The Effects of a Free Lunch
An Experimental Analysis on the Effects of Unconditional Guaranteed Payments on Labour Supply under Taxation
Master Thesis Elmar Cloosterman
Student number: 10083499
Master Economics (Track: Behavioural Economics & Game Theory) Supervisor: Ailko van der Veen
Date: 28-08-2017
Abstract
The idea of an Unconditional Basic Income (UBI) has been a persistent idea in western societies that gained a great deal of attention in recent years. The past few decades several large scale field experiments have been conducted by both the US and Canadian governments to investigate the effects of an UBI on labour supply. These experiments are often criticized for being constructed in highly politicised environments. Therefore, this paper investigates the effects of an UBI on labour supply and taxation by making use of laboratory experiments. Experiments conducted were real-effort experiments in which treatments differed in the conditionality of non-effort income. No significant differences for labour supply were found. Tax revenues were significantly higher when an UBI was in place and tax rates were above fifty percent. Results indicate that an UBI would not affect labour supply. However, due to relatively small number of subjects, one could raise questions on the robustness of these results.
Table of Contents
1. Introduction 5
2. Previous literature 9
2.1. Universal Basic Income 9
2.2. Labour supply and taxpayer behaviour 12
3. Experimental procedure 16
4. Hypotheses 20
4.1. Theoretical Predictions 20
4.1.1. The Baseline Treatment 21
4.1.2. The UBI Treatment 23
5. Results 25 5.1. Effort Levels 26 5.2. Tax Revenues 32 6. Discussion 36 7. Literature 39 Appendix A 42 Appendix B 50
1. Introduction
An unconditional modest income for every citizen, given out by the government, with no strings attached. This idea might seem utopian, but has a widespread base of proponents. Great thinkers ranging from Montesquieu (1689-1755) to Thomas Paine (1737-1809) and Milton Friedman (1912-2006) all independently thought about and proposed some form of a Universal Basic Income (UBI). Although the idea was thought up under a variety of different names such as ‘territorial dividend’ and ‘state bonus’ to ‘demogrant’ ‘citizen’s wage’ and ‘universal benefit’ over the last two centuries, it never had success of being actually implemented (Van Parijs, 2001). The closest the idea ever came to being reality was in the late sixties and early seventies of last century, as it gained sudden popularity in both Canada and the United States as a way to eradicate rising levels of poverty. In the summer of ’69, the then just-sitting president of the United States –Richard Nixon-, proposed a bill which closely resembled a basic income, as it would guarantee families of four $1600 a year, equivalent to roughly $10.000 in 2016 (Bregman, 2014). And while the bill easily got through congress, it was foundered in the senate and about to be forgotten. However, the idea of an unconditional basic income (UBI) crossed the ocean, and has gradually become subject of a fast expanding public discussion in certain countries in the European Union in the last two decades. While opponents see the proposal as crazy and obnoxious, proponents claim it to be an effective remedy for a range of social problems, including unemployment and poverty. And although the debate on the subject is fierce, a lot is still remains unknown about the societal and economic effects of an Unconditional Basic Income (UBI) policy.
UBI policy would entail ‘giving all citizens a modest income, irrespective of any income sources and without requiring any present or past work performance’ (Van Parijs, 1992, p.1). As most western welfare states have social security systems that are based on means test and reciprocity1, implementing an UBI would entail significant restructuring of these systems. And while most critics argue that an UBI scheme can only be
implemented when huge financing costs and a possible decline of GDP are taken for
1 That is, welfare payments are only given to individuals whose income is below a certain level (i.e. means
test). To (keep) receiving these payments individuals often have to fulfil certain work and/or job seeking requirement (i.e. reciprocity).
granted, which in turn would entail higher taxation rates to finance the scheme (Gilroy, Heimann & Schopf, 2013), others disagree with the unconditional character per se since it would exploit the people who are willing to work and thus will enhance parasitic behaviour (Van Donselaar, 2009). That is, workers generate domestic income (through taxation) which is needed to be distributed unconditionally. If recipients (i.e. those who receive the UBI but do not work) are not held to a ‘reciprocal objection’ to help produce the income needed for their unconditional payments, they are exploiting the workers that do (Van Donselaar, 2009; Wilderquist, 2011). This reduces labour supply and general productivity and thus reduces the sustainability of the UBI scheme.
However, to what extent an UBI exactly induces this parasitic behaviour, or more generally influences workers’ their labour decisions, remains relatively unclear. In an attempt to answer this question, several large field experiments have been conducted in both the US and Canada between 1968 and 1980. These experiments were government sponsored experiments in which certain neighbourhoods, or in one case even a whole village, were given the right to an unconditional guaranteed income2. In general, a small decline in labour supply was found, but a debate persists whether this decline was higher or lower than beforehand expected (Wilderquist, 2004). Also, these experiments are often criticized for being performed in a highly politicized environment (Noguera & De Wispelaere, 2006), due to the fact that welfare is a highly charged political issue. Hence, the results might have been used to push a certain political agenda of welfare reform, rather than objectively used for studying the exact effects of such a reform. Nogueare & De Wispelaere therefore argue for the use of laboratory experiments when assessing UBI policy.
Because labour supply is highly related to taxation, as is the hypothetical funding of an UBI, this subject relates to a strain of literature on taxpayer behaviour in presence of financing public services. In general, previously conducted experiments find that workers supply more labour when some form of redistribution or insurance is in place (Ortona et
2 Most experiments made use of a Negative Income Tax (NIT) which is defined by Harvey (2006, p.2) as
‘a system of refundable tax credits that guarantees eligible tax filers a minimum income. Tax filers with no income receive the full benefit in cash. Persons with taxable income receive a cash benefit only to the extent their NIT benefit exceeds their tax liabilities.’ In general, a NIT and UBI are considered theoretically equal in achieving the same net transfer of income (Van Parijs, 2004)
al. 2008; Ottone & Ponzano, 2011; Keser et al., 2015). That is, individuals supply more labour when either taxes are directly redistributed, used for the funding of a public good, or used as an insurance against economic risk. However, in the case of an UBI, individuals receive an unconditional payment which is independent of their (individual) work effort, which has both a redistributive as an insurance purpose3. As no previous work has addressed the unconditional nature of an UBI in a laboratory experiment directly, this paper therefore makes use of a laboratory experiment to answer the following research question:
To what extent does a Universal Basic Income (UBI) affect labour supply and does labour supply react differently to taxation when an UBI is in place?
By answering this question, it expands the work of Ortona et al. 2008, Ottone & Ponzano, 2011, and Keser et al., 2015, by introducing the notion of unconditionality. That is, effects on labour supply under taxation can be studied when individuals receive unconditional payments, rather than when tax revenues are (directly) redistributed or used as insurance. In a more general fashion, results can be used to uncover behavioural insights on how individuals would react in terms of labour supply and taxation when receiving an unconditional income.
To summarize results of this paper, it is found that: a) labour supply decreases monotonically as tax rates increase. b) there are no significant differences in labour supply when an UBI is in place compared to when there is no UBI in place. c) Tax revenues increase monotonically up to a certain tax rate; after which they decline (i.e. existence is found of a so-called ‘Laffer Curve’). d) There are no significant differences in average tax revenues when an UBI is in place compared to when no UBI is in place. However, tax revenues are significantly higher in the UBI treatment when only tax rates above fifty percent are taken into consideration.
