More detailed analysis of the relationship between spending and the Green Party of Canada vote.

In March, I published a post on the correlation between campaign spending and vote outcomes for the Green Party of Canada. I gathered together the Green Party of Canada Campaign spending data and votes results for 102 Ontario Ridings. Recently, Michael Moreau, a Green Party activist from Winnipeg commented on my basic correlation numbers, and pointed out that correlation is a poor analytic tool by itself. Since Mike is a mathematician, and knows what he’s doing, I sent him the raw data I compiled, and he produced some very interesting analysis. Mike’s blog is the Don Street Blog. What follows here is paraphrased from his reply to me, with some judicous cropping and editing.

 Mike ran a non-linear and linear regression analysis of the 102 ridings for which we have data. The non-linear graph tells us that there may be differing sensitivity of vote% gain to extra dollars – that in fact an extra $1 is worth more for the smallest of campaigns. However, there is little correlation improvement between the non-linear and linear, so we can use the linear model.

All reported Ontario campaigns: Linear

All reported Ontario campaigns: Linear

 

The linear model of the 102 reported ridings is the most important graph. It tells us that there is a massive correlation between the two variables. In fact, given a degree of freedom of 100, we can be more than 99.999% certain of the correlation existing! Now, the regression line tells us a lot, too. It predicts that for every $1000 in increased spending, we will get 0.1836% more vote total. In other words, it predicts that we will gain 18.36% over a base total by spending $100,000 in the riding. Unfortunately, at some max $100,000 in spending, the model predicts only 24% of the vote share for the Greens in an average riding. That means that at base support levels in 2008 for the GPC, no Green could be elected by only pumping in money. But, the money gets us closer.

Under $5,000 spending

Under $5,000 spending

Now, the other graphs zoom in on certain money ranges and tell us that the relationship between spending and vote% is fairly consistent at any level. The lowest range ($0-$5000 in spending) is a bit of a dog’s breakfast, though, since there are so many other factors at play in those locations. There is some evidence that extra dollars at that level are more effective – but not too much evidence. We cannot be statistically certain that an extra dollar spent is more effective in a riding with little money versus in a riding with more money – just that a dollar is effective. In other words, we can’t say that the GPC should funnel money to smaller EDAs to help kick-start their campaigns – but, we suspect that this is money is more efficient in those ridings than in the ridings with $40,000 already in play.

 

Finally, there is a 95% confidence interval to deal with. No one much cares about this at this stage, but for $10,000 spending in a randomly selected GPC

Ontario Over $5,000 spent

Ontario Over $5,000 spent

race in Ontario in 2008, the model predicts that vote total would be 7.723% plus or minus about 4%. That is to say that we can be 95% confident that with $10,000 spent, you would receive between 3.7% and 11.7% of the vote share. Such a wide range reflects the fact that there are many other factors at play. So, using the regression analysis to predict a vote share result entirely based on spending is faulty. However, we can say for certain that increased dollars equals increased vote % in a particular riding – and $10,000 gives us 1.836% more vote share.

 

STRATEGICALLY:

 

If the goal of GPC is to increase vote share overall, money can be sent anywhere, but we suspect (and have a little evidence) that “seed money” in small ridings can be the best use of resources. This money should only go to ridings where there is someone organized enough to spend it effectively and efficiently, though.

 

If the goal of the GPC is to gain “beach-heads”, then GPC should fully fund EDA’s where the Greens have a solid base of support, many volunteers, and a credible candidate. However, there should be some caution here. Only 4 green campaigns spent more than $42,000 in 2008, so we cannot be certain that the linear relationship between vote share and spending continues at higher spending levels. There could be any number of results.

Finally, regardless of the strategy, money should only be sent to ridings which meet certain criteria for federal funds. Those criteria should include – but not be limited to – number of members, vote gap between green vote and riding winner, organized EDA, evidence of past effective use of funds, and intangibles such as the candidate nominated.

This ends the first in (hopefully) a series of data crunchings from GPC Ontario 2008 and Canada 2008.  

The above is largely Michael’s analysis. The conclusions are his, and are certainly subject to discussion. The Data is what it is, and at least subsequent discussion will be based on honest to goodness data, instead of conjecture, and plausible but untested intuition and opinion. I for one will be revisting my past conclusions about beachhead vs. rising tide national strategy. I think that more than ever our strategy needs to be more sophisticated, looking at both rising tide, and targetted efforts. The better we can understand both the limitations, and opportunities that face us, the better decisions we may make in the runup to the coming election.

