Here’s a common argument:

The problem with the poor is that they haven’t got enough money. There’s ample empirical evidence backing this up. Therefore, the obviously-correct poverty intervention is to simply give the poor cash. You might be able to do better than this, but it’s a solid baseline and you should often expect to find that interventions are worse than cash.

There are technical reasons to be skeptical of cash transfers—which is why it is so important that the cash transfer charity GiveDirectly is carefully researching what actually happens when they give people cash—but until fairly recently, these objections seemed to me like abstruse nitpicks about an intervention that was almost analytically certain to be strongly beneficial.

But they’re not just nitpicks. Cash isn’t actually the same thing as human well-being, and the assumption that it can be simply exchanged into pretty much anything else is not obviously true.

Of course, saying “X is possibly wrong” isn’t very helpful unless we have a sense of how it’s likely to be wrong, under what circumstances. It’s no good to treat cash transfers just the same as before, but be more gloomy about it.

I’m going to try to communicate an underlying model that generates the appropriate kind of skepticism about interventions like cash transfers, in a way that’s intuitive and not narrowly technical. I’ll begin with a parable, and then talk about how it relates to real-world cases.

Two cities, two stadiums

In a world where the only thing people enjoy is baseball, there is a wealthy city. In this city, there is an excellently designed baseball stadium. The seats are amply sized and comfortable. Even the worst seats have a good view of the field. There are plentiful amenities, though the price of food and drink is high. There are awnings to block the rain. There are lights in case the game goes late.

Elsewhere, in a poor city, there is another baseball stadium. This one is shoddily built. The upper seats are tiny to cram in as many people as possible. The worst seats can hardly see the field, or are exposed to inclement weather. The aisles are too narrow. The mood is chaotic, and in the poorer areas of the stadium – but not in the expensive areas with the best views – people often get into fights. A smaller variety of cheaper, lower-quality food and drink is available. There is always a line for the bathroom, unless you paid for a box with a private one.

You happen to live in the rich city, and attend the rich stadium. You know two more things, from studying data within and between many cities in this world:

Within a city, the self-reported enjoyment of baseball scales logarithmically with individual wealth. In other words, if you ask people to rate their happiness on a scale from 1 to 10, then no matter how rich or poor someone is, they’ll be the same distance on the scale from someone with twice their wealth. (For instance, if people with an annual income of $10,000 report an enjoyment level of 5.2 out of 10 on average, and people with an annual income of $20,000 report 5.5 out of 10, then people with an annual income of $40,000 report 5.8 out of 10.)
Between cities, self-reported enjoyment of baseball also scales logarithmically with (average) wealth.

Clearly, enjoyment is a simple positive function of money; the problem with the people in the poor stadium is that they haven’t got enough of it. What’s more, since the function is logarithmic, a dollar goes much farther for the poor than for the rich. So you send some to the poorest people in the poor stadium, hoping that it will do then some large multiple of the good it can do you.

Because you know that good-sounding charitable interventions often fail for surprising reasons, you decide to test your assumptions, by following up with the recipients of the cash transfers. What do you find?

It turns out people care about two things: the quality of their seat, and food.

Some recipients buy more, or better food. Because the prices are lower over there, the difference in quality is large for them, even though the money would only make a small difference to them. Other recipients buy their way into better seats. Again, since the seats in the poor stadium cost less than the seats in the rich stadium, they gain a lot more than you lose. Overall, it looks like everyone who receives money enjoys the game a lot more, so your belief in the merits of cash transfers has been confirmed.

Then you learn about a third statistical regularity that flies in the face of everything you’ve learned so far:

As individual cities get richer, average enjoyment of baseball games does not increase.

What’s going on?

When someone pays for on a nicer seat, their experience of the game is improved. But if they’ve outbid someone else for that seat, that other person is now stuck in a worse seat. Buying someone a nicer seat in the same stadium does not improve the average game enjoyment, because – assuming fully booked stadiums – it makes someone else’s experience worse, because they’re now stuck in the bad seat. So it matters quite a lot how much of the variation in people’s well-being comes from things that can easily be got more of, like food, and how much comes from locally scarce goods, like choice seats.