The paper is setup as follows: In the next section, an overview of existing literature on Basic Income and taxpayer behaviour is given. Thereafter, the experimental design and
3 An UBI is redistributive in nature because the unconditional payment has a greater marginal effect on
individuals with a relatively low income (i.e. supply less labour) compared to individuals with a higher income. It has an insurance nature because even when individuals are unable to work due to economic risk (i.e. involuntary unemployment) they can fall back on this UBI.
procedures are described. Section 3 presents the hypotheses. In Section 4 results of the experiment are discussed. Finally, section 5 discusses the main findings and concludes the paper.
2. Previous literature
2.1. Universal Basic Income
After the most recent economic crisis, governments of most western countries have chosen to cut government spending. Most often by reducing public expenditure on welfare or by narrowing the eligibility criteria of already existing benefit systems. This policy line fits in the overall tendency of only slightly modifying welfare programmes in the last few decades in the western welfare state, despite the rapidly changing economy and labour markets (Koistinen & Perkiö, 2014). In contrast to this, the idea to introduce a fundamental social security reform in the form of a Universal Basic income (UBI) has become more persistent. From both ends of the political spectrum and in different socio-economic context, several initiatives for an UBI, or a welfare scheme that closely resembles an UBI, have been made in recent years. And while in no country the idea succeeded in becoming reality, proponents are getting more vocal and the idea is increasingly gaining in mainstream attention. Resulting in a more prominent position of the idea on the political agenda in countries where the idea was already known and the spreading of the idea to countries where it was previously unknown.
So what exactly is an UBI? The leading definition, formulated by one of the most vocal proponents of an UBI, Philippe Van Parijs, is as follows: “A basic income is an income unconditionally paid to all its members, without means test or work requirement. In other words, it is a form of minimum income guarantee that differs from those that now exist in various European countries through its being paid (1) to individuals rather than households; (2) irrespective of any income from other sources; and (3) without requiring any present or past work performance nor the willingness to accept a job if offered” (Van Parijs, 1992, p.1). Hence, an UBI ‘is granted by the virtue of an
unconditional entitlement’, and by the idea that ‘any income from other sources will come on top of the basis it provides’ (p.2).
In short, proponents argue that guaranteeing an income for all would eradicate poverty and reduce unemployment (Van Parijs, 2004; Standing, 2008), and emphasise the simplicity and transparency of a welfare system with an UBI, as it would reduce
reciprocal requirements of recipients (Van Parijs, 2001). Arguments of critics of a Basic Income are most often built around the premise of the high financing costs of such a system (as each citizen, rich or poor, would receive the unconditional payment from the government). Others expect a decline in overall productivity when an UBI is in place, as well as higher taxation rates necessary to finance the Basic Income (Hauser, 2006; as cited in Gilroy, Heimann & Schopf, 2013). To explain the drop in overall productivity, critics often refer to the classic economic theory of the homo economicus, in which it is assumed that work is considered unpleasant but necessary to provide a certain level of consumption. That is: work incentives come directly from the monetary transfers related to work and work itself is considered to be unpleasant, whereas leisure is something to be desired (Gilroy, Heimann & Schopf, 2013). When a Universal Basic Income is in place at the height of at least subsistence level, this would entail that some individuals would see this UBI as sufficient for their (basic) needs, and therefore might choose to work less or choose to not work at all. Unemployed individuals might refrain from seeking a job or employed individuals might even withdraw from the labour market. Hauser therefore expects a sharp decline in labour supply, especially among groups such as women
looking after children, elder employees, long-term unemployed, and possibly even young entrants of the labour market. He also argues that the increase in tax rates necessary to finance the UBI could induce employees to leave the labour market as it would decrease the marginal income (and thus utility) related to work. These factors would significantly drop overall GDP levels.
As the idea of an UBI is not new, several large scale experiments have been conducted to research economic and societal effects of a guaranteed income. Both the United States government and the Canadian government sponsored several large-scale guaranteed income experiments between 1968 and 1980. Collectively these experiments are known as the income maintenance experiments or the guaranteed income experiments in the literature (Wilderquist, 2004). In all experiments randomly selected recipients received some form of guaranteed income at the height of at least poverty line level. These experiments were setup to get an indication of the costs and feasibility of an UBI. Researchers were therefore mainly interested in in the effects of a guaranteed income on the work effort of the recipients. And while these experiments have been discussed in at
least 340 scholarly articles, opinions on how the outcomes of these experiments should be interpreted vastly differ, according to Wilderquist. Both proponents and opponents of an UBI have used the results of these experiments in support of their arguments. This disagreement is the result of the most general finding of all experiments: the treatment group worked less than the control group, all else equal. Which is what everyone
beforehand expected. However, the question of how much less, and whether it was more or less than expected beforehand, is the one up for debate. As no objective criteria were set for how much reduction in work effort would affect the feasibility of the guaranteed income scheme, observers can see what they want to see when looking at the results. Wilderquist therefore fittingly describes these experiments as ‘scientific Rorscharch tests’. Also, comparing the results of the five experiments is especially difficult as they differed in the levels of the guaranteed income (relative to the poverty line level) and socio-economic groups and regions.
These above mentioned field experiments are often criticized. Noguera & De Wispelaere (2006) argue that there are two distinct reasons why one might approach (the results of) these experiments with scepticism. First, field experiments that study social security are more susceptible to ‘political manipulation’, defined as ‘external interference with the research process or its outcomes for political reasons’ (p.2). Meaning that the setup of the experiment and/or the analysis of the results might be specifically designed to prove a certain political point. Second, they don’t see these field experiments fit to deliver a genuine understanding of the behavioural effects of an UBI. Meaning that while field experiments can be useful, they face considerable constraints that affect both the scope of the research (i.e. the range of questions that can be studied in a single
experiment) and the validity and robustness of the findings. They thus argue for the use of a laboratory experiment setting for researching UBI policy, as it offers a
‘comparatively closed environment’ in which all crucial stages of the research can be ensured. Also, by using laboratory experiments, a higher level of control in the research design is possible, which paves way to ‘study behavioural responses in relation to specific aspects of the labour market, social security, the welfare system, and so on’ (Noguera & De Wispelaere, 2006, p.4).
To my knowledge no such experiments in relation to UBI have been previously performed. This paper therefore proposes such an experiment. More specifically, the experiment is built around the question how and UBI affects individuals’ labour supply and if and how individuals react differently to taxation of their income when an UBI is in place. This way, it is also possible to examine the effects of higher taxation rates on labour supply when an UBI is in place. It can therefore, to a certain degree, test the arguments of opponents of an UBI, as described above, that an UBI would entail lower productivity and higher taxation needed to finance the scheme. Because the notion of taxation is included, this paper relates to a strain of literature on labour supply and taxpayer behaviour. An overview of the literature on this subject is given in the next section.
