Ability vs. hard work : how does our perception of success affect tax compliance?

(1)

A B I L I T Y V S . H A R D W O R K

HOW DOES OUR P ERCEPT ION OF S UCCESS AFFEC T TAX COMP LIANCE?

A thesis by Lucy Makinson (11087080) Supervised by Matthijs van Veelen

MSc ECO

Track: Behavioural Economics and Game Theory 15 ECTS

This paper investigates whether an individual’s perception of the primary root of their income – be it due to ability or effort – affects their willingness to comply with taxes on the income. Participants completed an online experiment in which priming was used to affect their perception of the earnings task.

Contrary to the predictions of the model, participants were significantly more compliant with taxation when they were told their income was due to their effort than when they were told it was due to ability. However, there is not sufficient evidence to conclude that this was due to changes in task

perception. Instead it raises questions of alternative priming effects, possibly regarding social norms.

(2)

1. INTRODUCTION

Taxation has long been at the heart of political debate, frequently providing an ideological dividing line between the political left and right wings. However, whilst the overall level of taxation often remains divisive, political parties are uniting in their condemnation of tax evasion (and in many cases tax avoidance too). Current estimates put the gross US tax gap at $458 billion, of which roughly 85% is due to underreporting of income, implying a compliance rate of less than 82% (International Revenue Service, 2016).

The increasingly global movement of income and capital makes it harder for governments to enforce taxation, particularly in the case of high earners who in general have far greater access to international financial tools (Christensen & Kapoor, 2004). This makes initial compliance

particularly important – only around 10% of the tax gap is recouped by the IRS through enforcement (International Revenue Service, 2016). A number of high profile tax scandals in

(4)

recent years, such as the Panama Papers and the HSBC scandal in 2015, has renewed pressure on both governments and economists to reduce evasion.

Despite the significant evasion observed, the level is still far lower than predicted by deterrence based models (Alm, McClelland & Schulze, 1992; Feld & Frey, 2002). This implies that factors other than selfish rationality motivate compliance decisions. Understanding these motivations is a crucial step in improving initial compliance.

This paper builds on existing experimental research surrounding tax compliance in asking

whether willingness to comply is dependent on the perceived nature of income – whether it is the result of ability or effort. Whilst there is a clear distinction between the ability and effort in both the philosophical and economic literature underpinning income taxes, this distinction has not yet been addressed in questions of compliance, although many experiments have highlighted how income source more generally does have significant effects.

This paper finds that, contrary to predictions, individuals are more compliant when they believe income to be derived from their effort than when they believe it to be derived from their ability. In the following section the existing work on both the topics of tax compliance, and the

distinction between ability and effort in the more broad tax literature, are outlined. Section 3 outlines the model behind optimal tax rates. Section 4 lays out the methodology of the

experiment, whilst Sections 5 and 6 lay out the results and discussion respectively. Concluding remarks are found in Section 7.

2. RELATED LITERATURE

TAX COMPLIANCE

The founding model of tax compliance can generally be attributed to Allingham and Sandmo (1972), who presented compliance as decision-making under uncertainty. The model was simple – labour supply (and therefore income) was taken as exogenous. On receipt of their income, an individual chose how much to declare (and therefore pay tax on). If any income was not declared there was a probability of being caught, with a fine proportional to the undeclared income. It is

(5)

assumed that all individuals are risk averse, and an optimising agent therefore declares a level of income to maximise their expected utility.

Whilst work on tax compliance has included empirical papers, data on tax evasion is difficult to accurately obtain and empirical studies have generally failed to test specific aspects of the theoretical models (Andreoni, Erard, & Feinstein, 1998). Laboratory experiments have therefore played an important role in understanding tax compliance. Their relevance has been boosted by recent evidence that behaviour in these experiments is a good match for behaviour in the field (Alm et al., 2015).