But richer cities have nicer stadiums! Doesn’t that mean that if you transfer enough money to people in poor cities, the poor city stadiums will get better? Maybe not! Stadiums are pretty hard to change substantially once built. And maybe it was never the money that made the stadiums better; maybe cities that have their act together tend to be better both at making money, and at stadium-building. You don’t have empirical reason to doubt this.

How should your strategy to improve people’s baseball experiences take this into account?

First, cash transfers can still be of some value. For instance, the very poorest spectators may go hungry. If you send them money, and they use the money to buy food, they might enjoy the game a lot more without harming anyone else. But if you keep giving them money, eventually they’ll have enough food. The only thing left to buy is a positional good: better seats.

Second, if there are high-wage cities with spare stadium seating, you might want to help people move there. They’ll enjoy baseball games more, since even the comparatively bad seats will be better, without harming anyone else. Some might turn down the opportunity out of loyalty to their hometown team, but others might take you up on it.

Third, if there is a way you can improve your home stadium, this is an important good. If you’re willing to learn foreign cultural norms and be genuinely curious about why poor people have worse stadiums, you might even be able to help them reform their institutions to get better stadiums built, which could make a big difference in the quality of their lives.

The second and third points go together well. If your town’s stadium is at capacity, immigration doesn’t help anyone enjoy a nicer game – but persuading the owner to add more seats can change that.

How this applies to the real world

In The Price of Glee in China, Scott Alexander points out a few key facts about the happiness literature:

Within countries, self-reported well-being seems to scale logarithmically of average income. (Again, that means that each doubling of income corresponds to the same, constant increase in reported happiness points.)
Between countries, a similar relationship seems to hold.
Within countries, per capita GDP growth does not appear to lead to corresponding increases in well-being.

These are the same three statistical regularities I gave in the baseball hypothetical, and we should draw similar conclusions. GiveDirectly is an excellent cash-transfer charity. It’s on the GiveWell top charities list in large part because it is taking such care to collect evidence about what actually happens when the global poor are simply given money.

To that end, I found this Vox article about GiveDirectly’s basic income experiment interesting. In particular, it’s interesting that the headline case is someone who sometimes went hungry, but is not going to spend the money on food. Instead, she’s going to spend it on what’s plausibly a positional good instead:

She is expecting her third child very soon. […] I asked Jacklin if she’s ever gone the whole day without eating; she has. I asked when the last time this happened was. She told me, “Last week.”

But when the nonprofit GiveDirectly told her that it would give her, and every other adult in her village, a basic income payment of 2,280 Kenyan shillings (about $22) a month for the next 12 years, she knew immediately that she would not spend the money on food.

Her plan is to save the money and then use it to pay her children’s school fees.

She is starving herself, while pregnant, in order to save for her kids’ formal schooling. This looks like a really bad outcome.

A brief digression on education

But education! She’s investing in her kids! Isn’t that good?

The education industry is already eating the developed world alive, with little apparent benefit. You might argue that there are diminishing returns – intuitively, education seems important. But, there’s ample evidence that developing-world education often doesn’t really improve people’s productive capacity; it often seems no better than obedience school, when it’s not being used purely as a certification of the ability to obtain a degree (and thus that you’re in the right social class for certain positions).

For a particularly dramatic example, consider Scott’s report on his experience in Haiti:

Even if you’re one of the lucky ones who can afford to go to school, your first problem is that the schools can’t afford paper: one of our hosts told stories of Haitian high schoolers who were at the level of Western 5th graders because they kept forgetting everything: they couldn’t afford the paper to take notes on!