2.2. Labour supply and taxpayer behaviour
Even though there is no clear consensus on how an UBI should be funded –some propose increased natural resource taxes while others argue that new tax instruments such as ‘Tobin taxes’ on (speculative) capital movements should do the trick-, there is assumed for simplicity in this paper that an UBI will be paid directly through taxes on income. This relationship, between labour supply and taxation, has been researched quite extensively (Agranov & Palfrey, 2014; Gamage, Hayshi & Nakamura, 2010; Keser, Masclet & Montmarquette, 2015; Lindbeck, 1982; Ortona et al., 2008; Schroder et al., 2013; Swenson, 1988). This section discusses the most important findings.
Swenson was one of the first to experimentally test how individuals react to taxation in a laboratory setting. He argued that taxes have a disincentive effect. In other words, taxes ‘discourage individuals from earning income’. This means that as income tax rates increase, work disincentives become large and their tax payments decline. That is, the higher the tax, the less inclined individuals are to work, resulting in less tax paid. This relationship often takes an inverted U-shape (with tax rate on the x-axis and tax payments on the y-axis) and is known as the ‘Laffer Curve’. The basic idea behind this Laffer Curve is that there exists an optimal point of taxation (from a government perspective) that maximizes tax revenue. Swenson tested these propositions in a laboratory experiment in which subjects had to ‘work’ by pressing two keys on the
keyboard of a computer and then hitting the return key. The higher the number of hits, the higher the cash reward. Swenson varied the tax percentage level and the level of free income (i.e. tax redistribution) between treatments. Subjects could choose to ‘work’ or actively participate in leisure, provided in the form of magazines, a computer game, or a card game. The main results of the experiment are that tax redistribution, preferences for leisure versus work, and after-tax income needs all impact individuals’ their earning incentives and thus their tax payments. This effect of the tax payments is translated to the shape of the Laffer Curve. In both situations, either when tax revenues were redistributed or was ‘burnt’, evidence was found for the existence of such a Laffer Curve, and tax revenues were optimized at nearly the same tax rate. However, when relatively high tax rates were in place, output levels saw a smaller decline when taxes were not redistributed compared to the situation where redistributive payments were made.
Keser et al. (2015) tested this premise in greater detail. They ran an experiment to test whether the use of the tax revenue influenced labour supply decisions. Settings were compared in which the tax revenue was not used, redistributed, or used to fund a public good (i.e. given to a charity). Keser et al. used a slightly different experimental setting than Swenson. Subjects had to decode letters into numbers, and would receive a cash reward for each decoded letter. Subjects were presented with all possible tax rates
(ranging between 0 and 100 percent) and had to choose the number of letters they wished to decode for each of the tax rates4. By using this so-called strategy method, Keser et al. were able to determine labour supply curves for each individual. They found significant differences in both labour supply and tax revenues under each of the settings. More specifically, they found that labour supply is significantly higher when taxes are either directly redistributed or used to provide a public good, rather than ‘destroyed’.
As social security policy has two main purposes –insurance and redistribution-, it appeals to two different notions of ‘fairness’ and thus behaviour. On the one hand, the redistribution of income makes society ‘fairer’ by reducing wealth differences in society. On the other hand, social security also makes society ‘fairer’ by strengthening the link between effort and the outcome of this effort by reducing the effect of random events that can drastically impact income (Esary, Salmon & Barrileaux, 2012).
Meaning that social security acts as an insurance to guarantee a certain level of income under unforeseen circumstances such as involuntary unemployment. Ortana et al. (2008) and Ottone & Ponzano (2011) take slightly different approach in their experiments to research the behaviour of taxpayers. In both papers the effects of taxation on labour supply is examined when tax revenues are used to reduce risks involved with exerted effort. That is, subjects had to throw dice to determine how much of the income earned by performing a certain task was theirs to keep. The outcome of the throw determined whether they could keep their full earned income or would lose either half or the full amount. Between treatments, the height of these insurance payments was differed. They find that labour supply increases as long as there is a ‘welfare state’ in place which (partly) accounts for the risks, paid through taxes.
However, none of these papers have included a treatment in which there are
unconditional payments given to the subjects. In all cases, height of the redistributive
payments depended on overall tax revenue (Keser et al. 2015), or tax revenues were used as insurance (Ortana et al. 2008; Ottone & Ponzano, 2011). Hence, no previous experiment has tried to research the effect of an UBI on labour supply under taxation in the laboratory. Because the idea of an UBI is not new because of its protection against economic risks (i.e. most western countries have social security programmes in place that help workers maintaining a certain income when becoming unemployed), but rather because of its
unconditional redistributive purpose, the experimental design in this paper is built around
this premise. There is however accounted for economic risks in the design. By using this research design, results can be used to test several things. First, it can be observed whether individuals reduce their supply of labour when receiving an unconditional guaranteed payment. Second, it can be tested how labour supply reacts to certain increases in tax rates when there are (no) guaranteed payments. Third, results can show how optimal tax rates, from a government perspective, differ.
In general, the research question of this paper is closely related to the previous research by Keser et al. (2015), Ottone & Ponzano (2011), and Ortana et al (2008), but adds the notion of unconditionally and economic risk to the studies. Because the experimental design by Keser et al. makes clever use of the strategy method to retrieve a relatively large amount of data for each subject, and because (financial) resources for this
paper are relatively scarce, a variation of this experimental design is used for this study to ensure a relatively large dataset for the resources available. Consequently, the way the hypotheses are structured and the structure of the analysis of the results of the experiment take a similar approach. This paper can thus be seen as a direct addition to the experimental work by Keser et al., by applying the methods to the case of an UBI. By using a similar structure, some of the more trivial results of this experiment (e.g. the general effect of an increase in taxation of labour supply) can be, to some degree, compared to the results by Keser et al. When these more trivial results are similar, one could argue that the less trivial results (e.g. the effect of the unconditional payments) can be considered more robust. The next section discusses the full experimental design in greater detail.
3. Experimental procedure
The experiment used is a real effort experiment. At the beginning of the experiment, each participant has to decide how much effort he/she is willing to invest for each possible tax rate t ranging from zero to 100 percent in discrete five-percent steps. By using this so called ‘strategy method’, data on effort levels, contingent on all possible tax rates, for each participant will be available. Because participants can quit the experiment sooner if they decide to work less, the experiment provides participants with a real trade-off between work and leisure.
The effort tasks consist of a decoding task, which is used first by Levy-Garboua et al. (2009), later by Charness et al. (2013) and Keser et al. (2015). Especially the experimental procedure described by Keser et al. is used as an inspiration to construct a fitting experiment to answer the research questions proposed in this paper. The main task of the experiment consists of decoding letters into numbers from a table with letters in one column and corresponding numbers in another, both displayed on the computer screen. Full instructions and an example of the decoding task is given in Appendix A. According to the previous mentioned papers, this task is considered boring, and is chosen to induce sufficient disutility of effort. About 12 letters can be decoded per minute.
