If you want to curb corruption and mismanagement, audit people unexpectedly, right? "Be unpredictable!" a certain ex-President once said.

Maybe not.

A thread on @wendynassrwong's job market paper.

Wendy looked at largest social welfare program in human history: India's employment guarantee scheme. It gives people public works-style labor when needed.

As @karthik_econ likes to say when people ask him about external validity in India: "WHO CARES?" It's so huge.

Don't worry, though, there's a generalizable lesson coming.

Wendy gets program data from Jharkhand, where the timing of audits is randomized. When a local coordinator expects an audit, how do they change their behavior?

Intuitively, most of us think that a small chance of an audit will be enough to change most people's behavior.

Implicitly, you're assuming there are decreasing returns to audit risk -- that people will get relatively less responsive as the certainty increases.

Also, you probably think that having times with 0 audit risk is just asking for bureaucrats to steal & mismanage.

Well, that's exactly what Jharkand did: randomize people's that they have zero risk sometimes, and known, quantifiable risks of audits (low and high) in other years

Wendy suggests our intuition is wrong -- there are times when people have increasing returns to audit risk. This Indian program is one of them.

Bureaucrats don't ramp up signs of misappropriation when they have 0 risk, and don't respond to low audit risk. Only high risk.

Think of it like cheating on taxes. Maybe a small risk of audit isn't enough to deter cheating. Some people will respond to high audit risk only -- a case of these increasing returns to audit risk.

Wendy develops a formal model to illustrate the logic, and estimates parameters from her data.

What's interesting to me is what this implies for organizations that want to curb cheating or corruption or mismanagement -- they have to figure out what kind of subjects they have, and their responsiveness to risk.

For academics, if Wendy's findings prove correct in other places, I think the next task is to figure out a more general model for what conditions generate increasing or decreasing returns to audit risk. There are implications for taxes, crime, employee management, etc.

Email me, @k_sonin, @LuisRMartinezA or Canice Prendergast for effusive letters!

Finally, the first photo in this thread comes from one of my favorite photography exhibits, "Bureaucratics", by Jan Banning. Check it out: https://www.janbanning.com/gallery/bureaucratics/

Here is Wendy's web page: https://wendynassrwong.github.io