Early experiments testing the predictions of the Allingham-Sandmo model of compliance were positive, insofar as they found that changes in the key parameters had the expected effects. Increasing the probability of audit and increasing the penalty rate both lead to higher compliance (e.g. Friedland et al., 1978; see Blackwell, 2007 for meta-analysis). When tax revenue was redistributed amongst experimental participants, with tax revenue being adjusted by a

government multiplier, the results continued to fit with a rational agent model of compliance – increases in the marginal-per capita return to the tax revenues also increased tax revenue (Blackwell, 2007). The model cannot make firm predictions in terms of overall compliance effects of a change in the tax rate, and ambiguous results are also found in experiments. In

general, high tax rates correlate with increased evasion (see Andreoni et al., 1998 for a summary) but there are notable exceptions to this relationship (Alm et al., 1995).

Yet despite the Allingham-Sandmo model (and other deterrence-based models) accurately predicting the effect of variable changes on compliance rates they could not completely account for the compliance levels observed, experimentally or empirically (Andreoni et al., 1998). Furthermore, whilst the model treated the compliance decision as a simple decision under uncertainty, the outcomes were not in-line with decisions over equivalent gambles: framing the problem as a tax decision has significant effects on behaviour (Baldry, 1986).

As a result of these issues, models of tax compliance generally shifted from deterrence-based models to models based around norms and the mechanisms surrounding taxation (Doran, 2009). Where early models had ignored the purpose of tax revenue except in such ways as it entered a

(6)

selfish utility function, government accountability and a sense of fairness in the tax system have since been shown to affect compliance.

The effects of fairness on compliance have largely been investigated in two forms – the fairness of the exchange between governments and taxpayers, and fairness in the treatment and actions of different taxpayers. The former includes the effects of reciprocity between government and taxpayers - for example the increased compliance when taxes fund public goods (e.g. Becker, Büchner, & Sleeking, 1987; Kim, 2002) – and the effects of government accountability.

Increased compliance is found when subjects have a say in how revenue is spent (Alm, Jackson & McKee, 1993), an effect echoed in the wider world where direct democracy is associated with increased compliance (Bennno Torgler & Schneider, 2004). More broadly, effective, stable and incorrupt governance is associated with increased tax morale and subsequent compliance (Feld & Frey, 2010) and compliance increases with trust in the tax authorities (Muehlbacher, Kirchler, and Schwarzenberger, 2011).

Fairness between taxpayers is investigated both in terms of differential treatment by government - where inequality in terms of tax rates and public good distribution reduces compliance (Spicer & Becker, 1980; Falkinger, 1995) - and the effects of reciprocity between taxpayers themselves, where compliance behaviour is strongly conditioned on the behaviour of others (e.g. Bordignon, 1993; Bosco & Mittone, 1997). The effects of inequity between government and taxpayers, and between taxpayers themselves, are jointly investigated by Bazart & Bonein (2014) who find that the effects inequity between individual taxpayers dominate the effects of inequity between government and taxpayers as a whole when the two effects are present.

More recently many experiments have considered how income source can affect compliance. This area is still in relative infancy, largely provoked by the key finding that subjects behave differently – both in overall compliance and in responses to tax rate changes – when income is earned rather than endowed (Boylan & Sprinkle, 2001). Given that the majority of income is earned rather than endowed, this finding has substantial repercussions for the study of tax compliance. Subsequent work on income source effects have established that compliance rates are more stable over time when income is earned (Durham, Manly & Ritsema, 2014), and that compliance is also higher when income requires higher levels of effort to earn (Kirchler et al., 2009).

(7)

Duch & Solaz (2015) investigated whether tax compliance differed when income was due to status, luck, or pure achievement. They found that being paid more than others for the same task (a “high status” effect) did not affect compliance, but for those paid less for the task the

relationship between ability and compliance was exacerbated – where high ability participants generally complied less, the gap increased within the “low status” group. They argue that tax avoidance by the rich will be lowest in contexts where they believe income to be derived from achievement rather than luck or status. They do not, however, distinguish between the root of achievement, be it ability or effort.

ABILITY AND EFFORT

Taxes are inextricably linked to redistribution, so at the heart of their justification lies philosophical theories of distributive justice. Amongst the four main theories of distributive justice - Rawls’s justice as fairness, Dworkin’s equality of resources, Steiner-Vallentyne common ownership theories, and Nozick’s entitlements theory – both Rawls and Dworkin treat ability as a factor that is deemed “arbitrary”, i.e. one which should not, in a fair society, affect the outcomes of the individual. In contrast, none of these theories treat effort as arbitrary (Allingham, 2014).

