I constantly try to convince candidates to try to reach voters in unconventional, but effective, ways. The looks I get are the same as dental hygienists get when they talk about flossing. Still, we must deliver the same little speech, hoping to find a receptive ear. Some people floss.
One of the tactics I push is gathering signatures. Not on fake petitions to demand Rick Snyder stop being himself, or that the nuclear industry shut itself down, but on garden variety nominating petitions. “Hi, my name is Mark Grebner and I’m running for re-election as County Commissioner. Would you sign my petition?”*
It’s not hard, and it leads almost automatically into a discussion of the fact I’m the incumbent, and that the voter already knows who I am, and tells me they’ve always supported me, and…
When the voter states an opinion, it is often one that did NOT exist before the encounter. Instead, the circumstances – I’m at their door with a simple request, and they half-remember who I am – that encourages them to blurt out something complimetary. As I leave, that inchoate semi-opinion tends to harden into reality. After all, they have now articulated something that sounds like a fact, and as Leon Festinger taught us, if you say something you’re apt to begin believing it.
And so, by spending three minutes on a doorstep, I may convert somebody into a long-term supporter. And you can do it too.
“Yes, but can’t I just file $100?”. Sigh.
Perhaps you’d like some evidence?
PPC was involved in various ways with the successful recall campaign in HD51 against Paul Scott. In the course of that effort, the district was canvassed for signatures, almost from one end to the other, on foot.
Now that the data is available showing exactly who turned out to vote in the November 8, we can see whether signing the recall petition – per se – caused people to vote. The evidence is very clear that it did.
In making this comparison, we have to control for differences between signers and non-signers of the recall petition. As you might guess, people who signed the petition tended to be more Democratic, and to have slightly better previous voting records. The best way to account for those differences would be to build a linear regression model where turning out to vote on November 8 is the dependent variable, and various demographic and political attributes of the voters are the independent variables. Then we see if adding the additional fact of whether each voter signed the recall petition improves the predictive power of our model. (Or, to go a step farther into obscurity, I could use a logistic regression model.)
The problem with the linear model is that almost nobody would know what I had proved at the end. So I’m presenting it in a way that I hope makes somewhat more intuitive sense, and yields almost exactly the same result.
The chart below divides the voters of HD51 into groups according to estimates I had made BEFORE the November election, of how likely they were to vote, based on their previous voting history, age, presence of other voters in the household, and use of absentee ballots. The first column, RANGE, shows my a priori prediction. As you will see, those predictions were reasonable, but not perfect. But the reason for including them is simply to make sure we’re comparing apples to apples when we look at people who either signed or didn’t sign the recall petitions. That is, people within each band are quite similar in their characteristics and likelihood of voting.
The second column, VOTERS, is the number of people who fall within that band of probabilities. SIGNED is the number of voters who actually signed the petition. VOTED% is the percentage of non-signers who cast ballots in the November recall election. SIGNED&VOTED% is the percentage of signers who did the same. Finally, GAIN is simply the difference between VOTED% and SIGNED&VOTED%. In other words, it’s the number of additional votes that were apparently turned out by the two-minute process of presenting them a petition and getting them to sign it.
RANGE VOTERS SIGNED VOTED% SGND&VTD% GAIN%
0- 5 12306 888 6 25 19
6- 15 28487 4305 19 40 21
16- 25 5795 1141 36 55 19
26- 35 3533 659 48 67 19
36- 45 2401 519 58 77 19
46- 55 1917 426 60 80 20
56- 65 1736 387 67 84 17
66- 75 1665 343 72 85 13
76- 85 2109 457 77 87 10
86- 95 3149 759 83 91 8
96-100 3747 845 88 93 5
TOTAL 66845 10729
I find two interesting results in this chart. First, that circulating the recall petition had such a powerful effect, particulary on the very weakest voters. When Paul Scott was re-elected in November 2010, some 20000 people failed to vote, and only 9% of the non-signers turned out in November 2011 to recall him. But among those same flaky people who signed the recall petition, 29% voted in 2011 – fully tripling the turnout.
The second point is that the center of the “effects curve” seems to be around 35%, which is almost always the case. Any intervention to encourage voting seems to have its greatest effect on people who are about 1/3 likely to vote on their own. Generally, we see the curve fall off below 10% and above 60%. In this case, the lower falloff point seems to be around 5% or even lower. But the important point is that it’s almost always a mistake to target HIGH PERFORMANCE voters for turnout – the best bang for the buck is 10-60% or so.
Fastidious statisticians will object that there may be latent, undiscovered ways that people who signed the recall petitions differ from people who didn’t sign – but I don’t believe such factors caused the large difference in turnout rates. People who don’t like Paul Scott were certainly more likely to sign the petition, but those people had voted in previous elections following the same patterns as the non-signers had. If not liking Paul Scott could make somebody twice as likely to vote as liking him, how in the world did he ever get into office? More to the point, controlling for political party doesn’t substantially attenuate the difference – people who signed the petitions turned out better, even if they were Republicans. And people who didn’t sign the petitions – including Democrats – turned out more poorly.
If anybody wants to test their alternative hypotheses, I will be happy to provide the full dataset. All I ask is that they publish their findings in a followup here on MichiganLiberal.
Finally, I want to thank Soapblox, for completely screwing up the formatting of my chart, and changing most of the text into Courier without being asked to.