GOTV - Part 13 - Built-in Research Component

by: Grebner

Tue Mar 22, 2011 at 05:03:45 AM EDT

If you've waded this far through the turgid waters of my prose, you should have noticed repeated references to control groups, careful record keeping, objectivity and other hallmarks of academic research.  This isn't just a matter of style, but a fundamental break from the legacy of running GOTV operations as an informal folk-way.

If we decide our goal in managing GOTV is to turn out the greatest possible number of additional Democratic voters - in other words, if we are trying to optimize the operation - we have to allow ourselves to be governed by scientific methods.  You might think the actual science might be done someplace else, and merely imported to Michigan - the way crop methods are perfected at agricultural colleges and transmitted to individual farmers by extension agents - but in the field of GOTV, there's so little genuine science that we'll have to do most of it ourselves rather than relying on distant experts.

In practice, building science into our methods means four things:

1.  We need to think in terms of clearly defined "interventions" to which some voters are exposed while others are not.  Thus, we need to think of a round of bulk rate mail as a "treatment" which we apply to a specified group of voters.
2. We need to hold back randomly selected, and statistically similar, groups as "contols" who do not receive the intervention, so we can later see if what we did actually made any difference.
3. The comparison must be done using actual statistical methods (at this point, insert all the forgotten terms from that Methods class you hated so much:  chi-square, Student's t, correlation, variance, etc.).  That is, the comparison can't stop with the question whether there was ANY difference, but must go on to ask whether it was "statistically significant" or whatever.
4. There needs to be some degree of external supervision of the process, to guard against short-cuts and distorted analysis brought about by insiders who have personal reasons to want to show their project succeeded.  In general, this means turning over a copy of the data to a trusted outsider BEFORE election day, guaranteeing that the control group was genuinely random and comparable.

At first blush, it sounds like adding science to our methods means adding heavy, useless overhead.  Even worse, it means undercutting our efforts, since some high-value targets will be deliberately ignored on election day, and left to fend for themselves.  But experience has shown the costs are small and the benefits are large.

In the first place, the control group may be a very small portion of the total target group.  If our intervention is (say) 500,000 people statewide, a control group equal to 1% of that number would be entirely adequate.

Second, the effect of excluding some people from the intervention is only to free up resources to use elsewhere.  If our initial plan is to walk 50 precincts with Dem performance >70% (to take an old-fashioned example) and we decided to exclude 10 of those precincts at random to serve as a control group, it may mean we can expand our list of targets to 60 precincts, by lowering our cutoff to a Dem performance > 66%.  In theory, we're wasting part of our effort by applying it to some voters with a slightly lower payoff, but if you work through the arithmetic, you discover the prospective loss is only a handful of votes.

Third, and most important, is the fact that much of what we do is actually a colossal waste of effort anyway.  If we would just STOP DOING THINGS THAT ARE USELESS, we'd free up time and money that could be applied where it would actually be productive.  But the first step is to figure out which activities are useless - and to prove the uselessness to all the people who believe in it.  As long as the only "evidence" that matters is the body of lore told by old-timers, we're going to continue trying to cure disease with leeches and poultices.  Once we insist that leeches be applied to a random sub-group, and we keep accurate records of who dies, we'll be in a position to move forward to that Pasteur fellow's claim that disease is caused by his invisible animalcules.  Not giving leeches to everybody isn't as big a loss as the old-timers think.  (As a practical matter, as long as the old-timers continue showing up at meetings, we're going to continue distributing a few leeches.  But they don't have to consume the major part of our effort.)

In my experience, our GOTV efforts should create and maintain a good relationship with one or more university Political Science departments.  For five years, I've been working with researchers at Yale, but there are any number of homegrown academics who would are interested in such work and would be competent both to provide advice and assure that analysis is performed objectively.