The earliest of these theories is Rawls’ Theory of Justice (Rawls, 1971). Rawls’ considers the most just allocation of “primary goods” as the one which would be selected by an individual under a “veil of ignorance” – where they are not aware of their own position in the allocation. Ability is classified as a “primary good” and therefore forms part of the allocation between which individuals must hypothetically decide. A lack of ability should be compensated by the provision of other “primary goods”, such as wealth, in order to ensure equality across primary good endowments. Under the Rawlsian philosophy transfers from those of higher ability to those of lower ability is a necessary condition to ensure equality. However, in contrast to ability, effort is a choice realised after the hypothetical allocation of bundles. Therefore, inequalities arising from differences in effort are not considered under the Rawlsian problem, and therefore not restricted by it – any outcome is acceptable provided the initial endowments (including ability) are just. Redistribution in the case of differential effort is not necessary.

(8)

This distinction between ability and effort is more clearly applied to redistributive outcomes in Dworkin’s “equality of resources” theory (Dworkin, 1981). Similar to the more colloquial term “equality of opportunity”, equality of resources states that inequalities of outcome can only be permitted insofar as they result from voluntary choices, and not from factors beyond an individual’s control. Dworkin’s particular thought experiment for determining fair allocations focuses on an auction for allocating resources. However, he includes a separate insurance-style system specifically for dealing with inequalities due to handicaps and differences in talents, in order that they should not affect equality outcomes.

Alongside Dworkin’s work is a substantial body of alternative theories falling under the more general term “luck egalitarianism” - roughly bounded by Dworkin (1981) and Cohen (1989) in particular. The consistency across the proposals in this group is the idea that inequalities due to unchosen circumstance (including ability) are not justified, but there is no equivalent lack of justification for inequalities arising through voluntary choice (Scheffler, 2005). This, as with Rawls, means that achieving the optimal distribution of outcomes requires redistribution from those of high ability to those of low ability.

The need to redistribute based on ability is at the core of income tax theory, the starting point for which is generally attributed to Mirrlees (1971). Mirrlees argued that taxes should be conditioned on an individual’s ability to pay and used income as a measure of otherwise unobservable ability. Specifically he streamlined the problem by excluding specific family and social circumstances affecting ability to pay, so that an individual’s ability to pay was equivalent to their “income-earning potential” – best estimated by (but not equal to) their income.

“One might obtain information about a man’s income-earning potential from his apparent I.Q., the number of his degrees, his address, age or

colour; but the natural, and one would suppose the most reliable, indicator of his income-earning potential is his income.”

– Mirrlees (1971)

It is therefore worth stating explicitly that the theoretical justification for income taxes is that income can be used to proxy ability, which is the actual variable of interest. Therefore the better

(9)

income is as a proxy for ability, the more efficient income taxes are in achieving their stated goal.

The most common criticism of income taxes is that they instead tax hard work - which is

undesirable both from a moral perspective, and from a tax efficiency perspective given the aims of income taxes in terms of redistribution (Tuomala, 1990; Mankiw, Weinzierl & Yagan, 2009). In short, the more income depends on ability relative to effort, the more justified income taxes are.

CONTRIBUTIONS OF THIS PAPER

In the extensive work surrounding tax compliance it is clear that levels of compliance are not simply a response to financial incentives and risks, but also a response to social incentives including norms (e.g. Elster, 1989; Naylor, 1989; Alm et al, 1999; Torgler & Frey, 2007). This paper extends existing work on the effects of income source on tax compliance, by

differentiating between income perceived as primarily ability-derived and income perceived as effort-derived. This distinction is clear in the moral arguments surrounding income taxation, but so far has not been addressed in the compliance literature.

This paper is also unusual in its focus on framing. Framing has been demonstrated to affect tax compliance decisions (e.g. Baldry 1986, Mittone 2006, Cadsby et al. 2006) but so far this research has been restricted to whether framing compliance decisions as a tax decision or a gamble affects compliance, and therefore has not addressed the effect framing can have within tax compliance decisions.

