GOTV - Part 4 - Marginal Costs

by: Grebner

Sun Nov 21, 2010 at 21:49:32 PM EST

You can skip today's lecture if either 1) you already understand the difference between marginal and average costs, or 2) don't feel any need to understand the arithmetic behind my arguments.  For anybody in the middle, I promise this post will be the low point of the series.

[Note - My GOTV essays, plus many more, are collected under Technical Politics] 

When an idiot (let's say, for example, a Detroit News reporter) tries to explain the financial decisions behind political techniques, they often divide the amount of money spent by the number of votes received, yielding a "cost per vote".  Such an approach is absolutely worthless.

In order to choose the best way to spend campaign resources, you need to compare the ratio of the cost of each choice to the number of additional votes it might bring, which we call the "cost-benefit ratio".  This isn't specific to campaigns or GOTV; the same comparison is implicit in every decision:  what kind of car to buy, where to build a new grocery store, how to change the tax law.  For many decisions, the relevant costs and benefits are so hard to specify that it seems a waste of time to consider them.  But where there are many courses open, with widely varying costs and benefits, the methodology at least frames the question correctly.

In real life, it's as hard to formally specify costs and benefits for GOTV as for any other activity.  Our costs refer mainly to two different resources, money and volunteers, which are not easily converted from one to the other.  The benefit generally isn't exactly measured in votes, because a vote in one place may be worth much more or much less than a vote cast a few miles away, depending on which targeted districts overlap each voter's precinct.  Our planning processes are beset by uncertainty of the effectiveness of our tactics, of how much money we'll finally have available and exactly when it will show up, of political and legal strings attached to the resources provided by various organizations.  In my forty years, I've never seen an actual written cost-benefit analysis in a campaign - not once.

And yet, with all its difficulty, our attempt to optimize the cost-benefit ratio of our activity is what every GOTV campaign is about.  Our estimates don't have to be precise; if they're accurate within a factor of two, we'd be doing pretty well.  The alternative is GOTV where money is spent on activities which are known to be entirely futile, where the decisions are made by people who have no talent or interest in actually winning elections, where what passes for a powerful idea is one put forth by whoever has been around the most years or who claims most vehemently that he's been successful.  As Charles Babbage once remarked, it's much better to act on inadequate data than with no data at all.

First, if we have both money and volunteers, we have to accept a crude method of converting from one to the other.  Let's say a volunteer is worth $10 per hour of useful labor.  (If that seems too high, or too low, it shows you're beginning to think along the necessary lines.)  Let's say we are running a GOTV in a small county - say Isabella - and we have $10,000 at our disposal, in a mix of cash and labor.

Next, we need to think about the range of techniques that are available to us.  (For the present discussion, I'll stick to old fashioned methods.  In later posts, I'll argue they're all futile and we need to re-think what we do.  But for now, I'll apply cost-benefit analysis to the familiar.)  We might think of door-to-door canvassing, leafleting, robo-calls, radio ads, and email for example.

If we do the absolute minimum, a certain number of people will vote anyway, driven by media coverage, candidate-centered advertising, their personal interest in politics, and so on.  We need to make rough allowance for how many "natural" voters there are, so we can consider how many additional voters our efforts would bring out.

Two political scientists at Yale, Don Green and Alan Gerber, have rigorously tested all the standard GOTV methods, and describe their findings in Get Out The Vote.  It's pretty depressing reading.  Most of what we do has very little effect on voters, and some has no effect at all.  The best technique (door-to-door canvassing) costs $30 per additional voter, and the cost rises rapidly from there.  Leafletting, email, robo-calls, media advertising, and (most) direct mail have no demonstrable effect whatever.

But that's not the end of the bad news.  We still need to discount our newly motivated voters for the fraction that will cast Republican ballots.  For techniques based on voter lists, how certain are we of the Democratic proclivity of the people we will contact?  If we're leafleting whole neighborhoods, for every 100 voters we mobilize, what is the net benefit to our Democratic candidates, after netting out any Republicans we accidentally push to the polls?  If 75% of our additional voters support the Democrats in the close races we're trying to affect, the result is to DOUBLE the cost per vote.

I suppose I've buried the lede, as Eric would say.  The terrible truth that arises from careful analysis is that the marginal cost of adding one vote to the Democratic ticket's plurality is at least $60, and our hypothesized $10,000 might possibly gain us 160 votes - if everything went right.  

In real life, things DON'T go right, and we routinely spend $100, $200, or $500 per additional net vote.  That's hard to believe, but it's true.

In November 2008, considering only battleground states, where the Obama campaign spent over $1 billion, the marginal cost per net vote was probably north of $1000 per vote.

How can it possibly be so hard to motivate additional voters?  The answer is that the first 90% of the people who vote are motivated by their own interests or by somebody else before we get to them.  They're part of the "average"; they're not "marginal".  The true marginal voters - the last 10% - take more effort.  They don't have much personal interest.  Or they've learned to lie about their voting behavior.  Or they have other self-defeating behaviors.  Or they have personal circumstances that distract them.  (Maybe they've just moved and haven't finished unpacking.  Maybe the wife is a few days from giving birth.  Maybe the teenager has just been arrested.  And so on.)

It always surprises people to hear an analysis that shows it costs so much to turn out additional net votes.  In response, the old-timers tell us about the thousands of votes that are routinely turned out by the very feeblest of contacts - soundtrucks, PSAs, walking-around money.  The guys at Yale have listened to all the stories, and they've challenged the old-timers to show their stuff.  You may believe the old-timers, but I personally believe the chi-square statistics.