The other problem is more systemic: schools teach everything by uninspired lecture even when it’s completely inappropriate: a worker at our camp took a “computer skills” course where no one ever touched a computer: it was just a teacher standing in front of the class saying “And then you would click the word FILE on top of the screen, and then you’d scroll down to where it said SAVE, and then you’d type in a name for the file...” and so obviously people come out of the class with no clue how to use an actual computer. There’s the money issue—they couldn’t afford a computer for every student—and a cultural issue where actually going to school is considered nothing more than an annoying and ritualistic intermediate step between having enough money to go to school and getting a cushy job that requires education.

These are both entirely consistent with a story where education improves individual outcomes by inducting people into the “educated class”. And Scott continues:

We heard horror stories of people graduating from nursing school without even knowing how to take a blood pressure—a nurse who used to work at the clinic would just make her blood pressure readings up, and give completely nonsensical numbers like “2/19”. [W]hen cornered this nurse absolutely insisted that the blood pressure had been ²⁄₁₉ and made a big fuss out of it.

Likewise, bureaucrats also do not seem to be able to do the sorts of tasks we would expect formal schooling to qualify you for. One such task is alphabetization:

Gail, our program director, explained that she has a lot of trouble with her Haitian office staff because they don’t understand the concept of sorting numerically. Not just “they don’t want to do it” or “it never occurred to them”, but after months and months of attempted explanation they don’t understand that sorting alphabetically or numerically is even a thing. Not only has this messed up her office work, but it makes dealing with the Haitian bureaucracy—harrowing at the best of times—positively unbearable.

Gail told the story of the time she asked a city office for some paperwork regarding Doctors Without Borders. The local official took out a drawer full of paperwork and looked through every single paper individually to see if it was the one she wanted. Then he started looking for the next drawer. After five hours, the official finally said that the paper wasn’t in his office.

While investing in education to get a higher future salary is a net good in a simplified economic model, in practice it’s often just buying what economists call rents. Even if those rents sometimes take the form of a job, it’s clear from Scott’s example that schooling is not enabling government bureaucrats to create more value for taxpayers than less-schooled people would be able to do – they’re just outcompeting the unschooled for a fixed pool of jobs where they don’t do much.

I suspect Haiti is especially bad for a bunch of reasons, but I don’t know how exceptional it is. And it would be weird for marginal education to be bad in exceptional basket cases like Haiti, bad in well-to-do countries like the US, but good in the middle.

Here’s another anecdote from a friend:

I had the opportunity to observe one poor school in India and indeed, school literally didn’t happen most days of the year for one reason or another. (Waiting for textbooks was one reason I remember them giving.) Also, the teachers were stealing the food the rotary club provided for free lunch. (Someone noticed and then they stopped, which I suppose is nice.)

And often when they showed up to school they were often grading papers from other schools for separate pay instead of teaching the kids.

At least they could get the teachers to show up to steal the lunches sometimes.

Banerjee and Duflo’s Poor Economics is consistent with the view that this is a common problem in developing countries. In Chapter 4, they write:

In 2002 and 2003, the World Absenteeism Survey, led by the World Bank, sent unannounced surveyors to a nationally representative sample of schools in six countries. Their basic conclusion was that teachers in Bangladesh, Ecuador, India, Indonesia, Peru, and Uganda miss one day of work out of five on average, and the ratio is even higher in India and Uganda. Moreover, the evidence from India suggests that even when teachers are in school and are supposed to be in class, they are often found drinking tea, reading the newspaper, or talking to a colleague. Overall, 50 percent of teachers in Indian public schools are not in front of a class at a time they should be. How are the children supposed to learn?

In 2005, Pratham, an Indian NGO focused on education, decided to go one step further and find out what children were really learning. […] Close to 35 percent of children in the seven-to-fourteen age group could not read a simple paragraph (first-grade level) and almost 60 per-cent of children could not read a simple story (second-grade level). Only 30 percent could do second-grade mathematics (basic division). […]

Unfortunately, India is not unique: Very similar results have been found in neighboring Pakistan, in distantKenya, and in several other countries. In Kenya, the Uwezo Survey, modeled on ASER, found that 27 percent of children in fifth grade could not read a simple paragraph in English, and 23 percent could not read in Kiswahili (the two languages of instruction in primary school). Thirty percent could not do basic division. In Pakistan, 80 percent of children in third grade could no tread a first-grade-level paragraph.