With the growing application of behavioural science to government policy, it is perhaps unsurprising that governments are increasingly looking to improve tax compliance through the use of the behavioural models being developed. Recent fieldwork has shown the effectiveness of these approaches in increasing tax compliance voluntarily (Kettle, Hernandez, Ruda, & Sanders, 2016; Leicester, Levell, & Rasul, 2012; UK Cabinet Office (Behavioral Insights Team), 2012) presenting new avenues for reducing both tax avoidance and evasion.

This paper’s focus on framing allows for more obvious policy prescriptions as a result of the conclusions – whilst it is very difficult to change the nature of work in the economy, changing

(10)

the dialogue surrounding work is comparatively straightforward. If framing can affect tax

compliance decisions, there is scope for the nature of public debate to also affect these decisions.

3. OPTIMAL TAXES: A MOD EL

The Mirrlees (1971) model of income tax aimed to base taxes on an individual’s “ability to pay”. Specifically it excluded specific family and social circumstances affecting ability to pay, so that an individual’s ability to pay was equivalent to their “income-earning potential” – best estimated by (but not equal to) their income.

Moving forward, the explicit statement of income tax as a tax on ability is often omitted, and optimal linear income taxes become a puzzle of efficiency trade-offs exemplified in the model of Sheshinski (1972). To simplify: the model considers the optimal tax rate, τ, to maximise the Social Welfare Function (SWF) subject to a government budget constraint. The key exogenous variables influencing the optimal tax rate are the shape of the SWF relative to the initial income distribution, and the elasticity of labour supply. The optimal tax rate is given by

𝜏 = 1 − 𝑔̅ 1 − 𝑔̅ + 𝑒

Where 𝑔̅ is the average “normalised” social marginal welfare weight, weighted by pre-tax incomes, and 𝑒 is the elasticity of aggregate earnings, with respect to the net-of-tax rate. This equation demonstrates the efficiency-equity trade-off. A higher 𝑔̅ implies a greater

preference for redistribution, representing the equity component. The efficiency element lies in the elasticity of labour supply – if a higher tax rate reduces labour supply (𝑒 is positive) then taxes will cause a negative behavioural response, reducing their efficiency.

[For a full derivation, see Piketty & Saez, 2013]

Under this model the root of income could affect the optimal tax rate in two ways - both through the elasticity of labour supply, and through the shape of the social welfare function. An increase in 𝑒 or 𝑔̅ implies a lower optimal tax rate.

(11)

We assume (as standard) that an individual’s utility is an increasing function of consumption, with decreasing marginal utility [fc > 0, f’c < 0] and a decreasing function of labour effort [fl < 0]. For simplicity we assume the following functional form

𝑈_𝑖(𝑐, 𝑙) = ln 𝑐 − 𝛾𝑖𝑙2

where γ is a measure of leisure preferences that varies by individual. It is worth noting that labour effort and consumption are linked, and so optimising levels of consumption (or labour effort) are determined by the relationship between consumption and labour effort.

It is first worth considering the structure of an economy in which ability does not have any effect on income. In such an economy, any differences in income are due to differences in preferences. Consider a simple, 2-person economy with only one good. As there is only one good, there is no trade – each individual produces only as much as they consume. Therefore an individual with a higher preference for leisure relative to consumption will produce and consume less than the other individual. With the addition of more goods trade occurs to exploit comparative advantage, increasing the joint consumption of the individuals; however differences in preferences still allow for different overall levels of consumption (income).

If income is determined only by labour effort then consumption is equal to labour effort, minus taxes (as wages will be the same across individuals) we normalise wages to 1.

𝑐 = (1 − 𝜏)𝑙

An individual maximises the following utility function, with respect to c. 𝑈𝑖(𝑐) = ln 𝑐 − 𝛼𝑖𝑐2

where

𝛼_𝑖 = 𝛾𝑖 (1 − 𝜏)2

(12)

When wages are determined by ability, with high ability individuals earning more for each unit of effort, the relationship between consumption and labour effort is given by

𝑐 = (1 − 𝜏)𝑤_𝑖𝑙

This allows an individual’s utility function to be written in terms of consumption as 𝑈_𝑖(𝑐) = ln 𝑐 − 𝛽_𝑖𝑐2

where a new variable can be defined as

𝛽_𝑖 = 𝛾𝑖 (1 − 𝜏)2_𝑤

𝑖2

It is clear that a population’s income distribution is defined by the distribution of α in an effort-driven economy or β in an ability-effort-driven economy, and that the two are equivalent when the distribution of α is equal to the distribution on β. Therefore a given income distribution, whether it is determined by ability or effort, is comprised of the same distribution of either α or β and is therefore comprised of the same utility functions. If all utility functions in the economy are the same, then all labour supply decisions are the same, and the elasticity of labour supply must be identical in the two cases. If ability affects income purely through increasing wages, there is no difference in labour supply elasticity between economies where ability is a factor and economies where it is not.