Each decoded letter pays €0,02 and maximum effort is solving 540 decoding problems. Exerting maximum effort thus generates an income of €10,80 before taxes. Prior to performing the decoding task, participants are informed that they have to pay taxes for each decoded letter. These taxes are automatically deduced from a participant’s earning. After participants have made decisions on how many letters they wish to decode for each of the 21 potential tax rates, one of the tax rates will be randomly chosen to be actually implemented. Subjects are then requested to perform their intended effort for that tax rate. The base payment for completing the decoding task for each individual is determined by:
Subjects cannot quit the experiment until their intended tasks are completed, otherwise they will not get paid. When they have completed their tasks, subjects are asked to throw two virtual dice: if the sum of these dice is 2, they will lose their income earned by the decoding tasks. This corresponds with a 2,78% (or 9:8) probability and is known by the participants. This risks corresponds with a general ‘economic risk’ and is used so taxes in are not only used for redistributive purposes, but also act as a sort of insurance.
The experiment will consist of two different treatments that differ in the (perceived) use of the tax revenue. The first treatment is the Baseline treatment. Subjects are told that the tax revenue will be used for one main purpose: insurance of the unlucky participants. In this treatment, participants who are unlucky receive a compensation payment of 𝐴 = €1,50. Participants are also told that if more taxes are generated than the maximum possible insurance to be paid out by the government (i.e. 𝑛 ∗ 𝐴 with 𝑛 being the number of subjects) these will be given to a ‘public good’ in the form of a charity. There is chosen to give excess taxes to a charity to give subjects the feeling they are still paying taxes when taxes exceed the amount needed for redistribution. The charity chosen is the Environmental Defence Fund. This tax revenue threshold corresponds to the ‘government budget’ that is available to insure people affected by the economic risk.
The second treatment is the ‘UBI treatment’, where each participant receives at least a guaranteed payment of 𝐴 irrespective of their labour supply and their outcome after the throwing of the dice. Similar to the baseline treatment, there is a government budget of 𝑛 ∗ 𝐴 and tax revenue surplus will be given to the charity. Subjects who are unlucky while throwing the dice will fall back to this unconditional payment 𝐴. Thus, the difference between treatments is the unconditional nature of the payment A. Where in the Baseline treatment only unlucky players receive this payment, in the UBI treatment all subjects, irrespective of their intended effort, receive this payment, thus increasing their initial payment. Hence, for the baseline treatment expected payment for each individual is:
35
36 1 − 𝑡𝑎𝑥 𝑟𝑎𝑡𝑒 ∗ €0,02 ∗ 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑐𝑜𝑑𝑒𝑑 𝑙𝑒𝑡𝑡𝑒𝑟𝑠 + 1
36∗ €1,50
While for the UBI treatment it is:
35
36 1 − 𝑡𝑎𝑥 𝑟𝑎𝑡𝑒 ∗ €0,02 ∗ 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑐𝑜𝑑𝑒𝑑 𝑙𝑒𝑡𝑡𝑒𝑟𝑠 + €1,50 + 1
Because there is risk and inequity involved in the decision-making, participants will be asked to make choices in two games before the main experiment. First, in a modified dictator game which tests for inequity aversion (introduced by Blanco et al. (2010)), participants are given a list of 21 pairs of payoff vectors and have to choose one of the two payoff vectors in all 21 cases. The left payoff vector is always (€0,80, €0), which gives the dictator €0,80 and the recipient nothing and thus can be seen as the ‘selfish choice’. The right payoff vector contains equal payoffs varying from (€0, €0) to (€0,80, €0,80) in discrete 4 cent steps and thus can be seen as the ‘unselfish choice’. Hence, participants first have to choose between €0,80 for themselves and zero for the other or zero for both. Then between €0,80 for themselves and zero for the other or €0,04 for both. Thereafter between €0,80 for themselves and zero for the other or €0,08 for both, etcetera. This continues up to the choice in which participants can choose between €0,80 for themselves or €0,80 for both. Each participant decides as a dictator and determines the payoff of one of the other participants. One of the 21 payoff vectors is randomly chosen for actual payment. Inequality aversion corresponds to the number of equal choices out of the 21 pairs of payoff vectors (i.e. the number of times respondents chose the ‘unselfish choice’ over ‘selfish choice’). Second, the Holt & Laury (2002) test for risk aversion will be used to be able to control for risk aversion, where subjects have to make choices in 10 different lottery choices. The number of safe choices corresponds with the level of risk aversion. see Appendix A for the instructions of these tests.
The experiment was conducted online, coded using the PHP language and distributed through online social networks and message boards5. All subjects were either
of the Dutch Nationality or currently residing in the Netherlands and were aged 16 to 61. Because the experiment was web-based, subjects were able to perform the experiment at home, behind their own computer, during any spare time they had. Thus, one could argue that using this setting, compared to an experiment in an official laboratory, subjects make a more direct trade-off between leisure and work (i.e. tasks performed in the experiment). In a traditional setting, where subjects are asked to perform tasks in an economic laboratory, subjects are often recruited with an indication of how long the experiment will last and how much (approximately) can be earned. Hence, if subjects arrive at the
experiment location they might have already taken into account that they will not use the upcoming X hours for leisure. In the words of Carpenter et al. (2006), subjects ‘come to play’: they decided in advance the time to be devoted to the experiment, and behave accordingly. Also, when they quit the experiment they are still located at the experiment location. As they are likely to need some time to travel to get home, there are some ‘fixed costs’ in their leisure equations. While when subjects are at home while conducting the experiment, they can immediately continue with their leisure after they finish the experiment. This makes the trade-off between leisure and work more direct.
After the experiment, participants will be asked to fill in a questionnaire, in which, besides general demographics, will be asked for their political orientation; their level of trust; and if subjects agree with several statements regarding welfare, societal responsibilities, paying taxes, and an UBI.
In total, 47 subjects completed the experiment. In total 5 subjects were dropped from further analysis for one of the following reasons: inconsistent preferences in either the risk or inequity aversion tests and/or not (completely) answering the questionnaire after the experiment. As the randomly chosen tax rate was 70 percent, subjects were paid, on average, €2,09. Including the payments in both the inequity and risk aversion tests.
4. Hypotheses
To formulate a hypothesis on how unconditional payments affect labour supply under taxation, a model with non-separative utility functions will be used. As the experimental procedure has a similar structure to the experimental procedure used by Keser et al. (2015), an adapted version of one of the two models proposed in that paper will be used. The version of the model presented here adds a level of economic risk (given by parameter 𝑝) and differs non-labour payments 𝐴 between treatments. This way it allows to investigate the effects of initial wealth (i.e. an UBI) and redistribution on labour supply when a level of economic risk is included. For purpose of readability, the full model is given in this section.
In general, the model predicts that income taxes are perceived as disincentive by individuals, since workers only care about their net wage after taxes. As a higher tax rate corresponds to a lower net wage, individuals are less willing to work when taxes increase. However, this effect may be partly offset because if net wages are low, an individual has to work more to maintain a certain amount of income. In other words: to satisfy the constraint of his or her budget line, one has to work more when net wages are low. Hence, there is a possibility that this ‘positive income effect’ may cause an increase in labour supply (Keser et al., 2015). In the particular case of an UBI the budget constrain is directly affected, as individuals receive a payment without work requirement that (partly) satisfies their budget constraint. The positive income effect might thus be offset due to the UBI. Hence, there is expected that an UBI significantly affects labour supply. The full model is explained in greater detail below.
