Registering new voters is probably the best way to make the elephant unhappy, given that non-voters who turn into voters tend to vote for the donkey. I’m the chair of the DFL of SD63, where we just completed a voter registration project. We were looking not just to register voters, but we wanted to learn about unregistered persons like where in our district they live, whether they’re eligible, why they don’t register and how to persuade them. We figured the high turnout of a presidential election would reduce the number of unregistered people to it’s lowest point, so we would be looking at the people least likely to vote, and skipping the relatively easy registrations of people who vote regularly, but moved between elections.
We registered some new voters, we learned quite a bit unregistered people in our district, and we learned how to learn about them. What we learned will make future voter registration efforts more effective, and will help with canvassing in general. I wrote a report for our senate district, and I’ve combined our forms, scripts, and instructions into a kit which is available to any other DFL unit that wants it. Though what we learned about the unregistered people in our district applies specifically only to our district, anyone can take what we did and learn about their own district.
So what did we learn about unregistered people, and how did we do it?
We tried to take the approach that we knew nothing about our district or unregistered persons until we actually tested our guesses, even if those guesses were based on long experience with canvassing and long residency on the district. We want as much as possible to trust our data rather than our assumptions. We did in fact guess right about some things, though now we’re not guessing, we know. What was more interesting was what we guessed wrong about, especially if assumptions were completely blown away.
We put together scripts and forms, taking our best guesses at what would work to collect useful data and convince eligible non-voters to register. We produced lists of all registered voters in a turf and used that as a skip list, meaning we knocked on the doors that weren’t on the list. The jargon for what we did is a “reverse doorknock”.
We did test knocks with just a core groups in a couple different parts of the district. Our initial approach to targeting was to pick turfs at random so as to get the most accurate sample possible of unregistered persons. However, we discovered in the tests that some areas have so few unregistered persons, that it’s possible to go an entire block and every address is on the skip list. Our concern was that sending volunteers to turfs with so few people to talk to would make them feel we had sent out for nothing but a long walk, which could mean they wouldn’t come back. So we tried to split the difference between being random and giving volunteers a chance of actually getting registrations.
We had canvassers identify themselves as volunteers with the DFL, and we made buttons with the SD63 logo, so the people they spoke to would know they were speaking to Democrats. We guessed this would cause Republicans to filter themselves out by their refusal to register with us, so the people we registered could be regarded as Democratic-leaning. That means that besides finding Democrats, we’ve already done party ID for the people we register.
We made revisions to forms and scripts following the test knocks, so we were able to send out volunteers with some confidence about what we included in their clipboards.
I’m not going to go into demographic details because I doubt anyone outside our district would care, and I’d rather we didn’t hand things to the Republicans. Let them grind it out as we did, not that I expect them to start registering new voters instead of suppressing them, but still, no point in making it easy. So I’m just going to write broadly about what we learned, with a reminder that this applies particularly just to our one district, and may not be true in yours.
We learned right away that people in MUBs (multi-unit buildings: apartments and condominiums) are less likely to be registered than residents of single-family homes, so we chose to target areas heavier in MUBs. We looked for buildings with fewer registered voters than units. We were able to do this because in 2012, our district worked on a multi-unit building project. This entailed gathering the address and other data about every MUB through information available on the Web and by going street by street, a precinct at a time, through the whole district. Yes, it was as tedious as it sounds, but it’s given us a base of data we’re pretty sure our opponents don’t have. We also picked some areas with clusters of MUBs, and otherwise random areas where at least a couple buildings were located.
There was a tradeoff to targeting. Our sample is not as random as we hoped. The choices we made for targets introduced biases into our results, even if they resulted in more registrations. Where I indicate uncertainty about our findings, the reason will usually be this bias. It’s possible unregistered persons in MUBs are different than those in single-family homes, or those in MUBs where relatively few were already registered may be different than those where most residents were already registered. It’s possible we missed differences between different parts of the district. None of this means our findings are wrong, merely that they’re more uncertain than we’d like. We have results we can use for future canvassing, but we’re going to have to be watchful for data we missed.
We had guesses that turned out to be very wrong, but fortunately not wrong in the sense of losing the election because we got something wrong. Not like “Romney insisting the polls were skewed” wrong. These were more enjoyable discoveries of error, and the most important such error was possibly our guess that we would be frequently told to leave the MUBs we managed to get into, maybe a third or half the time. It happened only once in two full-scale doorknocks and all the test knocks. We have too little information to know if one sort of building is more likely to contain residents who tell canvassers to leave. Can’t tell from a sample of one. What we know for sure is that if we can get into a building, we’ll almost surely be able to doorknock it. This doesn’t apply just to voter registration, but to any canvassing we do. We need to encourage campaigns to reconsider their emphasis on more easily canvassed single-family homes.
As mentioned in the section on why we targeted instead of picking turfs at random, we guessed unregistered persons would be clustered, but that they would be scattered enough that we could find some anywhere. It turns out some parts of the district are so close to fully registered, that it’s possible to find blocks where every address is on the skip list. There were many where there are so few doors to knock, that given how most people don’t answer their doors for whatever reason, a canvasser could potentially walk a large turf and not talk to anyone.
We guessed some eligible persons would express general disgust with politics, saying things like all politicians are crooks or elections are rigged, but this was quite rare. We tried to anticipate these sorts of objections in the detailed instructions for canvassers. They might have felt that way, but they didn’t care to say so.