SOCIAL WELFARE FUNCT ION

That the optimal tax rate in a society should depend on the preferences for wealth distribution within the society is hardly surprising. Clearly a society with Rawlsian social preferences will favour higher redistribution than a society with utilitarian preferences, because under the Rawlsian preferences the utility of any individual other than the least well-off is irrelevant. Therefore even with a highly inefficient tax system the optimal tax rate on everyone except the least well-off is 100%. A more general point can be made intuitively: the more accepting a society is of income disparities, the less redistribution is required to achieve the social optimum, and the lower its optimal tax rate will be.

(13)

To demonstrate this formally, consider the definition of 𝑔𝑖 and 𝑔̅. 𝑔𝑖 measures the dollar value of increasing consumption of individual 𝑖 by $1.

𝑔𝑖 =

𝜔_𝑖𝐺′(𝑢𝑖_)𝑢 𝑐𝑖 𝑃

Where 𝜔_𝑖 are the Pareto weights assigned to each individual 𝑖 (generally assumed to be independent of 𝑐, consumption and 𝑧, income)1, 𝐺′(𝑢𝑖)𝑢_𝑐𝑖 is an increasing transformation of utilities, and 𝑃 is the government multiplier.

𝑔̅ is defined as the ratio of the average income weighted by social welfare weights, 𝑔_𝑖

𝑔̅ =∫ 𝑔𝑖𝑧𝑖𝑑𝑣(𝑖) 𝑍

Some basic points can be noted surrounding 𝑔_𝑖 and 𝑔̅. If a society has no preference for redistribution then 𝑔_𝑖 ≡ 1, and therefore 𝑔̅ = 1 implying optimal taxes are zero. In contrast, if society favours complete redistribution (the only criterion being the utility of the least well-off individual) then 𝑔𝑖 ≡ 0 for all except the least well-off individual. Assuming the least well off individual has an income of zero, this implies 𝑔̅ = 0, and optimal taxes are equal to

𝜏 = 1

1 + 𝑒

which is equal to the revenue maximising tax rate. This confirms the initial intuition: a society that has a greater preference for redistribution will favour higher taxes.

From the moral philosophy surrounding income distribution it is clear that high incomes are justified when they are the product of effort, but not when they are the product of ability. We would therefore expect a greater preference for redistribution when income is due to ability (and less justified), and therefore a higher optimal tax rate.

Mathematically the idea that those with high incomes are entitled to them can be shown as a greater welfare weight being assigned to those with higher incomes. Rejecting the assumption

1_{income, 𝑧, is used here in place of labour effort, as used in our previous formulations. This is largely for}

consistency with the general formulations. However it does not affect the overall analysis as any adjustments can be

(14)

that Pareto weights should be independent of 𝑐 and 𝑧, if those with a higher income have this income due to effort (and are therefore considered more deserving of income) 𝑐𝑜𝑣(𝜔_𝑖, 𝑧_𝑖) > 0. If Pareto weights are positively correlated with income then 𝑔_𝑖 will be more positively correlated with income, 𝑧𝑖, than for otherwise identical social weightings which are not correlated with income. Provided the sum of the weights 𝑔_𝑖 are normalised, a positive correlation between 𝑔_𝑖 and 𝑧𝑖 results in a higher value of 𝑔̅ compared to when they are uncorrelated. Incomes derived from effort are more justified and hence the average marginal social welfare weight is higher, resulting in a lower optimal tax rate.

4. METHODOLOGY

The objective of this study is to examine the effect of income framing on tax compliance. Specifically it investigates whether income framed as a reflection of ability is treated differently from income framed as a reflection of hard-work.

