Two Problems with Designing a Basic Income Experiment

WARNING: I don't do (micro) Public Finance. And I don't do experimental economics. Those who have read the literature may tell me they are well aware of these problems, or that I'm wrong about something.

Suppose someone asked me to design an experiment to test whether Basic Income would be a Good thing. (They wouldn't ask me, but let's suppose.) I can see two major problems.

1. Temporary experiment vs permanent policy.

Let us suppose that the worst fears about Basic Income were true. If you gave anyone (say) $10,000 per year whether they worked or not, everyone would stop working. Just suppose. Would the experiment be able to tell you that? Because if the experiment weren't able to tell you that, if it were true, it wouldn't be a very informative experiment.

If the experiment were temporary, and known to be temporary, it probably wouldn't be able to tell you that. Because getting $10,000 per year for one year is very different from getting $10,000 per year for life. (Colby Cosh tells me the Dauphin experiment did run for 4 years.)

If people do spend the extra money on taking time off work, that doesn't mean they spend it all immediately. If the average person in the experiment will work for (say) 20 more years, and people spread their extra time off work evenly over their remaining years ("consumption-smoothing"), the effects of a lifetime of basic income will be 20 times bigger than measured by a one year experiment.

It could be even worse than that, because people tend to bunch their leisure towards the end of their lives. We call it "retirement". Maybe the effect of getting $10,000 per year for one year will be that people will retire one year earlier. In which case, the experiment will show zero effect for everyone, except the small fraction of people who were planning to retire soon. If a one year experiment cuts working life from (say) 40 years to 39 years, the effects of a lifetime of basic income will be 40 times bigger than measured by a one year experiment.

You might need to run the experiment for three score and ten years to get some idea of the life-cycle effects. Which is a bit awkward.

2. Funded experiment vs self-funded policy

It's OK if experiments cost money. It's worth paying (sometimes) to get the information. But a basic income policy needs to be self-funding. Free ponies don't grow on trees. If you simply give everyone in the experiment a free $10,000 per year, then (as Stephen Gordon tweeted) good things will very probably happen. But what we want to know is whether good things will happen if we give a group of people $10,000 per year each and that same group of people have to somehow pay for that $10,000 per year each. What we want is a self-funding or revenue-neutral experiment (except for the cost of paying the researchers). The subjects would have to give up all other forms of support that the basic income is designed to replace, and pay the extra taxes that would be needed to make up the difference. Labour supply is not just about income effects; it's about substitution effects caused by changing the marginal (net) tax rate.

Almost any change in tax/transfer policy creates winners and losers. And with informed consent, the losers won't sign the Consent Form. Which would make it a bit awkward getting it past the Research Ethics Board.

It's the external validity of experiments that matters, not the damned T-statistics.

But, as a (moderate) supporter of Basic Income, I think I might just have a Cunning Plan. Which I will leave till later, if and when my head is clearer. But I think we are thinking about this Basic Income thing the wrong way.

Comments

[Kevin Milligan]

Thanks Nick--these are two serious problems with the "pilots" that seem to be planned in Ontario and Finland. I'm really not sure what we would learn from them--especially since the experiment in Finland is apparently only going to use current benefits recipients as the treatment group.

There are also issues with mobility and availability (even just contingent availability) of the vast array of social supports that might not be there if a true UBI replaced all income security.

About your own Basic Income plan. Having interacted with you a bit on this, I have a good sense of what you have in mind. But you should be clear that what *you* mean by basic income is very different than most of the policy chatter on this topic, so I don't know you want to taint your ideas with the 'basic income' label!

[Nick Rowe]

Thanks Kevin. I am only slowly learning that how I think of UBI/GAI/NIT isn't how some of its (their?) supporters think if it. Maybe the different names actually do mean different things?

Mobility. Hmmm. Hadn't thought about that. The experiment would have to follow the subjects, not be location-specific?

[Avon Barksdale]

Friedman was pretty clear on all this stuff a half century ago.

[Nick Rowe]

Avon: Permanent Income Hypothesis? Yep.

[Henry]

"Permanent Income Hypothesis?"

Keynes may have briefly beaten Friedman to the punch in considering such a proposition.

Page 95 of GT:

"Changes in expectations of the relation between the present and the future level of income. We must catalogue this factor for the sake of formal completeness. But, whilst it may affect considerably a particular individual's propensity to consume, it is likely to average out for the community as a whole. Moreover, it is a matter about which there is, as a rule, too much uncertainty for it to exert much influence."