Residents of MUBs seemed more likely to answer their doors than residents of single-family homes, but this is an impression, not quantified. Assuming it’s accurate, combined with the short distance between doors, this makes MUBs much more efficient to knock.
Sometimes we guessed right, but now we’re not guessing: we know. To wit:
We confirmed our guess that residents of MUBs are more likely to be unregistered. We can’t be precise, that X% of MUB residents are unregistered compared to X% of single-family residents, but we don’t need to be exact for this information to be highly useful.
Residents of new buildings are the most likely to be unregistered and were the most likely to register.
Most immigrants not already registered are not eligible. This seems like a good place to repeat the caveat that this is just our one district.
Our district’s immigrants are overwhelmingly Hispanic or Somali. We encountered few of any other group. We were able to provide canvassers with phrases in Spanish and Somali, and those proved essential in some turfs. Even though the people who spoke no English turned out to be non-citizens, we were at least able to communicate why we were at their door. Hopefully, if they become citizens, they’ll remember that the DFL contacted them.
The majority of unregistered voters fell into one of two groups, immigrants (mostly non-citizens) and people who moved after the last election. However, this is where the biases might have been added by our choice of targets. We may have overemphasized areas with more immigrants, areas with more short-term residents, or areas with newer buildings where many residents are new, even if they will be there long term. I’m 70% sure we have it right, which in practical terms means I’m unwilling to say we know definitively those two groups are the vast bulk of our unregistered persons, but I’m sure enough to use that information in planning doorknocks, though with an eye out for evidence that’s wrong.
We’re still entering the last data on demographics so I’m giving an impression from having looked though walksheets — take this with a big asterisk. It looks like our unregistered persons are about half and half white and non-white, in a district about 80% white. However, so many of our unregistered persons are non-white immigrants who are non-citizens, I suspect the eligible unregistered voters will look like the district. Their ages seem to cluster in the 20′s, 30′s and 40′s. They seem equally split between genders.
By far, the most frequent objection to registering was the immediate lack of time. “I don’t have time” could mean multiple things. Sometimes that statement came with a request to leave a form, which canvassers took to be a sincere interest in registering. We’ll eventually be able to check to see if the addresses where forms were left ended up in the voter database, and then we’ll know what percentage really do fill it in. Usually though, canvassers came away with the impression of being put off. “I don’t have time” could be a euphemism for “I don’t have any interest”, “I won’t register with a Democrat”, “I don’t really understand this,” or “I hate having people come to my door and I’m trying to not be rude.” There’s no way to know. The persuasion was normally that registering now would save time on election day, and sometimes that worked, plus we haven’t found a better answer yet. We are sure though that the majority of people who don’t want to register will deny having time rather than give a reason.
Incidental to comparing the number of units in MUBs to the number of registered voters at that address, we confirmed that MUB registered voters are more likely to be “no-data”, which means the database has no data to determine a likely party preference. It appeared the proportion was more than half, whereas the whole district is about a quarter no-data. This doesn’t affect voter registration, but will be very useful as we focus on no-data voters in future doorknocks. My speculation is the cause of the difficulty is getting into MUBs, which I presume is behind the practice of focusing on single-family homes, which means we don’t get better at getting into MUBs.
There are some things we still don’t know, that might be worthwhile for future research:
Do most eligible voters know about this year’s local elections? We were speaking to people who were unregistered, so if they were regular voters, then they probably had moved, and our impression was most didn’t know. But that’s just an impression, and the fact is we often mentioned it without trying to figure out if they already knew, since a secondary goal was letting them know.
We operated under the assumption that if one person at an address is registered, then all eligible voters are registered. That’s playing the odds when it comes to picking targets, but it isn’t true 100%. Is there a way to identify addresses where some eligible voters are registered but not all?
Will our impression that MUB residents are more likely to open their doors stand up if we start quantifying that? If accurate, why do MUB residents answer their doors more? It could be they aren’t as accustomed to being canvassed and aren’t as unwilling to open their doors when they don’t expect anyone, in which case they might become unwilling if they get canvassed more often. It could be as simple as they don’t have a backyard or garage or basement where they can’t hear the doorbell. We can only speculate now, but we have to be aware in case something changes and canvassing becomes less effective, just as changes like caller ID and the frequency of GOTV calls have made phone banks less effective.
Would the results be different if this weren’t the year after a presidential election? Maximized turnout last year reduced the pool of unregistered eligible voters, but such hard-core non-voters might not be representative of unregistered eligible voters after a low turnout election.
How much time has to pass before it’s worth covering the same ground? That is, will there be enough turnover after X number of years to make our data out of date and require doing the project over if we want to be sure it’s right?
We lumped in MUBs together, but are there any differences between apartments and condominiums?
Which parts of our district have more transient populations, and are these populations less likely to register and turn out?
Could we use data on income or poverty rates to match up our findings from doorknocking and use them to be more precise in seeking eligible non-voters?
Again, I invite DFLers who want to try this in their districts to contact me about getting ahold of the kit. I put everything in the kit so my district doesn’t have to start over or even need anyone who worked on it before, so it ought to be usable anywhere. You’ll presumably need to customize it to your district, but unlike us, you won’t have to start from scratch.