PREVIOUS RESULTS AND HYPOTHESES

The primary hypotheses of this paper centres around the legitimacy of income taxes when they are a tax on ability, compared to when they are a tax on effort. If principles of fairness affect compliance, then we would expect higher levels of compliance the more income is believed to be derived from ability. There is a wealth of evidence supporting a link between compliance and fairness, for example; Kim (2002) who finds increased compliance when the monetary exchange with government was considered to be fairer, Feld & Frey (2010) show that tax morale and subsequent compliance are significantly affected by the accountability and quality of

government, and Bazart & Bonein (2014) who find increased compliance by those who benefit from current inequity (see Bazart & Bonein, 2014, for a full review of literature relating to fairness in tax compliance).

Therefore, we would expect compliance to depend positively on the perceived strength of the relationship between income and ability.

H1: Participants who attribute the source of income to ability will be more likely to comply with taxation.

(15)

Duch & Solaz (2015) find that levels of compliance are highly dependent on structural

advantages and disadvantages. Specifically they find that at high (30%) tax rates, none of those in the “low” status treatment fully declare their income. Similarly Bazart & Bonein (2014) find that disadvantageous inequity in tax rates increases the number of evaders, but advantageous vertical inequity decreases the level of evasion (though not the number of evaders).

If ability is perceived as a structural factor, then those that believe themselves to be of lower ability will be less likely to fully comply with taxation in the ability treatment. This does not apply to the effort condition, as effort is not a structural disadvantage.

H2a: Participants who believe they are of lower ability relative to the group are less likely to comply with taxation in the ability treatment.

In addition, the effects of overconfidence in ability relative to others (“better-than-average” overconfidence effects) are well documented (e.g. Svenson, 1981). We therefore anticipate overconfidence in task performance.

H2b: Participants will exhibit overconfidence in perceptions of their own performance.

A potential effect of H2a and H2b is that the majority of participants in the ability condition may feel structurally disadvantaged by the tax regime, and may declare less income as a result. Provided H2a and H2b hold, this could create a contradiction to H1 whereby those in the ability condition declare less income.

H2c: Conditional on H2a and H2b, participants in the ability treatment will declare less income than those in the baseline treatment.

Finally, the treatments may not only affect tax compliance, but also task behaviour. Mueller & Dweck (1998) find that when a task is perceived as requiring effort to succeed, participants exert more effort. In addition, Kirchler et al. (2009) find that compliance falls under low-effort

conditions. Therefore, we anticipate that participants in the effort treatment will exert more effort, and resultantly score higher than those in the control treatment, and also have higher compliance rates.

(16)

H3: Participants who believe effort is a large determinant of income will perform better on the task and have higher tax compliance rates.

EXPERIMENTAL DESIGN PARTICIPANTS

A total of 76 participants were recruited, primarily students and recent graduates. There were 34 male participants and 42 female participants with an average age of 25 across all participants.

PROCEDURE OVERVIEW

The experiment was conducted online over a 2 week period, with participants being sent a link to the experiment. It consisted of 2 stages, an earnings stage and a declaration stage, and was

preceded by instructions and followed by a brief questionnaire. Participants’ final payoffs were decided by these two stages. All participants were randomly and anonymously allocated into groups of four. Following their performance in the earnings stage they were asked to declare income. Any undeclared income was kept by the individual participant. Any declared income was taxed at 30% and the tax revenues were split equally between the 4 group members. Final payments for each participant were given by

𝑃_𝑖 = 𝑦_𝑖 − 0.3𝑑_𝑖 +1 4∑ 𝑑𝑗

4

𝑗=1

where yi is their initial earnings, and di is their declared earnings.

One group of four was randomly selected for payment once the experiment had been completed by all participants2.

2_{By selecting a whole group the optimal strategy for each player is unchanged by the payment structure. If 4}

individuals were chosen at random, risk-averse individuals may be more likely to favour redistribution amongst the group, and therefore comply more with taxation.

(17)

Before beginning the task the participants were presented with full instructions detailing the procedure for themselves, how payments would be calculated and enacted, and how the data would be processed to ensure anonymity. In particular the instructions made clear that

participants were not required to declare all their earned income, and understanding of this fact was subsequently checked. All participants had to answer 3 short questions to ensure that they had read the instructions fully.