UPDATE JUNE 23: Alice Funke at “The Pundit’s Guide” published a post on the relationship between spending and voter outcomes last month. For those of my readers who are involved in planning the next GPC campaign, they should read this post, and draw the appropriate conclusions regarding the likelihood of NOT earning a financial return on a rising tide strategy. (There are some good reasons to broaden the target, but nort broaden it to 308 target ridings.)

Add to: Facebook | Digg | Del.icio.us | Stumbleupon | Reddit | Blinklist | Twitter | Technorati | Furl | Newsvine

Advertisements

16 Responses

  1. Wow! I feel bad now that I haven’t posted on my blog in 2.5 months. Thanks for the hat tip.

    Let me correct one thing: I’m a math teacher, not a mathematician. Otherwise, good to see the graphs up.

    Strategically: Money good. EDA to spend money better. Local riding-by-riding intangibles are way more powerful an analytic tool then my regression analysis. Still, good to find correlations.

  2. Maybe we can throw membership numbers into the mix? Not to publish the raw numbers, but the conclusions will definitely have some lessons for all Green Party activists, locally or centrally.
    Now for my usual needless criticism, if you earn your living teaching mathematics, does that not make you a mathematician?
    Strategically, this discussion has modified my thinking to a certain degree. For example, if it really does take $10,000 to increase the vote share by 1.8%, then in the typical riding this will bring in less than 900 votes. In terms of return on investment, it isn’t wise to overspend in the expectation of recouping the money through pay per vote subsidy. If the threshold for 10% of the vote is doable, the financial math becomes favourable. This implies that a nuanced strategy may look at broad based support for ridings at the cusp of the 10% mark, while the balance of election funds might go towards meeting other objectives.

  3. As an interested outsider looking in (and somewhat cynical onlooker) I find it sad that there is ANY correlation between money spent and vote obtained! It SHOULD all be a level playing field but we know that is not true, it takes money to let the public know about your ideas (or to spin and lie your way to power as some have done!). That said is it not as much important HOW the money is spent as how much is spent? Could this not be one of the many variables, perhaps a major one?

  4. Hi Rural,
    Money has played a central role in electoral politics since the days of the Roman Republic! This analysis is kind of background information that conclusively establishes that money is in fact important. I would love to test other variables, like how the money was spent, but whatever variables we are measuring will need to be consistent across all reporting ridings. The impact of HOW the money is spent is the $64 question, but the answer is not simple. If it were, then that would have been the subject of this post. I for one would like to see more analysis with different variables added into the mix. The more robust your data, the more meaningful the conclusions. Hopefully there will be future posts with additional variables tested.

    This broad kind of question has it’s uses. There have been, and will continue to be discussions in the Green Party about the relative merits of funnelling our campaign spending in different directions. This now gives us some concrete data to support, or reject conclusions about how the money should be directed.

  5. @Rural:

    While I agree that money for votes isn’t ideal, this really isn’t a case of money for votes but rather information for votes. Without providing information to people, multiple times in multiple ways, they won’t know you, understand your position, or choose to vote for you. Flyers, signs, pamphlets, billboards, websites, phone banks, etc…all cost money. So, money is one good measure of how much campaigners put into reaching out to voters. It’s not the only measure of how they reached voters, though.

    Without money, I don’t know how a campaign reaches 50,000 people in 30 days with multiple, convincing messages that both convince them that you are the candidate to support and that you can win.

  6. Entirely agree that money is at least one measure of getting the message out even if I dont like that it is so!!
    But rather than “Without money, I don’t know how a campaign reaches 50,000 people in 30 days ” should that not be “reaching as many people as possible 365 day a year” election or not?
    But, yes information is power and these analysis are important, you may be sure the “big boys” have full time staff working on such things.
    Gess I am an idealist as well as cynical, is that possible??

  7. To Rural:

    Parties wish to reach voters 365, and to some extent try to, but that way is even MORE expensive than during elections, for many reasons.

    One is that election expenses are refunded 60% by the government if you get 10% vote or more (most Lib, Con, NDP, Bloc do, and a growing number of Greens) so it is the cheapest time at less than half the cost.
    Second, the public are far less likely to pay attention to partisan messages outside of elections – even during, it takes a lot of head-shaking to get people to pay attention. (You literally have to bang on their door and wave it in their faces in many cases).
    Third, you may convince a voter in January and move on, but have lost them by the October election. So you have to do it all again during the election despite any pre-election contact.
    Finally, you only get to spend each dollar once. Any that you spend between elections can take away from what you have during elections (especially for smaller/poorer parties) and if your outreach fades during the election, all your pre-election work is lost.

    So for those four reasons, party communications are generally concentrated during election periods. The main exception is Conservative attack ads, because they have extra money to burn.

