Thursday, December 21, 2017

Basic Income Experiments and the Case for the UBI: What can we learn?




I like the idea of a universal basic income (UBI). I have written about it numerous times over the years and I think a strong political and moral case can be made in its favour. I also think a particularly strong case can be made for it at the present moment, given the ongoing structural and technological changes to the nature of work.

At the same time, I think there are obvious practical and political hurdles that lie in its way. These hurdles include fears about how a UBI might incentivise boredom and apathy, disincentivise work, and be economically unsustainable due to the fact that a smaller and smaller cohort of economically active people will have subsidise those who are not.

On the face of it, UBI experiments seem to provide an antidote to these fears. Around the world, several small and largescale experiments have been conducted that involve many of the features of a UBI: people are given a guaranteed minimum income without needing to prove that they are seeking work or disabled. These experiments almost invariably have positive results. Those who are paid this guaranteed income see their lives improving along several metrics of well-being, and the work-disincentive is minimal.

These results have led some people to be enthusiastic cheerleaders for the idea of a UBI. Rutger Bregman, for instance, in his book Utopia for Realists expounds on these empirical results for several chapters, using them to build what seems to be a strong case for a UBI. While I share many of Bregman’s political and economic goals, I worry that the empirical case for the UBI can be overstated.

My scepticism largely stems from what I understand the case for the UBI to be, and from my reading of two reviews of the empirical literature, in particular those from Karl Widerquist and Philippe Van Parijs and Yannick Vanderborght. I want to outline the case for cautious scepticism in the remainder of this post.


1. What are we arguing for?
The case for cautious scepticism is easily stated. If you are going to use the basic income experiments to argue for a UBI, then there must be some ecological validity to those experiments. In other words, the findings from the ‘world(s)’ created by the experiment must be extrapolatable to the world of the UBI. Unfortunately, I think this is where UBI experiments tend to fall down. There are some significant disanalogies between the experimental-world(s) and the UBI-world, which make the former much less useful for understanding the latter than is commonly assumed. In formal terms, the argument is this:


  • (1) If the findings from basic income experiments are to provide good evidence as to what would happen when we implement a UBI, then those experiments must have some ecological validity (i.e. the experimental world must be sufficiently similar to the UBI-world).

  • (2) The basic income experiments tend to lack ecological validity (i.e. the experimental worlds are too dissimilar to the UBI world).

  • (3) Therefore, the basic income experiments do not provide good evidence as to what would happen if we implemented a UBI.


I’m going to take it that the first premise of this argument is relatively uncontroversial. The second premise is the one that requires support. This support can be provided in two phases. First, by explaining what the world would look like under a UBI and second, by explaining the main features of the experimental worlds set up under the basic income experiments.

Let’s focus on the UBI-world first. I appreciate that people might have different things in mind when they discuss the UBI, and that there are a number of distinct policy proposals that could be clustered under that label. I have a particular conception of what the UBI would entail that I believe conforms with the most widely-discussed variants of the UBI. That conception states that a UBI is:

(a) An income paid to all people (or, at least, all adult people) within a given state/jurisdiction, irrespective of their ability and willingness to work (i.e. a universal and unconditional income);

(b) Significant/generous in monetary size, i.e. provides enough income for someone to live on. The precise amount required might be disputable, but I would estimate something in the region of 30-50% of per capita GDP within a given state/jurisdiction. For reference this would be just under $18,000 per annum in the US and just over $20,000 in Ireland (where I currently live) on 2016 figures. The exact figure would have to take into consideration the cost of living in the particular jurisdiction.

(c) Funded by taxation revenues within the state/jurisdiction in question. In other words, the money for the UBI is not provided from outside resources but from a tax on wealth and/or income (probably both) within the state. This means some people will be net beneficiaries of the scheme and some will be net contributors.

(d) Offered in an economy in which work is becoming both less feasible for the majority (due to automation and inequality) and less desirable (i.e. less relevant or important for self-worth).

You could challenge each of these assumptions about the UBI-world. For example you could argue that the UBI should be funded by some international system of wealth transfers (so not necessarily funded from within the state) or that the figure should be lower/higher or that it could be offered in a world in which work remains as relevant as ever. But I suspect most people would not drop all of these assumptions, and even if they did, they would have to replace them with their own. Either way, those assumptions would need to be borne in mind when critically reflecting on the ecological validity of the experiments.