4.1. Theoretical Predictions
Preferences are represented by a Cobb-Douglas utility function first formulated by Keser et al. (2015). The utility function represents the trade-off individuals make between consumption 𝐶 and leisure 𝐿 such that:
𝑈 𝐶, 𝐿 = 𝑈 𝐶, 𝑇 − 𝐿 = 𝑈 𝐶, 𝑒G = 𝐶 H 𝛾 − 𝑒
G 8JH (1)
Where 𝑇 is total time available such that working effort is given by: 𝑒 = 𝑇 − 𝐿. With L representing leisure. The trade-off between leisure and consumption is captured by parameter 𝛼.
The maximum available tasks per worker are given by 𝛾. Each worker has a standard budget constraint given by 𝐶 = 𝑒G𝑤 1 − 𝑡 + 𝐴8,M. Where consumption 𝐶 depends on the workers’ net wage after taxes 𝑤(1 − 𝑡) (with 𝑤 representing the wage and 𝑡 the tax rate) times work effort 𝑒G plus the exogenous unearned income 𝐴8,M, with 𝐴8representing the unearned income for the baseline treatment and 𝐴M for the UBI treatment. For this experiment 𝐴8 = 0 and 𝐴M = 𝐴. Insurance in the Baseline treatment is also equal to 𝐴 in this experiment.
There is chosen to let income/insurance 𝐴 to be exogenous because in reality the private marginal return from work effort on redistributive/insurance payments is rather low because a state is composed of millions of inhabitants. Also, in reality, government deficits are possible when collected taxes fall short. Making it possible to still make redistributive and insurance payments in the short term. As said above, this experiment consists of two treatments which differ in the conditionality of redistributive payments. For the Baseline treatment (BT) the model is constituted as follows.
4.1.1. The Baseline Treatment
Consider first the Baseline treatment where participants do not receive any unearned income, but are eligible for an insurance payment of A when they belong to the unlucky participants that lose their income due to chance.
The utility function for each worker is given by substituting 𝐶 into equation (1), which gives:
𝑈(𝐶, 𝑒G) = OP(1 − p)𝑒G𝑤(1 − 𝑡) + 𝑝𝐴RSH∗ (𝛾 − 𝑒G )8JH (2)
with 𝐴 the insurance amount transferred to the unlucky participants who were randomly selected for losing their total income earned in the decoding tasks, which happens with probability 𝑝.
Alternatively, using a monotonic transformation, and some mathematical simplifications, the Cobb-Douglas utility function can also be represented as follows:
𝑈 𝐶, 𝑒G = 𝛼 log (1 − p)𝑒G𝑤(1 − 𝑡) + 𝑝𝐴 + 1 −a log g − 𝑒G (3)
The optimal effort requires that each worker maximizes his or her utility given by equation (3) over 𝑒G. This gives the maximization problem:
max
ZG 𝛼 log (1 − p)𝑒G𝑤(1 − 𝑡) + 𝑝𝐴 + 1 −a log g − 𝑒G (4)
The following first-order condition defines the optimal work effort choice of each worker
i 𝛼 1 − 𝑝 𝑤 1 − 𝑡 1 − 𝑝 𝑒G𝑤 1 − 𝑡 + 𝑝𝐴+ 1 − 𝛼 𝛾 − 𝑒G = 0 (5)
This first order condition can be solved to obtain the optimal labour supply for the baseline treatment 𝑒[\∗ such that:
𝑒G[\∗ = 𝛼𝛾 1 − 𝑝 𝑤 1 − 𝑡 − 1 − 𝛼 𝑝𝐴 1 − 𝑝 𝑤 1 − 𝑡 (6) which simplifies to 𝑒G[\∗ = 𝛼𝛾 − 1 − 𝛼 𝑝𝐴 1 − 𝑝 𝑤 1 − 𝑡 (7)
From (7) it can be observed that since all parameters are positive 𝑒G[\∗ is increasing in wages 𝑤 and decreasing in taxes 𝑡. With regards to non-labour payments, or insurance payments in this case, there is observed that optimal effort decreases as these increase. As the experiment is designed to test how labour decisions are influenced by tax rates and non-labour payments (i.e. the unconditional payments
as part of the UBI), the following testable hypothesis (complying with the first hypothesis from Keser et al.) is formulated:
Hypothesis 1: Effort decreases as taxes 𝑡 increase 4.1.2. The UBI Treatment
Now consider the case where all participants receive non-labour payment 𝐴 independent of the outcome when throwing the two (virtual) dice. Thus, participants who are ‘lucky’ receive the payment earned through labour plus the non-labour payment 𝐴 and participants who are unlucky receive only 𝐴. The utility function and budget constraint are now given by:
𝑈 𝐶, 𝑒G = 𝐶 H 𝛾 − 𝑒
G 8JH (8)
With the constraint:
𝐶 = 1 − 𝑝 𝑒G𝑤 1 − 𝑡 + 𝐴 + 𝑝𝐴 (9)
Substituting (9) into (8), some minor simplifications and using a monotonic transformation yields the following maximization problem for each worker 𝑖.
max
ZG 𝛼 log 1 − 𝑝 𝑒G𝑤 1 − 𝑡 + 𝐴 + 1 − 𝛼 log [𝛾 − 𝑒G] (10)
The first order condition of (10) yields the optimal labour supply 𝑒G`[G: 𝑒G`[G∗ = 𝛼𝛾 − 1 − 𝛼 𝐴
1 − 𝑝 𝑤 1 − 𝑡
(11)
The comparison between optimal effort levels in the Baseline and the UBI treatment indicates that effort levels in the UBI treatment are lower as long as probability 𝑝 takes a non-zero value. This indicates that when individuals receive additional unconditional income it will make them less inclined to work, as their needs are satisfied when exerting less effort. From this, the following hypothesis is derived:
Hypothesis 2: Expected effort levels are higher in the Baseline treatment compared to the
UBI treatment.
From a behavioural perspective, this model cannot say much as it is assumed that all individuals are identical. However, in reality, there exists heterogeneity between individuals in both behaviour and preferences. Attitudes towards taxation might differ from person to person. Some individuals might be intrinsically motivated to pay taxes (which is sometimes termed as ‘tax moral’) (Keser et al. 2015), while other might be highly ‘tax-averse’. It is often shown that these tax morals are related to some demographics including age, gender, marital status as well as social preferences (Clotfelter, 1983). Also, how an individual values redistribution is likely to be heterogeneous (Alm et al. 1992). Individuals who value government spending highly might be more willing to pay taxes than individuals who attain less value to government spending. There might also be significant differences between subjects in risk and inequity-aversion. Subjects that are risk-averse and/or inequity averse might be more willing to pay the proposed taxes to guarantee a certain income for themselves and/or others. Based on these studied and the formal testing of these propositions by Keser et al. (2015) it is expected to observe heterogeneity in attitudes towards taxation, redistribution and risk across individuals. Also, individuals with higher aversion towards risk and/or inequity are expected more likely to comply with taxation in when an UBI is in place, as this reduces both risk and inequity.