He's saying in the real world that uncertainty about future income precludes any such consideration.

For Keynes, uncertainty is what essentially renders classical theory invalid.

[Eric Pedersen]

It seems like there might already be (quasi) experimental data for this... aren't a lot of lotteries set up to give people the choice between a one-time payout and a fixed per-year payout? How do people who receive relatively small awards (equiv. to 10,000 per year for someone's working life, so I would guess comparable to ~$500,000 or so) change their working behaviour relative to matched people who didn't win the award? It also deals with the self-funding issue in a way, as lotteries are already funded via other people paying to enter the lottery.

[Jussi]

"experiment in Finland is apparently only going to use current benefits recipients as the treatment group"

IMO not true, government's prestudy dated 3/30/2016 suggest using both random and weighted random study for the experiement. Weighting method will be used to gain more information about special groups (current benefits recipients?).

[Nathan W]

If I was unemployed or decided to take fewer hours in a dead end job thanks to $10,000 a year guaranteed income, I would invest in specific skills building, have more time to join clubs where I could demonstrate skills and build networks, and would have more money to invest in small business development.

Maybe not everyone's like that. But when there is no clawback (as opposed to welfare), I think most people are rather more like me than the "lazy bums" we're warned about.

[Frances Woolley]

Nick: Thanks for doing some clear thinking on this. One quibble:

"But what we want to know is whether good things will happen if we give a group of people $10,000 per year each and that same group of people have to somehow pay for that $10,000 per year each."

I'm not sure about that. I agree that the program as a whole has to be revenue neutral, which means someone is going to have to pay for it. But think of it as social insurance - there but for the grace of God go I. Ex-post, income distribution is not self-funded in the sense you describe above, i.e. some people gain more than they contribute over their lifetime. But ex-ante it can be self-funded in the sense that everybody's expected benefits over their lifetime are zero (or actually slightly negative, because of admin costs as you say) - just as car insurance is self-funded.

[Some One]

Nick Rowe wrote:

What we want is a self-funding or revenue-neutral experiment

I think that's a way too strong requirement that severely limits the scope of the policy change artificially.

Many of the benefits of Basic Income are expected to be intangibles:

• improvements in competitiveness: under Basic Income more young adults can afford college - or can go through re-training - which improves the general educational levels of the workforce
• an increase in quality of life: seniors having a peace of mind, not being forced to work in jobs they find physically challenging to perform, etc.
• general health improvements: seniors not having to work menial jobs - these could in fact increase costs

Furthermore, many of the expected benefits of Basic Income are compound effects driven by strong network effects:

• reduction in petty crime in a given city that has Basic Income (due to people who fall out of the social net not being forced to subsist)
• reduction in corruption (due to low wage jobs not forcing existential corruption - like it happens in many poorer countries)
• reduction in social safety net administration costs (bureaucracy)
• improvement in business flexibility and confidence (for example due to no minimal wage requirements: Basic Income allows the elimination of a minimal wage)

Due to tragedy of the commons effects and due to network effects such policies will be most effective if they are 'complete' geographically and socially.

Plus, there are policy variants where the 'costs' are spread out so much that literally there's no direct 'losers' of the BI policy:

• for example if Basic Income is financed through accounts and money created by the central bank electronically, and the monetary base is directly increased by Basic Income (i.e. the Basic Income is not sterilized and there's no matching revenue for it), then the main 'cost' effects of Basic Income are an increase in inflation and/or a weakening of the exchange rate in export-heavy economies.

All of these are public goods that may 'cost money' if you consider them in terms of a revenue neutral experiment, but which are very expensive to purchase directly (sometimes they are impossible to purchase directly, such as a reduction in crime or an improvement in health), so even though they may cost revenue, they are a significant net plus from a general policy POV that is very much tangible.

Revenue-neutral small-scale experiments won't be able to capture the full benefit of such benefits.

Small-scale experiments should be instead be funded independent of revenue (not limited to be revenue neutral), so that they are representative of the long term policy, and then economic studies of the cost effectiveness should import expected long term benefits.

[Patrick]

Can't comment on the technical issues, but as a practical matter I worry that we'd get the worst of both worlds - the nanny state nonsense would stay and we'd just pile on an expensive new program with screwed-up incentives.

And I say this as a leftie.