In addition, parents seem to view education as a way to buy a credential, as a ticket into an “educated class” job, not a way for their children to pick up valuable skills continuously:

Parents seem to see education primarily as away for their children to acquire (considerable) wealth. The anticipated route to those riches is, for most parents, a government job (as a teacher, for example), or failing that, some kind of office job.

Parents thus would rather pay for the complete education of their highest-potential children, then pay for a mostly-complete education for all their children. (Poor Economics does claim that the all-or-nothing credentialist view is unrealistic and there are substantial gains for each year of education.)

I believe very strongly that learning is an important form of real capital, and it’s entirely possible that formal schooling is more like this in the developing than in the developed world, but I haven’t see the evidence, and at this point I think that if you try to justify the benefits of a social program by pointing to the bare fact that people are using it to buy more formal schooling, I think this is mostly adverse rather than positive evidence. (I’m more optimistic about educational programs like SOLE that try to route around the hierarchy, and focus exclusively on learning rather than credentials.)

Some investment is real

While some investments people make to better their or their children’s circumstances are positional, others seem much more like the sort of real investment we might hope for, under the usual economic framework. For instance, one common use of GiveDirectly’s cash transfers is to buy a metal roof, which lasts much longer than the more commonly used and cheaper thatch roofs. The most conservative estimates GiveWell cites suggest that the annual return on investment for metal roofs is 7-14%, though the other estimates they report are substantially higher.

Importantly, the return on investment for a tin roof is not plausibly extractive. Even though purchasing the roof imposes a cost on the world (by increasing demand for the relevant inputs), it also reduces a cost (by reducing demand for thatch roof inputs), and the reduction seems to be substantially greater than the increase.

At least some recipients of cash transfers try to start businesses, which is maybe the paradigmatic case of an investment for which we should expect a return. But while locals know their local economy much better than I do, I am skeptical of a business plan where the basic income is being used to subsidize variable rather than fixed costs. As Vox reports:

Samson [...] explained his plan to go into cage fishing at the lake. He’d already bought the fish and just needed to buy feed, and the feed — per a catalog he showed us — is expensive. So he’s going to use the GiveDirectly money for that part of the operation.

Business acumen, like anything else, is a skill that can take time to develop—while there are serious disadvantages to making decisions for people from afar, we should at least not expect that cash transfer recipients will immediately make reasonable business decisions.

How should we think about cash transfers now?

I hope that most readers will be see that the point here is not that cash transfers are bad. The point is that when you look at the evidence about what happens when rich people send poor people money, good news won’t look like a tedious confirmation of an obvious truth, but rather like an encouraging outcome where one should have been actually uncertain beforehand. And you should be able to recognize ambiguous or bad news as such, and not assume that it’s good by definition.

I hope that it will also motivate some amount of additional justified skepticism of nominal rates of return, as estimates of the social value of an investment. You, as someone living in a rich country, might not think that your access to higher-wage jobs is a reliable measure of how much value you’re able to provide to the world. (If you did think this, the most benevolent thing you could do would be to maximize your nominal wealth.) If you are in fact skeptical of the meaningfulness of your income as a metric, you should be similarly skeptical of the meaningfulness of variations in income of people in poor countries.

Finally, while abstractions like income levels and average self-reported happiness can be good starting points for generating hypotheses, I hope I’ve been somewhat persuasive that they are not adequate metrics for making philanthropic decisions—one has to engage with what’s concretely happening in the world.

(Cross-posted on my personal blog and LesserWrong)

Cash transfers are not necessarily wealth transfers

Two cities, two stadiums

How this applies to the real world

A brief digression on education

Some investment is real

How should we think about cash transfers now?