error based on the sample size .. as the sample size increases, the margin of error rooted in sampling decreases.
So, it works out pretty well to aggregate the multiple polls together, and almost all of the time, leads to a better, more statistically reliable result. There are sometimes good reasons to exclude "outlying" points, but that is a pretty rare circumstance.
Jim Johnston,
Lambton-Kent-Middlesex
Opinions expressed are my own
methods & margins
A bit of consultation with someone with training in statistics backs up my main concern, that the claim about error margin seems illegitimate if it is taken as referring to the aggregate of poll respondents, this because of possibly differing methodologies. Even if "the methodology is basically the same in all companies", there must be more analytical clarity before a claim that aggregating companies' results effectively drops the error margin. This concern would follow as well from Jim Harris' comment at http://greenparty.ca/en/node/3199#comment-2396, "[...] the company is under scrutiny over its methodology." Still, if understood as an exercise different from bearing on diminished error margins, as simply a study of pollsters & their polls, it is welcome & of some importance.
Hey Daryl This would be
Hey Daryl
This would be interesting to discuss with Greg Morrow of DemocraticSpace. When you average these polls together he adds all the sample sizes and then re-calculates the margin of error based on the larger sample size.
Jim
Jim Johnston's Analysis is Fantastic
Jim's analysis is simple, straightforward and presents undeniable trends. In short it's FANTASTIC!
Thank you, thank you, thank you for this!
I've written about Jim's methodology in GPC's Relentless Rise: Ipsos-Reid Under-estimates Green Support by 2%: Inflates Conservative's Support 5%+ which you can read at http://greenparty.ca/en/node/3199 . (The discussion on Jim's work is at the top of the post).
Jim
Such a beautiful graph: Maybe I'll make it my desktop background
Jim
This is such a beautiful graph I can't stop staring at it. . .
Maybe I'll make it my desktop background image
Jim
Love the idea - can someone post?
Ooh, I like that, but want one better.
Can someone put together this GPC-growth graph (the one that has only us above the zero) with the GPC 3rd place screen capture (the one that haunts Jack Layton when he closes his eyes) into a single image that I could grab and use as my desktop background? That would be nice to show off every time I open my notebook.
Erich Jacoby-Hawkins
Barrie, ON
The views I express on this blog are purely my own and should not be construed to represent the official position of the Green Party of Canada - the same goes for all other people's posts & comments.
Erich Jacoby-Hawkins, Barrie ON - although I'm on Cabinet (Nat'l Rev. and Ecol. Fiscal Reform), views here are my own and may not reflect official GPC positions. Please visit www.ErichtheGreen.ca
How's this? (desktop graphic)
How's this?
http://www.greenparty.ca/files/GPC%20poll%20deskto...
I've posted it here on my blog as an attachment as well.
Cheers,
Cameron
- - -
The blog section of the GPC website is a place for GPC members to share their personal opinions and views. The views I express here are my own and are not the official position of the Green Party of Canada.
Use for some purposes, not others
Lambton Kent Middlesex EDA (SW Ontario)
There is a clear value in having individual polls taken and graphed, instead of averaging over a quarter. The benefit for this styule of graphoing is that we are able to see more clearly when in time what shift occurred, perhaps map shifts to news of the day. It tells us what we did right or what others did right to create that shift.
When we average over a long time, like a quarter we can see long term trends, and lose our ability to map them against very recent news items or advertising.
Both objectives are of value, but different value.
If we take a number of polls taken at different times by different methods, and graph them with a connecting line, not averaging them, we will not see a smooth line the way we do with quarterly averaging. But back your viewing distance off a couple feet and you can see both the overall trend and the individual contributors.
Graphing all parties with scores divided by their score on last election has a benefit that we can see how they have all been changing since then. But do note that if you graph the same data based on a poll taken when one of the parties was at a high point, their whole graph will appear negative. If you choose a point where a party was at a low, everything on the graph will be in positive territory.
If GPC tends to get fewer votes than polls indicate, then for GPC the last election would be expected to be a low. We should expect a recovery to polling numbers.
To compensate for that, we could use polling numbers a month before last election, but even that has the problem that we are cherry picking our polls. Use a full quarter before the last election to avoid that problem. This is to compare apples to apples, polls to polls, rather than intermixing polls with elections.