All participants first completed a 90-second real-effort task to determine their initial income for the experiment. The task consisted of non-verbal multiple-choice questions of increasing

difficulty. Participants received 10 experimental units for every question answered correctly. No participant completed all the questions before the time expired.

Once the time had ended, participants were immediately shown their earnings for the task. They were reminded that any declared income would be taxed at 30%, and that tax revenues would be equally distributed amongst group members. On the same page they were presented with a box to declare their earnings.

Following the earnings declaration participants were asked to complete a short questionnaire. The content of the questionnaire is discussed below, but the full script can also be found in the appendix. At the end, participants could provide their email in order to be contacted for payment, should their group be selected. 63 of the 76 participants chose to share their email3.

TREATMENTS

There were three treatments which affected only the way in which the real-effort task was framed; a baseline treatment (with no framing), an ability treatment (which framed the earnings task as a good measure of ability), and an effort treatment (which framed the earnings task as a good measure of effort). Mueller & Dweck (1998) show that praising children for either their effort or ability following a task significantly affects their perception of the relative importance of effort compared to ability in task achievement. This demonstrates that framing can have significant impacts on task perception. In this paper the framing occurred at two points. In the initial instructions participants were told the nature of the task with the description being

3_{There was no significant difference in any of the variables between those who shared their email and those who did}

(18)

treatment-dependent, and in the questions following the instructions (to ensure they had been read) the description of the task was treatment-dependent. This second framing was aimed at improving the salience of the task-description, without allowing participants to ascertain the experimental question.

For full wording of all stages of the experiment, see the appendix. The table below shows the changes in wording across the three treatments.

Instructions Follow-up questions

Baseline Treatment

You will receive 10 experimental units for every question answered correctly, and there will be one example question before the task starts.

In the first stage you will be asked to answer multiple choice questions. How long will you have to answer these questions? Ability

Treatment

Your achievement in these questions is strongly determined by your ability. You will receive 10 experimental units for every question answered correctly, and there will be one example question before the task starts.

In the first stage you will be asked to answer multiple choice questions to measure your ability. How long will you have to answer these questions?

Effort Treatment Your achievement in these questions is strongly determined by the effort you put in. You will receive 10 experimental units for every question answered correctly, and there will be one example question before the task starts.

In the first stage you will be asked to answer multiple choice questions to measure your effort. How long will you have to answer these questions?

EARNINGS TASK

The key criterion for the earnings task was that the questions set could reasonably be perceived as both a test of effort, and a test of ability. Raven’s Standard Progressive Matrices (e.g. Raven, 1958) are frequently used as a test of non-verbal fluid intelligence, both within academic studies and as selection criteria. Mueller & Dweck (1998) used these tests to examine the effect of effort- vs. ability-based praise on performance and found that the type of praise affected the

(19)

perception of the task, and that individuals praised for effort subsequently performed better. This shows that, whilst Raven matrices measure ability, they are strongly affected by the level of effort exerted and this is reflected by individual’s perceptions of the questions which can be manipulated by framing. The questions have the additional advantage of being non-verbal. Given that a large number of participants may be foreign students this reduces the chance of

participants feeling discriminated against within the task. Whilst Raven matrices were not used for budget reasons, questions of a similar style were sourced from an online database.

An additional consideration was the level at which the puzzles should be pitched. In order to reduce fatigue, the task was kept to 90-seconds. The questions within this time were pitched at a reasonably low level – all participants should be able to complete the questions relatively easily even under time pressure. This allowed for higher scores, which would likely yield greater variation. In addition, it increased the relative importance of effort relative to ability. Given that Raven tests are generally used to test ability, shifting the focus to effort was considered an advantage. The average score over the 90-second exercise was 15, which is not too dissimilar from the 12-13 point average in Duch & Solaz (2015).

TAX RATE

Given the constraints in terms of participant numbers this paper uses the same tax rate for all participants. Duch & Solaz (2015) used 3 different tax rates; 10%, 20% and 30%. They found significant differences in the compliance behavior of “low” and “high” status types at the 30% tax level (where “high” status types were more likely to be compliant), but the differences were not apparent at the lower tax levels. “Low” status types evaded tax more than at lower tax levels, whilst “high” status types evaded less. This suggests that feelings of fairness play a greater role at the higher tax rate. As fairness is a key reason for distinguishing between ability- and effort-based income, this paper uses a 30% tax rate.