[Majromax]

> If the experiment were temporary, and known to be temporary, it probably wouldn't be able to tell you that. Because getting $10,000 per year for one year is very different from getting $10,000 per year for life. (Colby Cosh tells me the Dauphin experiment did run for 4 years.)

There are two distinct but equally-important "treatment groups" for basic income:

*) The first group is those that are already on social assistance, who would see this assistance replaced with a basic income.
*) The second group is those who are not already on social assistance, who would see a direct transfer payment for the first time.

For the first group, we already know there are strong work disincentives. One principal goal of a basic income experiment would be to assess if there are any work incentives to the simpler structure, in that it eliminates welfare walls and extremely high marginal clawback rates.

This second group might experience work disincentives, but we must also be aware that this is a second-order effect. The work disincentive of welfare clawback is built right into the marginal clawback rate; the work disincentive from the "income effect" comes into play only when we look at subjective utility-versus-income. (That is, an agent with linear utility of money will experience no work disincentive at all from a lump-sum payment). This sort of secondary effect also gets balanced against other secondary effects such as income-stability leading to greater human capital accumulation.

(On the other hand, the benefits suggsted by Some One above are tertiary benefits, in that they relate to sociological interaction between the recipient and broader society.)

> Let us suppose that the worst fears about Basic Income were true. If you gave anyone (say) $10,000 per year whether they worked or not, everyone would stop working. Just suppose.

We can dismiss this possibility out of hand, because people don't retire as soon as they can fund a $10k per-year annuity. We only need to design the experiment to differentiate between already-plausible cases, and here we have several competing but reasonable ideas on why people work:

1) To avoid abject poverty and provide for the necessities of life,
2) To avoid relative poverty and maintain social standing with peers,
3) To provide for luxury consumption,
4) To provide for income security against the future, either though job security proper or savings,
5) To maintain social standing when work is normatively expected.

Basic income directly deals with points #1 and #4, and it may also impact #3 depending on individual preferences with respect to luxury goods. #2 and #5 are more along the lines of social policy, and we can have good (working) equilibria and non-working equilibria -- see the problem some nations have with "NEETs", or extremely high-unemployment ghettos.

A well-designed basic income experiment would look at the introduction of basic income with respect to work-effort in both high and low-unemployment regions. It might be (hypothetically) that basic income works well when we have a poor individual in an otherwise high-functioning economy, but it may work terribly if an entire community is out of work.

[Majromax]

@Frances Wooley:

I agree with you on the "self-funding" aspect. It's important that an experiment looks the same as a final implementation with respect to the marginal, low-income worker, but we already have a good idea on the labour-market responses of tax changes elsewhere in the distribution. If someone proposes funding basic income via (say) a GST increase, we can immediately ballpark-estimate the labour response to the tax increase part of the system.

[Avon Barksdale]

No Nick, not the PIH. I'm talking about Friedman's discussion on a guaranteed minimum income. If you get a chance, watch Friedman's PBS Free to Choose series.

[Nick Rowe]

Eric: "It seems like there might already be (quasi) experimental data for this... aren't a lot of lotteries set up to give people the choice between a one-time payout and a fixed per-year payout?"

Interesting. I think I remember reading some economist(s) studying the effect of lottery wins. But: the sample who buy lottery tickets may not be representative?; the "taxes" that fund lotteries are not income-contingent; perhaps most importantly, it's a random transfer from unlucky to lucky, so I'm not sure if the lessons are applicable to BI (but some should be).

Frances: thanks. OK, it has to be self-funded for the sample in the expected value sense (assuming it's a representative sample of the population). I think that's right. But then if the sample gets hit by the same negative shock that hits the population, the sample project should only be allowed to go into deficit to the same extent the existing system goes into deficit?

Some One: I was sorta with you up to this point:

"Plus, there are policy variants where the 'costs' are spread out so much that literally there's no direct 'losers' of the BI policy:

for example if Basic Income is financed through accounts and money created by the central bank electronically, and the monetary base is directly increased by Basic Income (i.e. the Basic Income is not sterilized and there's no matching revenue for it), then the main 'cost' effects of Basic Income are an increase in inflation and/or a weakening of the exchange rate in export-heavy economies."

No. It's arguments like that that destroy all credibility of BI proponents.

Look, my day job is writing posts on things like Helicopter Money. And there may or may not be a case for helicopter money in some circumstances. ***But that case applies equally to funding the existing tax/transfer system, or infrastructure investment, or whatever***. When proponents of BI say "Don't worry about the cost, because we can fund it with free money!" I know they don't understand very basic things like opportunity costs. They simply haven't thought it through. Hippies for BI!