5. Results
As the first part of the analysis, potential differences are investigated between treatments in both demographics and answers given to the post-experimental questionnaire. Table 1 shows sample characteristics of the participants in both treatments. It is observed that there are no significant differences between treatment with regards to characteristics of subjects. Both treatments consist of 21 participants of which gender is relatively equally distributed. The average age of respondents is approximately thirty and slightly higher in the UBI treatment. Around fifty percent of respondents think that, in general, most people can be trusted. Subjects in the Baseline treatment are slightly more inequality averse with, on average, 14,33 safe choices (out of 21) in the inequality aversion test versus 12,95 in the UBI treatment. Subjects are considered to be somewhat risk averse with on average 6,67 and 6,05 safe choices out of 10 in the Baseline treatment and UBI treatment, respectively. It is also observed that respondents have a relatively high tax moral (i.e. they agree that paying taxes is one’s responsibility towards society) and have answered positively to the statement whether it is one’s responsibility towards society to be professionally active (professionally active moral). These results thus suggest that individuals were sufficiently randomly distributed across treatments, and that potential differences in behaviour between treatments cannot be directly accounted to differences in socio-demographics and/or answers to the questionnaire.
Table 1. Means and standard deviations in socio-demographics
Baseline Treatment UBI Treatment
Socio-demographics N 21 21 Male 66,67% 52,38% Age 29,62 29,76 (9,46) (9,15) Political preference 4,76 4,95 (2,02) (2,28) Trust 66,67% 71,43% Inequality aversion 14,33 12,95 (6,21) (5,87) Risk aversion 6,67 6,05 (1,59) (1,97) Tax moral 4,43 4,33 (0,73) (0,84) Work moral 2,19 1,95 (1,22) (1)
Professionally active moral 4,05 (1,00)
3,90 (1,38)
Notes: no significant differences found between treatments
Political preference corresponds to the answer to the following question “What is your politicalpreference?” 1 = left 10 = right. Trust corresponds to the answer to the question “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” 0 – “yes most people can be trusted” 1= “No, you can never be too careful in dealing with people. Inequality aversion corresponds to the number of equal choices out of 21 pairs of payoff vectors. Risk aversion corresponds to the number of safe choices a respondent made. The statements for Tax moral, work moral and professionally active moral were “Paying taxes is one’s responsibility towards society”, “Receiving money without working for it is humiliating” and “Being professionally active is one’s responsibility towards society”, respectively. Respondents could answer with 1= strongly disagree 2=somewhat disagree 3=neutral 4=somewhat agree 5=strongly agree.
5.1. Effort Levels
Table 2 shows summarizing statistics (i.e. average effort, and standard deviation of effort) of the chosen effort levels per tax rate between treatments. It is observed that average effort levels between treatments are relatively similar and that effort levels decrease as the tax rate increases. The average effort level for all tax rates is 93,99 in the Baseline treatment and 110,21 in the UBI treatment. At the zero percent tax rate effort is slightly higher in the UBI treatment. However, there are almost no differences in effort levels between treatments in the 25 to 55 percent tax range. Figure 1 gives a visual representation of these effort levels.
Table 2. Summary statistics of effort levels per treatment
Baseline Treatment UBI Treatment
Tax rate Mean Std. dev Mean Std. dev
0% 216,90 166,92 260,48 220,90 5% 209,23 166,68 248,29 214,08 10% 198,85 166,08 220,43 197,72 15% 187,71 160,43 209,62 193,94 20% 181,76 159,44 196,76 185,03 25% 169,76 150,45 167,14 167,92 30% 162,19 148,24 147,57 157,54 35% 123,38 120,42 122,48 137,92 40% 119,57 120,30 108,81 128,40 45% 98,61 108,96 100,86 126,04 50% 81,10 90,21 86,57 122,50 55% 61,10 81,09 65,95 117,16 60% 40,62 42,39 61,95 117,06 65% 29.,62 29.,54 56,43 116,68 70% 24,05 24,17 52,76 116,96 75% 19,76 20,83 50,14 117,47 80% 14,76 16,92 42 116,60 85% 12,38 15,30 41,05 116,84 90% 8,29 11,88 35,86 117,03 95% 7 11,38 35 117,21 100% 7,19 11,94 4,33 7,74 Total 93,99 129,18 110,21 161,93
Note: Effort levels correspond to the number of intended tasks in the experiment. N=21 for the Baseline Treatment and N=21 for the UBI treatment
It is observed that, consistent with Hypothesis 1, in both treatments effort levels decrease with rising tax rates. This is consistent with previous findings from Swenson (1988), Sutter & Weck-Hanneman (2003) and Keser et al. (2015). Average effort is slightly higher in the UBI treatment when tax rates are below 20 percent and above 55 percent. This is surprising given the theoretical predictions that state that effort is higher in the Baseline treatment. Hence, this indicates that no support is found for Hypothesis 2 when looking at average effort levels.
Figure 1: Results of average effort levels per treatment
Consistent with the behavioural expectations heterogeneity in attitudes toward taxation across participants are observed (see Figures a, Figure b and Table c in Appendix B). Appendix B shows that most participants have monotonically decreasing effort as tax rates increase (81,82 and 72,73 percent of individuals in the baseline treatment and UBI treatment, respectively). A small part of respondents exerts constant effort and then decrease sharply their effort until zero effort after a certain tax rate (4,55 percent in the baseline treatment and 13,64 percent in the UBI treatment). 9,09 percent of individuals in the UBI treatment chose to exert no effort unconditionally to the tax rate. In the Baseline treatment none of the respondents chose to exert no effort. Also, some respondents made ‘inconsistent’ effort choices. With inconsistent meaning that they chose a higher effort level at a certain tax rate compared to lower tax rates (i.e. effort is not decreasing monotonically, for instance choosing a low effort at tax rate t, then a higher effort at tax rate t+1 and again a lower tax rate for tax rate t+2). In total 13,64 percent of respondents in the baseline treatment exerted inconsistent effort versus 4,55 percent in the UBI treatment.
These individual effort levels seem to indicate a few things, which comply to findings from other research who have used this decoding task (e.g. Charness et al. (2013) and Keser et al. (2015)). Firstly, some participants chose to exert zero effort. This seems to indicate that The participants found the task ‘sufficiently boring’ so that it incurred a high enough level of disutility for some to not exert effort. Second, because some individuals chose to exert maximum effort it can be said that the task was considered feasible. When looking at the effort levels at the 0-percent tax rate, it is observed that participants are, on average, not willing to exert maximum effort. With the average effort at this tax rate being 216,90 in the Baseline treatment and 260,48 in the UBI treatment. This indicates that participants made a trade-off between work and leisure, even at the 0-percent level, with an opportunity cost in terms of leisure.