2. Findings from the Basic Income Experiments

So much for UBI-world. To complete the assessment of premise (2) we need to consider the features of the experimental world. Obviously, there is no single ‘world’ here. There are, rather, multiple different worlds established under the auspices of multiple different experiments. Each of these worlds varies in terms of the size of the income paid to the experimental subjects, the social/cultural environment in which it was paid, the source of the payment, the conditionality of the payment, and more. I’m not going to be able to review every single basic income experiment in detail, but I will give an overview of some of the major ones for which we have data and findings (there are some other ones that are intended or in progress that I exclude from consideration).

These experiments include the famous North American ‘negative income tax’ (NIT) experiments from the late 1960s and 1970s, as well as some more recent experiments in developing nations and Western Europe. The best data comes from the NIT experiments, which are said to have been “the first large-scale social experiment[s] to use the scientific method of randomly assigning human subjects into treatment and control groups’ (Widerquist 2005). Their results have been analysed and reanalysed several times over years. There is no simple consensus on what the causal consequences of these experiments was, but there are some general patterns. The other experiments have generated less literature, though this could change in the future. What follows is a brief overview of the main experiments. I’ll describe some of their key findings afterwards:

The Manitoban Mincome Experiment: This was an experiment conducted by the Canadian government in Winnipeg and Dauphin from 1975-1978. It covered 1300 families living in urban (Winnipeg) and rural (Dauphin) regions, separated into different treatment groups. The families had to earn less than C$13,000 per annum to qualify and they were paid a sum between $3,800 and $5,800, depending on their treatment group, for the years of the experiment (approximately 25% of the per capita GDP at the time). This was combined with a ‘take-back’ or ‘clawback’ rate of between 35-75% which was charged if they earned additional income through work (this clawback rate is a common feature of negative income tax and basic income proposals: the idea is that the guaranteed income is reduced (proportionally) if you earn some additional money yourself).

The New Jersey-Pennsylvania NIT experiment: This was an experiment that ran from 1968-1972 in urban parts of New Jersey and Pennsylvania. It initially covered 1,216 two-parent families ‘with a male head’ from Black, white and Latino communities. Single parent families were not included. This number dwindled to 983 after drop outs. The families had to earn less than 150% of the official poverty line to be selected for the payment, and were paid between 50% and 125% of the poverty line, depending on treatment group. For guidance, the payment in the 100% version of NJ experiment in 1968 was $1,000 per annum (approximately 21% of per capita GDP). The clawback rate was between 30% and 70%. The payments were received for three years.

The Iowa-North Carolina NIT experiment: This was an experiment that ran from 1970-1972 in rural parts of Iowa and North Carolina. It was intended as a rural supplement to the New-Jersey/Pennsylvania experiment. It covered 809 two-parent and female headed families. This number dwindled to 729 after drop outs. They had to earn less than 150% of the poverty rate to qualify and were paid between 50% and 100% of the poverty rate, subject to a 30-70% clawback rate.

The Seattle-Denver NIT Experiment: This was an experiment that ran from 1970-1980 in parts of Seattle and Denver. It covered 4,800 Black, white and Latino families with annual incomes below $11,000 (if single-parent) or $13,000 (if two-parent). This was the most complex and longest-running of the NIT experiments. The families were paid between 75% and 148% of the official poverty rate and subject to a clawback rate of 50%-70% (though two other non-linear clawback rates were tested at 70% minus 0.025 times income and 80% minus 0.025 times income). The incomes were paid for six years initially, but experimenters got approval to continue for an additional fourteen years for some of the families. However, the experiment was cancelled in 1980 so the families only received the income for nine years.

The Gary, Indiana NIT Experiment: This was an experiment that ran from 1971-1974 in Gary, Indiana. It covered 1,799 families, mostly single-parent Black families. The total number dropped to 967 by the end of the experiment. This was the highest drop out rate in any of the experiments. The families had to earn below 240% of the poverty rate to be selected and were paid either 75% or 100% of the poverty rate for three years. This was subject to a 40% or 60% clawback rate.