Patrick: as a righty, I agree that would be my worst fear. And when I hear some proponents, I fear that your fear may be realistic.

Majro: the way I approach BI is I think similar to you. We already have a BI system -- we call it "welfare". It's just a crappy BI system with lots of exceptions and weird and wonderful stacked clawback and tax rates.

Avon: ah, OK. Saw/read Free to Choose ages ago. I've probably internalised it.

[Tel]

Rather than doing wealth thransfer pf a sizeable amount for a limited time, seems that a better experiment would be to transfer a limited amount for a much longer time and then measure the effect of that. If it seems promising, go for a slightly larger amount and measure again.

Oh wait! We already did that, in the US it's called food stamps. The measurement sure seems to show correlation between more food stamps and less employment.

Some will no doubt tell me that correlation is not causation, but I can't imagine you guys getting any sharper results from your experimental design. I mean, most economists already know what answer they need to get, and after that it's just a matter of choosing the best interpretation to get the desired outcome.

[Nick Rowe]

Tel: fair point. But could be reverse causation in that graph, or simply some third variable causing both. The beauty of experiments is that you can have a random sample that you can compare with a control group or the population.

[Miami Vice]

Do you have to be looking for work to qualify for food stamps or is poverty the only requisite?

[Miami Vice]

https://research.stlouisfed.org/fred2/graph/fredgraph.png?g=48bc

[Majromax]

@Tel: You cannot infer causation from that graph because it is not clear what the "treatment" consists of.

Remember that SNAP is an ongoing policy. Absent change to that policy, there is a "food stamp response function" that adjusts the amount of aid provided in response to poverty, which is in turn correlated with non-employment status.

If food stamps cause less employment (which seems a reasonable hypothesis, through mitigating abject poverty), then much of that response is built in to the passive, background environment. That is to say, food stamps could not cause employment drops in this environment since the labour market was already in equilibrium with the program.

In terms of policy changes, one page notes that maximum benefits were extended from 2009 to 2013. That coincides with the tail end of the recession, but it could not have caused the stronger drop in employment during the first 2/3 of the recession.

Finally, the program is not equivalent to any sort of basic income test, in that in many cases benefits are fully withdrawn well before a worker would have full labour-market participation (that is, full-time employment even at minimum wage). For households in these situations, SNAP is not a universal income support but instead one received conditionally upon not working (much), which would be expected to enhance any negative labour-force effect.

[Evan Calford]

Nick, as an experimental economist I'd like to nitpick on one little thing (it may be a touch off topic here, but you put it in bold so I felt the need to respond!).

You wrote: "It's the external validity of experiments that matters, not the damned T-statistics."

Which is completely accurate in the context of an experiment designed to test the effects of a UBI (and I agree that GE effects are important here and probably not possible to capture in an experiment).

But, in general, external validity of an experiment is not the only thing that matters. Particularly for lab experiments, the goal is often to isolate (and test for) a single mechanism. If we can't get the mechanism to "work" in a controlled environment where we can shut down confounds, then either the mechanism isn't "real" or we don't understand it properly. On the other hand, if the mechanism does "work" in the lab, then we can tentatively take steps to try and test and embed it in more realistic environments. During this process, a combination of theory and more general experiments can be used to build a case for external validity of the mechanism.

For studying and understanding basic behavioural mechanisms, the drive for external validity can sometimes be a hindrance rather than a help.

[Nick Rowe]

Evan: thanks. Good to hear a real experimental economist chime in. Seems to me you make a fair and reasonable point point. I was being slightly hyperbolic in that bolded bit!

[Bram Stolk]

I expect the basic income to cover food, shelter and little more.
Do you think the modern consumer will forgo the iPhone7 and the holiday abroad by stopping working?

There's little danger of happening that, I think.

On the other hand: legions of wannabe creatives will cancel employment, and do their hobby writing/painting/videogame-authoring, what have you.

Bram

[Noni Mausa]

One thing I would enjoy watching would be the disciplinary effect on employers. Employees would be able to leave bad employers much more easily, and could afford to work for good employers at a lower wage level (or as volunteers or trainees) Even if the employment to population ratio stayed the same, the effect would be to tighten the labour market, with corporate culture becoming a much more important part of the mix.