Sunday, March 30, 2014

Satiation, imitation, human error, and variable reinforcement: 4 lessons from psychology for economic theory


 - by New Deal democrat

(You know the drill.  It's Sunday, and I take a break from graphs and data and voice some opinions.  Regular nerdiness will resume tomorrow.)

Noah Smith ignited a discussion this past week with a post about the backlash against behavioral economics. He wrote:
There are two main knocks against behavioral econ. These are:

1. There is no "grand unified theory" of behavioral econ; instead it's a bunch of specific little theories for different situations. What we want is a widely applicable, unified theory of economic behavior.

2. The behaviors produced in a psychology lab are extreme effects produced by extreme, coordinated manipulations; in the real world, lots of stuff is going on, and what we care about is the average.
 Brad DeLong had a repost here. [Update:  Tim Harford's original post is here.  Cullen Roche's succinct, and devastating, rebuttal is here.]

This feeds into one of my long time grudges against the imperialism of economic theory.   Once upon a time I took my B.A. in psychology and was accepted into a hifalutin "public policy" Ph.D. program at a top ranked school.  I very quickly discovered that, despite claims of an "interdisciplinary" approach, in fact it taught "theoretical economics uber alles."  When I confronted a professor with my knowledge that the claimed outcomes contradicted actual, empirical psychological studies, he simply said "it all randomizes out."  I transferred out of the program.

So, let me briefly discuss 4 specific routes towards a "grand unified theory" of behavioral econ.

1.  Satiation.

Econ theory starts out with an assumption that drives are insatiable.  We all know that's not true (well, ok, sex might come close).  I may thoroughly enjoy steak, but not at every meal all the time.  Once we start from the proposition that drives are satiable, demand curves in particular change their shape. Once I have enough of something, I don't want any more at that or any price.

There is a very real world manifestation of this in the demand curve for labor.  Econ teaches that the more you pay someone, the more they will work.  If anything, over the long term, the reverse is true. Most people work in order to save for one or more purposes: whether paying this month's rent or feeding the kids for the poor, to saving for a house for the middle class, to saving for the kids' college education and retirement for the affluent middle class.

Once all of the targets have been achieved, there is no more economic reason to work (although there may be strictly social reasons in terms of keeping active and interacting with others).  If I have decided that I will retire once I have $500,000 set aside, and you double my pay, you have just halved the time I will continue to work.  Far from increasing my labor, you have decreased it.

2.  Variable reinforcement.

One of the oldest and most durable results from psychology is that the way to maximize a behavior is *not* to reward it all the time.  Once the behavior is learned, you decrease the number of times you reward it, and reward it on a random,  non-regular schedule.  Since your subject is working for a reward, you will get lots more of the behavior if it takes many repeated cycles to get the reward than if the reward is given every time.

I once pointed this out to Prof. Greg Mankiw on his blog.  He was not amused.

This point fits in very well with the discussion of satiation above.  If my lab rat can press one of two bars in order to get food, and one bar gives out food every time, and the other on a variable schedule, my rat is going for the bar that always rewards him.  He'll get satiated quicker, and then he'll stop.

But if I am an employer, or a casino, (of even Kos of DailyKos), what I want is to maximize the behavior that rewards *me.*  Casinos are an obvious example, but that's also why so many employers gradually reduce perks over time once an employee is hired, and why they are instituting variable annual bonuses.  Once they have the employee trained, variable reinforcement maximizes the employee-equivalent of bar-pressing, and complex bonuses dependent on many metrics increases the employees' performance on each, since they don't know which one will give rise to the biggest reward.

And if you participate in a big bulletin board like DK, the comment system ensures that you will keep coming back for interaction, and the comment recommendations that add up like pinball scores serve exactly as variable reinforcement.  I have no doubt that if we could hook up most Kossacks to an fMRI, we'd see endorphins released into the brain every time they see the number of recommendations for a comment increase.  It does wonders for Kos's page views.

3.  Imitation

"Monkey see, monkey do."  Much of, and maybe most, human learning takes place by imitation.  You imitate your parents.  You imitate your older siblings. You imitate your 'cool' classmates.  You imitate your professors and mentors.  You imitate a successful profit-making strategy.

In psychology, it is known as "modeling."  You model your behavior on others.  Economists really haven't explored this at all.  I once asked Prof. Mark Thoma about whether economists had ever thought to study learning, and got back a response indicating that some economists had used a system of successive approximation.  But brute force practice is not imitation.  The closest economists have come to discussing imitation is in the concept of herd behavior.

Imitation shows how a strategy, once novel, becomes commonplace and sometimes overshoots with catastrophic results.  A great example of the former is franchising.  Pioneered by McDonald's, and by Harlan Sanders, the founder of Kentucky Fried Chicken, franchising was quickly imitated and now there are probably thousands of franchising corporations, explaining why virtually any mall and any large plaza or strip mall looks virtually identical anywhere in America.

The problem arises when a behavior can only be imitated by a limited number of people without turning toxic.  A great example of this was house-flipping a decade ago.  Originally it was  a great success.  The flipper buys an old property, fixes it up, and resells it for a profit.  During the housing boom, then bubble, the number of flippers probably grew at an exponential rate.  At some point the market reached satiation, and then went right past it.  A successful profit strategy for some became a disaster for a far larger group of people.  Modeling as a psychological concept explains how that happens over time.

4.  Human error

It is a trite truth that all humans have different skill levels.  And probably everybody engages in some self-destructive behavior.  Even in a perfectly competitive market, different participants will better react to changes, or even to extended periods of stability.  If there are no barriers to entry, the less skillful (or lucky) participants leave the market, and newer ones come in.

But over time, most businesses tend to become oligopolies.  And oligopolists can engage in poor or unlucky decision-making as well.  When they do (Arthur Anderson in the Enron scandal, Circuit City), the oligopoly shrinks, tending further towards monopoly.

Too often even the most prominent economists seem to assume that the "free market" participants always make the most "optimal" decisions.  We know this isn't true, and the impact of poor human decision-making over time in conditions other than perfect competition appears to have been totally neglected by economists.

TO SUM UP:  Psychology isn't just about a bunch of discrete tiny behaviors that randomize out.  It is about human variation, how we learn, and how we engage in large systematic behavior patterns.  Satiation, imitation, human error, and variable reinforcement are general concepts that have major implications for any "grand unification" economic theory.