Namibian (Otjivero) Pilot Study: This was an experiment that ran from 2008-2009 in the village of Otijvero in Namibia. Every resident in the village (approximately 1,000 people) under the age of sixty received a monthly payment of 100 Namibian dollars (approximately $8 US dollars). Those over sixty received a state pension of over 500 Namibian dollars. The payment to those under sixty represented approximately 2% of per capita GDP in Namibia at the time. The project was funded mainly by a variety of international donors, many based out of Germany.

Madhya Pradesh Study: This was an experiment that was conducted in the Indian state of Madhya Pradesh from 2011 - 2012. Every adult in eight randomly selected villages was granted an unconditional income of 200-300 rupees per month (about $4-6 US dollars). This represented 6.5% of per capita GDP in that region of India at that time. Children were also entitled to half the payment received by adults. The results in the eight villages were compared with twelve other villages in which no payment was made. The project was funded by UNICEF.

Give Directly Cash Transfer Evidence: Give Directly is a charity that specialises in giving large direct cash transfers to people in developing countries. Between 2011 and 2013 they ran a program in rural Kenya that was analysed by the economists Johannes Haushofer and Jeremy Shapiro. The program sent unconditional cash payments of between $404 US dollars and $1,525 US dollars to families that met a means test criterion (did they live under a thatched roof as opposed to a metal one?). The payments were given on different schedules (monthly or single lump sum) and were targeted at different members of the household, in order to study differences in response. The payments represented a considerable sum of money, over 30% of per capita GDP for the minimum payment, rising to over 130% for the maximum payment. Give Directly calculated that the payments represented more than double the average consumption expenditures of villagers in the relevant regions. The Haushofer and Shapiro study looks at 503 households who were given the payments.

Miscellaneous Others: A number of other smaller scale studies have been conducted over the years and are often discussed in relation to the UBI. A popular one is the Win For Life program in Belgium. This was a program for lottery winners that allowed them to receive monthly payments for life as opposed to one single payment. There is also the Berlin MeinGrundeinkommen experiment that has funded basic incomes of $1,000 per month for over 40 people, and an experiment on 13 homeless men in London in 2009 that gave them each a personalised budget of £3,000. There are other examples too, but this suffices for present purposes.

So what are the findings from all of the studies? As you might imagine, they are quite nuanced and differentiated. But, in general, the outcomes are positive. The guaranteed income payments are found to have a positive impact on individual health and well-being, and do not disincentivise work to the extent that some would predict. In fact, in some cases they increase work.

Take the various North American NIT experiments (including the Manitoban mincome experiment) as an example. As described, there were several different treatment conditions across these experiments, but in most cases the payment of the income was found to correlate with positive outcomes. For example, households receiving the income typically saw an increase in educational performance and an improvement in birthweights for children born to those families (Levine et al 2004). Recipients did reduce the number of hours they worked per year. Across the five studies, the reduction ranged from 20-130 hours per year for men, and 0-166 hours per year for women. According to Widerquist (2005), most of this reduction is accounted for by workers ‘remaining nonemployed for longer if and when they became nonemployed’ and not because they reduced the number of hours they worked per week. The reduction was also highest among younger people and those with young children, suggesting that it represents an investment of time in education or child-rearing as opposed to any tendency towards apathy. The experiments also calculated how much additional tax would need to be collected as a result of the reduction in work, estimating that it would add 5-10% to the total cost.

Similarly positive results are recorded from the experiments in developing countries. But in those cases all the reports suggest that the increased income correlates with an increase in economic activity, not a decrease. The Give Directly study by Haushofer and Shapiro found that the payments significantly increased (self-reported) psychological well-being and that people receiving the payments invested significantly in assets that could help them in their businesses (mainly agricultural). This led to an increase in revenue from those businesses, but not necessarily an increase in profit. The study did not find any significant increase in consumption of alcohol or drugs. The Namibian pilot project reported a number of positive outcomes, including large increases in economic activity and GDP within the village, large reductions in crime, and improvements in nutrition and well-being. This was combined with an increase in inward migration (which is something that tends to worry people contemplating basic income proposals) Some of these findings are disputed and alleged to be exaggerated but they have been defended by those involved in the program and are broadly consistent with other findings. The Madhya Pradesh studies also reported positive outcomes, with improvements in sanitation, nutrition, education and increases in labour and work.