    On the comparision to election spending dating back to the Roman Senate, a critical difference of the last few decades is that modern election spending is primarily communication – including polling and message-crafting – while the primary expense from Roman times up until the 19th century (later in some corrupt areas) the spending was directly on bribes to key voter blocks. At least today’s spending is rather “clean” by historical comparison, despite the “dirty” attack ads. Current laws are pretty good at forbidding bribery and outright lying.

  8. Thanks for posting this analysis. I can spot myself on the graph – the little dot just to the left of $60K (58), just above 10% (11.1).

    We came in well below the line, yet still had one of the better results, well within top 10% (24th out of 303 in Canada, 9th out of 106 in ON).

    Of course, the graph makes no allowance for any number of factors, as you admit. In our case, our spending was disproportionate to our campaign size by at least 50% due to our being able to secure large loans and spend more. Which is to say, our campaign was more like a $40K one that had an extra $20 to spend. Not all of it was spent at top efficiency, either. If we’d spent only $40K then we’d be much closer to the line. I could probably pick out $20K which (in retrospect) didn’t do much for us in garnering more votes. But of course this is the kind of learning that you really only get by making a few mistakes (along with several successes). Certainly it’s clear that one can’t simply spend one’s way to winning.

  9. Hey Erich, you’re welcome. If it helps you to make better decisions locally, and influence better decisions in Ottawa, then it’s done it’s job. I would like to collect more variables, and plug them into the mix. I wish it were possible to get consistent data on actual spending patterns at the riding level.

  10. @Erich:

    Your dot and one other dot above 25% of the vote are the largest outliers. Interestingly, the biggest outliers are at higher spending. This leads me to suspect that our conclusions about money are not well-tested above $40K. The four biggest spending campaigns are not enough to show us whether the trend continues, spikes up, or reaches a plateau.

    Likely, there are just as many myriad factors at play in big-spending ridings as in smaller-spending ridings. But, the big measures may be (1) efficiency of resource use, and (2) ability to efficiently use resources. Without lots of volunteers to use the resources and give them legs, the money is only so useful.

    Another great measure that I’d like to test, but can’t, is size of GOTV effort relative to vote share. Did you have a strong GOTV push? Did your campaign ID lots of voters? Does lots of money translate into many flyer drops and billboards – but no more Voter ID? Could be. To be effective with money, the best things – IMO – are the kind of canvassing and stickers that you did recently. That is extreme efficiency. At $50/poll, you could conduct that exercise in the entire riding for less than $10,000! Boots on the ground is another key factor…

  11. Actually, I have tested GOTV in a more direct, and accurate way. During the London North Centre by-election, there were over 10,000 ID’d voters, most of the polls were scrutinised, and there were several different GOTV tools used.
    The data was prepared ahead of time, with carefully chosen control group on whom no GOTV efforts were made, a second large group who received recorded telephone GOTV message, another large group that received phone calls. Then there was the group that overlapped, and received both message, and telephone call. With all the polls scrutinised, we should have had a direct count of which electors voted, and who didn’t out of all our ID’d voters. This would have quantified precisely how effective GOTV efforts are. Unfortunately, the campaign manager(s) didn’t preserve the data, and all the scrutineers records of who voted were sent to the shredder, without being scanned into the database. (All the scrutineers sheets were bar-coded to make this easy). Actually, that was the reason I stopped volunteering, and let my membership lapse. I had a strange idea that it was deliberate.
    Anyhow, it’s a lot of work, but the next time the GPC is running a campaign with 100 election day volunteers, they should take the chance to get some important data out of it.

  12. Professor Stanbury, formerly of the Royal Commission on Party Finance and Electoral Reform, and I published a two-part series in the Hill Times on the relationship between candidate spending and vote share, which has been reprinted at the Pundits’ Guide.

    Essentially the correlations are pretty strong, except for candidates who get close to winning. In those cases, nearly everyone spends at least 75% of the limit (unless a long-standing incumbent with little competition); and of course the more competitive candidates in the race, the more the spending rises and conversely the lower the vote share required to win the seat in a first-past-the-post system.

    I’ve taken the liberty of linking those two articles here:

    Part I – An overview of candidate spending by party, 1997-2006

    Part II – The relationship between candidate spending and electoral outcomes

  13. That PG! That’s a great site, btw.

  14. Thanks, not that…oops

  15. I knew what you meant, Mike. Thank you.

  16. […] be leveraged by collecting as much low hanging fruit as possible across the whole nation. It was easily proven that spending $1000 in a small riding would win more votes than adding an extra $1000 to the […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: