Why save democracy when you can save dictatorship? Incentive compatible survey design

“Good question. Yes, we have your best interests at heart.”

There is a kind of problem so fundamental to organizing that we sometimes forget to think of it; common as day. You see it when

  • People answering a survey tell you what they want you hear instead of the truth
  • Someone lies at an interview
  • Just about any time that people aren’t incentivized to be transparent

Those are all examples of mis-alignment in the sense that individual incentives don’t point to system goals. It’s called the problem of “incentive alignment” (also known as “incentive compatibility”).

The phenomenon of “buyer’s remorse” gives a clean economic example of the idea. In a normal auction, where people are bidding for a thing, it turns out that the structure of the decision doesn’t actually incentivize an honest evaluation by buyers of what they think a thing is worth. In real world auctions people often overbid, in part because they are influenced by the fear of losing. So typical “first-price” auctions are actually not incentive aligned.

But there’s an auction design out there that actually does incentivize honesty. It’s the “second-price” auction. In a second price auction the winner doesn’t pay the price they bid, but the next highest price. Why does that change anything? To see the trick you have to think a bit. At first thought you might just think that the smart strategy is to name a crazy high price and pay the losing bidder’s fair price. But what if all bidders think that? Then you’re going to overpay. You don’t want that: you don’t want to pay more for a thing than it’s worth to you. Where this reasoning gets you is that all bidders in a second-price setting will decide to name the price that they are actually willing to pay, no more, no less.

This is a good example because it also shows how small elegant tweaks you can restore alignment. In so many real world settings, incentives don’t support honest disclosure. We have workarounds in most parts of our life, but the problem still matters, and it attracts a lot of attention from economists. Their work depends crucially on the idea that incentives determine behavior.

Why save democracy when you can save dictatorship?

Incentive compatibility is especially challenging for survey design. How old are you? How much do you make? Who did you vote for in the last election? It turns out that we can’t always trust the answers to these questions. That’s important because surveys are the least bad way to learn things about people in a standardized way. But what if it was possible to pose any question—How often do you have non-PC thoughts?—in a way that people felt an incentive to answer truthfully?

For some things the problem is easy to solve once you’ve spotted it. Say you’re studying philanthropy, and you ask “Do you donate more or less to charity than your peers?” But you realize that most people will say that they donate more than they do. The incentive compatible way of getting an (honest) answer will be to invite people to non-hypothetically donate some of their survey reward to a charity. If they donate and that donation is smaller or larger than average, you the researcher found out if they donate more than others without actually having had to ask. By replacing hypothetical questions with costly behavior you get honesty.

Another strategy is to ask questions with verifiable answers. Instead of asking “What is your height and weight?” you might say “What is your height and weight? We will measure you after this and you’ll only get paid for participation if the difference is 0.” But if you’re verifying then why ask in the first place? And what if verification is impractical? And, most relevant for us, what if it’s impossible, such as with subjective self-evaluations (“Are you kind to others?”)?

Where it matters most, incentive compatible survey design is actually a real can of worms. The problem there is clearest if we hit pause on saving democracy and take a moment to try and save dictatorship. As a longtime scholar and organizer of self-governing communities, I’m comfortable saying that many communities could do worse than structure their governance under a benevolent dictatorship. A lot of groups, organizations, and communities that I admire do. In its most ideal form, benevolent dictatorship is not that different from democracy, because the dictator, being benevolent, is caring, curious, and motivated to understand and integrate everyone’s needs. As a result, the dictator will generate just the kinds of solutions that a healthy democracy would, and they’ll probably so it much more efficiently than a large governing body.

So why not replace everything with benevolent dictatorship? The main problem is fragility. Nothing systemic keeps the “benevolent” in there. If your competent leader was replaced by another competent leader, it’s generally luck. And you have to keep getting lucky because your first dictator won’t be your last. Benevolent dictatorship slips very easily into the non-benevolent kind that has reliably attended humanity’s darkest moments. Whether it’s through bad succession or the corrupting influence of power, no tool we have can reliably keep a benevolent dictatorship benevolent.

Incentive compatible survey design

Well there might be one tool. What if we had incentive compatible personality tests? It’s easy to imagine the important questions you would want to ask a candidate for dictator.

  • “How likely are you to abuse power?”
  • “How do you respond to disagreement?”
  • “How do you respond to insults?”
  • “If a brakeless trolley is hurtling toward a loved one, and you’re at the switch that can divert it on to another track with n people you’ve never met, what is the largest n you’ll tolerate.”
  • You’re infuriated after reading a personal attack by a journalist in a major newspaper. Will you act on that journalist? If so, what will you do?

Asking is easy; what’s hard is to know if their answer is honest. If there was a way to know what someone really thinks, you’d just disqualify the people who give bad answers and appoint the people who give good answers.

I have lots of bad ideas on how to solve this, such as what I call “double-blind policy“, and is based on the premise that you can’t lie about a question if nobody knows what was asked.

But generally this is more likely to remain the name of a major challenge rather than the name of a class of solutions. Still: if we could solve it, I’m not entirely positive that I’d remain a scholar of classic democratic systems. I mean I would, but it would be harder not to admire the green grass on the other side.


This entry was posted on Wednesday, May 22nd, 2024 and is filed under Uncategorized.