When compared to previous research which used a similar task, several similarities and differences can be found. Frist, as stated above, effort decreases (monotonically) as tax rates increase as observed in all previous work. Second, general attitudes towards the task (e.g. some exert zero effort, some maximum effort, no average maximum effort at the 0-percent level, etc.) also comply with previous work. This indicates that the task was performed correctly and induced the expected incentives. Third, heterogeneity in attitudes towards taxation is also found in this experiment. However, when the results of the Baseline treatment of this experiment compared to the Baseline treatment performed by Keser et al. (2015)6, effort levels are considerably lower in this experiment. The average effort level in the Baseline treatment in Keser et al. was approximately 292,32, which is more than three times the average observed in the Baseline treatment in this experiment, which is 93,99. The difference in exerted effort can be constituted to one or more of the following reasons: First, payments were significantly lower in this experiment (€0,02 per decoded letter vs. €0,04 per decoded letter). Second, there are differences in the experimental design. In this experimental design a certain level of economic risk is included (i.e. the probability that a participant could lose his/her earned income and would only get a certain amount of ‘insurance’), this thus lowers the expected return of effort. Whereas in Keser et al. there was no economic risk nor insurance payments. Third, as
explained on p.18 above, this experiment was conducted online resulting in a more direct trade-off between work and leisure, compared to the physical laboratory setting in which the experiment by Keser et al. was conducted. Fourth, participants in this experiment were all residing in The Netherlands, while the participants by Keser et al. came from several countries. As Keser et al. found significant differences in effort levels between countries, this difference in nationality could constitute (for a small part) in the difference in observed effort levels, although it is unlikely that this explains the full difference.
Going back to the results of this experiment, in contrast to Hypothesis 2 the average effort level considering all tax rates is higher in the UBI treatment (110,21; SD: 161,93) than in the baseline treatment (93,99; SD: 129,18). However, a Wilcoxon-Mann-Withney test indicates that the difference between treatments is statistically insignificant (p=0.3839; two-sided). Hence, it cannot be concluded that average effort levels differ between treatment. These findings are summarized in Result 1.
Result 1. a) In both treatments, effort levels decrease (on average) monotonically with
rising tax rates. b) Across individuals there exists important heterogeneity in the attitudes towards taxation. c) Overall, there is no significant difference in average effort levels between treatments.
Table 3 shows regression estimates on the determinants of effort levels in support of Result 1. Random effects Tobits are used because of the data has a panel data structure and because multiple subjects chose to exert maximum effort (i.e. data is ‘censored’ at 540 units of effort). Because it is observed that average effort is higher in the UBI treatment when the tax rate is above 50 percent, regression results conditional on these tax rates are also included to test whether the treatment has a significant effect at these tax rates.
Table 3. Determinants of effort levels
All tax rates Random Effects Tobit
All tax rates Random Effects Tobit Tax Rate > 50% Random Effects Tobit Tax rate > 50% Random Effects Tobit Variable (1) (2) (3) (4) Tax Rate % -2,44*** (0,09) -2,44*** (0,09) -1,07*** (0,12) -1,06*** (0,12) Treatment 16,22 (29,05) 29,54 (27,55) 22,07 (22,65) 26,79 (22,09) Male 48,63 (30,14) 38,05 (24,16) Age 0,63 (1,60) 0,18 (1,28) Political preference -7,69 (8,54) -8,77 (6,85) Trust 30,40 (36,66) 4,03 (29,40) Inequality aversion 0,47 (2,96) -1,75 (2,37) Risk aversion 5,83 (10,73) -1,17 (8,61) Tax moral -24,94 (19,53) 12,20 (15,67) Work moral 4,45 (14,49) -0,59 (11,62) Professionally active moral 24,57*
(14,12) 15,73 (11,32) Constant 213,43*** (21,01) 137,21 (108,87) 104,98*** (18,39) 142,66 (87,70) Observations 882 882 420 420 Cens. observations 0 0 0 0 Log. likelihood -5448,49 -5206,19 -2161,08 -2158,15
Notes: standard errors in parentheses *** Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level
The 882 observations in columns (1) and (2) corresponds to the 42 respondents observed for 21 possible tax rates. For column (4) there were only 10 observed tax rates.Treatment variable: 0=baseline treatment 1=UBI treatment. Tax rate % variable ranges from 0 to 100 in discrete steps of 5.
In Table 3 there is controlled for tax rates, treatments and sociodemographic variables to determine the effects on effort levels. In support of Result 1, it is observed that effort decreases significantly with tax rates. On average, a one-step increase (i.e. a 5 percent increase) in taxes reduces effort by 2,44 and 1,06 units (holding all other variables constant) conditional on all tax rates and all tax rates above 50 percent, respectively. The UBI treatment has a positive coefficient in both models, but is not significant. With regards to
socio-demographics, none of the variables is significant except the ‘professionally active moral’ in model (1), which is weakly significant at the 10 percent level. Hence, subjects who answered more positively to the statement ‘being professionally active is one’s responsibility towards society’ (on a five-point scale) are more likely to exert higher effort. However, this significance is not found when only taking tax levels above fifty percent into consideration, reducing the robustness of this finding. These findings thus suggest that subjects base their effort decisions completely on tax rates, and that the way these taxes are used, in either the form of insurance or direct redistribution, does not affect their decisions. Or, put differently, these findings suggest that higher initial wealth, paid through taxes, does not affect labour supply decisions.
5.2. Tax Revenues
In Figure 2 a visual representation of tax revenues is given. The figure shows subjects effort levels for each tax rate from zero to 100. From the plot it is observed that tax revenues are near similar up to the 20 percent tax rate. For the Baseline and UBI treatment optimal tax revenue levels are observed at the 30 percent level (average revenue equal to 21,03) and the 45 percent level (average revenue equal to 19,52), respectively.
Figure 2: Average tax revenues for both the Baseline and UBI treatments.
Note: Average tax rate are calculated; (𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑒𝑓𝑓𝑜𝑟𝑡 ∗ 𝑛 ∗ 𝑤𝑎𝑔𝑒 ∗ 𝑡𝑎𝑥 𝑟𝑎𝑡𝑒)
After the 40 percent level, tax revenues decrease steeply in the Baseline treatment, whereas in the UBI tax revenues remain near constant between the 55 percent and 95 percent level. In general, the explanation of the shape of the Laffer curves on tax revenue is twofold. First, as tax rates increase the marginal income for every additional unit of effort decreases. Therefore, individuals are less inclined to exert effort, which drives down tax revenues. Second, the perception of tax rates is important. According to Levy-Garboua et al. (2005) and Keser et al. (2015), high tax rates might be considered to be ‘unfair’. This results in negative emotions which drive workers to exert less effort to ‘punish’ tax authorities by not exerting effort. This in turn reduces the tax revenues. As tax rates have the same marginal effect in both treatments, this difference cannot be explained by the first notion. Hence, this difference might be due to the unconditional payments in the UBI treatment, which in turn lets subjects perceive higher tax rates less unfairly.
In general, no significant difference is found between average tax revenues. when using a Wilcoxon-Mann-Withney test (𝑝 = 0,3833; two-sided). In addition, Table 4 reports coefficient estimates on the determinants of tax revenues. Coefficients shown are OLS estimates with clustering on individuals. Column (1) shows the determinants considering all tax rates, while for column (2) only tax rates above 50 percent were taken into consideration. The independent variables include tax rate, tax rate squared and the dummy variable for treatment. For column (2) the tax rate squared variable is not taken into consideration as no squared relation is expected when tax rates are above 50 percent.
