How Do We Know If Guaranteed Minimum Income Works?

A commenter on my last post about a Guaranteed Minimum Income raised some interesting questions. Do we have any evidence that a GMI would actually work?  And if not, how would we get such evidence?

This is certainly a worthwhile question.  After all, if we’re going to totally change the distribution of production for the entire country, it would be nice to know that we’re not going to ruin everything. Remember, we’re dealing with economic incentives, and when you mess with incentives, there are often unforeseen consequences.

I think that we don’t know whether GMI would work, and we’ve unlikely to have conclusive evidence of whether it will work or not without actually trying it. The economy is too complicated to predict and Guaranteed Minimum Incomes would be a huge change.

But that’s a boring, unpalatable answer. So, instead of stopping there, how about we start to look for ways that we might get closer to an answer.

What is success?

The most important question to start with is to decide what “actually work” means.  For this exercise, what is success? Reasonable people could have one of any number of answers, including:

• Maximize GDP per capita
• Maximize GDP growth
• Maximize median income
• Reduce income inequality
• Maximize average happiness
• Reduce happiness inequality
• Reduce the percentage of people below the poverty line
• Reduce catastrophic economic outcomes
• Improve social cohesion
• Create a stable and self-sustaining program

Theoretically, as a scientist, you’d typically figure out which of these areas you care the most about, and examine those through experimentation. As a politician in 2015, on the other hand, you’d decide whether your ideology supports GMI and select whatever dimensions support that ideological view from whatever studies are already out there.

I’m not sure what the right criteria for success are, but I think I think pretty important ones are sustainability and average happiness. In fact, I’d say sustainability is a requirement–without that, none of it is real. And if we’re not doing this to make people happy, then why bother?

On the other hand, if we regressed to become a bunch Neanderthals sitting around banging rocks in a cave, but were really happy doing it, I wouldn’t consider that a success either.  So it feels like success needs to have some element of maintaining economic progress as well.

Becoming a mad scientist

So if we vaguely know what success looks like, then how to we create an experiment to test GMI?

The problem with the social sciences, including economics, is that often, you have to be completely deranged to run real experiments. For instance, suppose you wanted to find out if being a victim of domestic violence as a child leads to perpetrating domestic violence as an adult. Despite the clear value in knowing the answer to this question, many would consider it wrong to divide up a classroom of children into two groups, and beat all the children in group A every day (while giving group B only the number of beatings that a child living in our society would normally expect).

Thus, social scientists are limited in the types of experiments they can run. They typically try to avoid things that hurt people. But even if they can create an ethical experiment, the fact that they are running an experiment can influence the outcomes.

For instance, one natural way to test this theory would be to take a large subset of a population, divide it up into two groups, and give one subset a GMI, and nothing to the other for a long period of time. But would the non-GMI group even care about the study if they weren’t receiving money? Would the GMI group feel like they had to work less to “take advantage” of their good fortune to be in the GMI group? If you add progressive taxes for the GMI group to pay for the program, would it make the wealthy in the GMI group resentful of the wealthy in the non-GMI group, and want to drop out of the study?

Freeloading Canadians

Then there’s the fact that the GMI group benefits from the value provided by the non-GMI group. The pharmaceutical industry is a good example of this today. Drug companies spend billions discovering new drugs because they know that when they find one, they can recoup their costs by charging Americans $10 for a pill that costs $0.03 to manufacture.

But the price for drugs in Canada is far less than those in the USA. Because Americans pay high drug prices, the pharmaceutical companies will create medications that they never would have created if drug prices were as low as in Canada. Yet Canada still gets the benefit of those drug discoveries. In effect, Canada is a freeloader on US pharmaceutical research.

A similar problem would exist with small-scale GMI studies. Suppose some non-GMI participant, incented by lower taxes and the chance to become rich, creates a next whizz-bang gadget. The people in the GMI group will still benefit from the gadget, but you won’t know whether that gadget would have even been created if everyone were in the GMI group.

So what’s the poor social scientist to do?

Social scientists have solved these sorts of issues in several ways.

Data Mining: One way is to become data miners.  Basically, instead of doing experiments, do statistical analysis on data that has already been gathered by surveys and other means.  If you have the number of welfare recipients in a bunch of counties and you know the average GDP in those counties, perhaps you can do correlation analysis and find out something interesting.

There are huge problems here, of course. We’re looking for causation, but correlation doesn’t imply anything about causation. If A is correlated with B, perhaps A causes B. Or maybe B causes A.  Or maybe C causes both A and B.  Or maybe it’s D….

Even worse, the chance that the data contains the exact information that you’re looking for is low.

Natural Experiments: A natural experiment is one where, independent of any experiment, individuals happen to have been assigned randomly into one of two groups, only one of which ends up being exposed to experimental conditions. Thus, in such cases, you end up with an experimental group and a control group with randomly assigned members.

An example of such a natural experiment was the Georgia land lotteries in the early 19th century. Lots of land, formerly owned by the Cherokee Indians, were given away to thousands of Georgians in a random lottery. By comparing the outcomes of those who won the lottery to those who entered but lost, you can determine the long-term impact of an influx of a moderate amount of wealth on Georgians.

The problem, of course, is you have to get lucky to have a random experiment that sheds light on the exact question you want answered.

Say, “Screw It”: The third option is to try to set up an experiment that isn’t really all that much like the question that you want answered, but at least is ethical. State openly the problems with the experiment and how the huge flaws in the experiment’s design are likely to invalidate the conclusions of the experiment.  Then, get picked up by the media, who expand the rather weak conclusions of the experiment into an all-encompassing definitive statement while completely ignoring the massive issues with the experimental design.

The bottom line

Determining whether GMI would work is hard (impossible). Because it’s a social science, you can’t run the experiments that would give you the evidence you’d need to be confident in your conclusions. Nevertheless, keeping in mind the difficulties inherent in these sorts of experiments, there have been several studies that begin to examine these issues.  I’ll talk about some of those in my next blog.