The miscellaneous other studies that I mentioned highlight similar positive outcomes. All of which suggests that the case for the UBI has some robust empirical support.


3. The Mismatched Worlds of Experiments and UBI
But this is where we have to return to the ecological validity question. Taking the studies at face value, and assuming that all their results are accurate, is it true to say that the ‘worlds’ established by the experiments are sufficiently similar to the world envisaged by supporters of the UBI? Not really. There are important differences between the world of the UBI and the various experiments that makes it difficult to extrapolate from one to the other.

One of the main problems with the experiments — and this will always be a problem until we actually trial a UBI — is that they are not truly universal. All the NIT experiments were targeted at specific population groups in specific regions, making it difficult to draw firm conclusions about what would happen if the scheme was rolled out more generally. They had a variety eligibility conditions attached to them, were paid to households as opposed to individuals, and were of limited duration. The recipients of the income knew the payments were only going to happen for a limited time and this could have altered their behaviour, making them more inclined to invest the income in education or other activities. The one exception to this might be the Seattle-Denver experiment which was supposed to last for 20 years for some recipients. Interestingly, that experiment is the one that saw the largest reduction in hours worked, with experimenters noting that it took 2.4 years for people to adjust their behaviour to the program (Widerquist 2005).

In addition to this, the NIT experiments were not compulsory. People self-selected to receive the income, which could be problematic since the sample of people receiving the income might be those who were most likely to make good use of it, and we should remember that there were high drop out rates in some of the experiments. Furthermore, as Van Parijs and Vanderborght point out, all the NIT experiments excluded households that would belong to an income category that would make them a net funder of the scheme; only net beneficiaries were included. This means we have no evidence for ascertaining the behavioural effects on the net funders if we introduced a UBI.

The experiments conducted in developing countries also have significant problems. It is true that in India and Namibia the incomes were given to everybody within selected villages. But the sums paid in those cases were extremely modest (less than 10% of GDP per capita) and were for a limited duration. The Give Directly cash transfers were much more generous, but they were also limited, one-time payments, given to specific sub-groups within selected regions. I believe Give Directly is changing this now and trying to provide payments to everyone within a particular locale, but I have not read any data on those schemes. Furthermore, in all of these cases the funding came from outside the jurisdictions in which the payment was received.

The miscellaneous small scale experiments also suffer from a number of weaknesses that make it difficult to extrapolate from them to the world of the UBI. They tend to be unique, one-off experiments, of minimal or uncertain duration, affecting very specific or narrowly defined sub-populations.

Future experiments may correct for these issues, but more important than all of them is the fact that no UBI experiment can examine the effects of UBI in a world in which work is less important, and less available. At the moment, anyone receiving a basic income is doing so in a world in which they are otherwise massively incentivised to work. It makes sense, in this social environment, for at least some of them to invest the additional income in building capacities for future employment. But what happens if, as many futurists claim, part of the motivation for the UBI is that work is less of an option for the vast majority of people? Furthermore, what happens if one of the policy goals behind a UBI is that it makes work less of an economic necessity for people? To me, this is one of the most important features of the UBI: that it frees people from the economic compulsion of work. UBI experiments cannot explore this possibility. Van Parijs and Vanderborght sum up the problem pretty well:

Even the best-designed experiments cannot capture the effects on the labour market which are at the core of the case for basic income versus conditional minimum-income schemes…the universality of basic income opens the possibility of saying yes to some jobs that are currently not viable and its freedom from obligation [conditions] opens the possibility of saying no to other jobs, which will therefore need to pay better or improve if they are to have incumbents. Such effects…have no chance of showing up in the experiments, partly because of their limited duration but above all because they affect only a few hundreds or thousands of individuals in labor markets of several millions. 
(2017, 143)

That said, the experiments discussed above do provide good evidence for something. They all suggest that guaranteed income poverty reduction schemes, that are not premised on the willingness or ability to work, and that are targeted at particular populations, can have very positive outcomes. We certainly should not stop experimenting with these schemes. But we should be cautious about using them to build a case for the UBI. There is a real sense in which implementing the UBI itself would be the first major experiment.




No comments:

Post a Comment