Table 4. Determinants of tax revenue: OLS estimates with clustering on individuals
All tax rates Tax rate > 50%
Variable (1) (2)
Tax rate 2,60***
(0,57)
-0,96** (0,41) Tax Rate Squared -2,74***
(0,47) Treatment 0,16 (0,24) 0,34*** (0,12) Constant 0,12 (0,16) 1,05 (0,33) Observations 882 420 R-squared 0,06 0,03
Notes: standard errors in parentheses *** Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level
The 882 observations in columns (1) and (2) corresponds to the 42 respondents observed for 21 possible tax rates. For column (4) there were only 10 observed tax rates.Treatment variable: 0=baseline treatment 1=UBI treatment. Tax rate variable ranges from 0 to 1 in discrete 0.05 steps
The regression results from column (1) show that tax revenue increases with tax rate as this coefficient is highly significant. The statistically significant negative coefficient estimate on the ‘tax-rate-squared’ variable supports an inverted U-shaped relationship between tax rate and tax revenue (i.e. a Laffer Curve). Column (2) also shows a significant relationship between tax revenue and tax rate. In addition a significant difference between treatments is found in this regression. This leads to the following results:
Result 2. a) Considering all tax rates, no significant differences are found in tax revenues
between the Baseline treatment and UBI treatment. b) When only tax rates above fifty percent are taken into consideration, tax revenues are significantly higher in the UBI treatment compared to the Baseline treatment. c) In both treatments, Laffer curves of tax
revenues take an inverted U-shaped form, resulting in a tax rate for which tax revenue is optimal.
However, it should be noticed that in this particular experiment non-labour payments are (perceived) to be paid through taxes. So to what extent are these payments feasible given the observed tax rates? The ‘government’ should reserve 𝑛 ∗ 𝐴 from taxes to pay to their citizens in the UBI treatment, while in the long run the government should reserve 𝑛 ∗ 𝑝 ∗ 𝐴 in the Baseline treatment. In this case 𝑛 ∗ 𝐴 = 21 ∗ €1,50 = €31,50 and 𝑛 ∗ 𝑝 ∗ 𝐴 = 21 ∗9:8 ∗ €1,50 = €0,875. As observed in Figure 2, optimal tax revenue in the UBI state is equal to €19,52 (at the 45 percent level), which implies that the distribution of these unconditional payments is only possible when the government runs a deficit. Hence, in this experiment, as €19,52<€31,50, labour taxes are not sufficient to support unconditional payments as distributed in the UBI treatment. In contrast, tax revenues are higher than the amount needed for the insurance payments in all tax rates above zero for the Baseline treatment. Moreover, when these results are directly translated to a real-life welfare policy, one could argue that an UBI is only possible when the government runs a direct (and rather large) deficit. This is in line with how some proponent (e.g. Bregman, 2014; Van Parijs, 2004) see an UBI financed: not directly through taxation on labour, but (partly) by taxation on capital movements (i.e. Tobin tax), higher taxes on consumption, and/or savings on bureaucratic costs of the welfare system and reduced poverty in general (which in turn is argued to reduce crime and lower health-care costs). Whether these proposed financing options are sufficient to fully reduce the proposed deficit is far beyond the scope of the paper, but would make up for interesting subsequent research.
6. Discussion
Recent research has investigated how labour supply and tax revenue are influenced by taxes on labour. In this study, that research expanded to the case of an Unconditional Basic Income (UBI). It thus stands in relation to previously conducted field experiments performed by both the US and Canadian governments to test the effects of a guaranteed income on labour supply. As this research makes use of a laboratory experiment, it entails high levels of control in the research design, resulting in the possibility to focus on the behavioural responses of an UBI with regards to labour supply.
Consistent with previous findings and the proposed theoretical model, there is found that labour supply decreases as the tax rate increases. In both treatments (i.e. both in absence and presence of an UBI) average labour supply declines monotonically as tax rates increase. On average, subjects supply more labour when an UBI is in place, especially when taxes are below 25 percent and above 55 percent, compared to the state where individuals receive no unconditional payment. However, these differences are found not to be significant. Hence, it cannot be concluded that an UBI affects average labour supply.
Also consistent with previous findings, it is found that the tax revenue curve is inversely U-shaped and thus takes the form of a Laffer curve. Optimal tax revenues are found to be at the 30 and 45 percent tax rates in case of absence and presence of an UBI, respectively. Tax revenues are observed to be significantly higher after the 50 percent level when an UBI is in place. Thus, it can be concluded that an UBI positively affects tax revenues, as long as the tax rate is above fifty percent.
These results place this research in a somewhat inept position when taking the general discussion about UBI and previous research into consideration. As most opponents argue than an UBI would decrease individuals’ incentives to work and therefore significantly lower labour supply, this research finds no such effects and thus can be used as an argument against that premise. This contradicts findings from the large scale field experiments conducted several years ago. Also, research on taxpayer behaviour almost unanimously find that labour supply increases significantly when taxes are used for either redistributive or insurance purposes, compared to when tax is not used for these purposes (Keser et al. 2015; Ottone & Ponzano, 2011; Ortana et al., 2008). As the UBI payments in this research are considered to be redistributive payments and no significant differences
between treatments are found, one could raise questions about the robustness of the results found in this paper. Especially when considering the relatively small amount of subjects (42) that were used for analysis. It is therefore argued that this research design would benefit replication with a greater number of subjects. In addition, it might be useful to replicate the experiment with a) higher payments, b) a varying economic risk, and c) in both a laboratory and online setting to find an explanation for the relatively low average effort participants exerted compared to Keser et al. (2015).
A possible extension of this work may consist in introducing inequalities in the population and investigate the effects of an UBI when these inequalities exists. That is, the wage subjects receive per decoded letter could be heterogeneous within a certain range across the population. Then the effect on labour supply on certain wage classes can be studied, as the marginal effect of the unconditional payment differs across the population. A more important extension with regards to the Basic Income discussion may lie in the inclusion of different types of jobs. As for most people, the reason to provide labour is a combination of both extrinsic (i.e. financial compensation) and intrinsic motivation (Benabou & Tirole, 2003). That is, a great deal of people simply find some form of enjoyment in the work they do, thus affecting their labour-leisure decision. Although difficult to incorporate, a research design in which different types of jobs that yield different types of motivations are included could provide more insight of the why and how individuals make decisions concerning their labour supply.
From a policy perspective, one could argue based on the results discussed in this paper that an UBI would not affect labour supply as much as expected and that more tax revenues can be collected in case of an UBI when relatively high tax rates are in place. However, in this highly stylized experiment costs of the unconditional payments greatly exceeded tax revenues. When translated directly to a real-life situation, this would mean that governments introducing an UBI should be able to run a large deficit, and that this deficit should be levelled by either taxation on different sources than income, and/or indirect savings that are a consequence of an UBI. And while proponents of an UBI often accredit massive savings to an UBI ranging from health care cost (as the standard of living would be raised), crime fighting costs (due to a decline in criminal activity by reducing poverty) and due to the downsizing of bureaucracy, the question still remains whether this
would be sufficient to finance the redistributive guaranteed payments to all citizens on a monthly basis. Therefore, this research only adds a drop in the bucket in the broad discussing regarding the potential implementation of an UBI. It does however, give insight to some of the key behaviours of taxpayers when receiving unconditional payments. Which when extended further could be of significance when the implementation of an UBI is considered.