QUESTIONNAIRE

Alongside asking for age and gender, the questionnaire contained four further questions. The first two related to participants’ perception of the task – whether it required effort or ability, and what they considered their relative performance to be. These questions were included to examine

(20)

how task perception affected compliance behaviour. The second two questions were related to attitudes to taxation and were taken from the World Values Survey (used by several other papers investigating tax attitudes, e.g. Alm & Torgler, 2006; Daude, Gutierrez & Melguizo, 2012; Tekeli, 2011; Torgler & Schneider, 2007). This allowed analysis of how tax attitudes explain compliance behaviour.

For full wording of the questionnaire, see the appendix.

5. RESULTS

SUMMARY STATISTICS

Despite participants being randomised to treatment groups, allocations to group 2 (the ability group) were significantly higher than to the baseline and effort groups. There is no significant difference in age between treatment groups, however those in the ability treatment are

significantly more likely to be female with 75% female compared to 50% in the other treatment groups.

Table 1: Demographic statistics by treatment

Overall Group 1: base Group 2: ability Group 3: effort Average age 25.12 (5.26) 24.05 (2.19) 25.63 (6.36) 25.38 (5.40) Male 34 11 19 4 Female 42 11 19 12 Female % 55% 50% 50% 75% N 76 22 38 16

Figures in parentheses indicate standard errors

Gender, however, is not significantly correlated with any other explanatory variables (including test score). The only significantly correlated explanatory variables are tax cheating with both percentage compliance and income equality – those agreeing with the statement “Cheating on taxes if you have a chance is never justified” declared a higher share of their earnings and were also more likely to support the statement “Incomes should be made more equal”.

(21)

Table 2: Pairwise correlation coefficients between explanatory variables G ender Ag e Sco re % co mp lia nce T a sk P er ce ptio n Rela tiv e P er fo rma nce Inco me E qu a lity T a x Chea ting Gender 1.000 Age 0.092 1.000 Score -0.113 -0.211 1.000 % compliance 0.223 0.114 -0.174 1.000 Task Perception -0.116 0.006 -0.006 -0.186 1.000 Relative Performance -0.181 0.006 0.176 -0.048 -0.033 1.000 Income Equality 0.035 0.081 -0.153 -0.160 0.117 0.022 1.000 Tax Cheating 0.019 -0.206 0.029 -0.289* -0.181 -0.163 0.3131* 1.000 *p<0.1, **p<0.05, ***p<0.01 MEASURING COMPLIANCE

There are 3 possible ways of measuring compliance. The first is to measure compliance as the percentage of total income declared, and the other two options are to measure compliance as a binary variable which either reflects whether an individual fully complied or not, or reflects whether an individual has partially complied or not. Across the 76 observations half (38) fully complied by declaring all their income, 27 partially complied and only 11 participants did not declare any income.

(22)

The primary measure in this paper is a binary variable that measures whether individuals

completely complied or not and (unless otherwise stated) references to tax compliance will refer to this measure. This measure was the most responsive to the independent variables – both changes in treatment and also questionnaire responses. In particular, the percentage compliance was particularly unresponsive when the extensive margin was ignored (participants who were fully compliant or non-compliant were excluded), demonstrating that any variation in percentage compliance was due to movements into- and out-of- zero or full compliance.

The choice of using a binary variable indicating complete compliance as opposed to partial compliance was both because of its increased responsiveness, and also because it is a more relevant measure – from a legal and political standpoint an individual either complies or they do not. Furthermore, using a binary variable of full compliance is in line with Duch & Solaz (2015). It is worth noting, however, that the effect of each independent variable took the same direction on all measures of compliance – only the magnitude and significance was affected by the choice of variable (see Table 3).

Clear predictions could be made regarding the relationship between the income equality and tax cheating measures and compliance. In both cases a negative effect was anticipated, indicating that those agreeing with the statements “incomes should be made more equal” and “cheating on taxes if you have a chance is never justified” would be more likely to comply with taxes. That significant negative relationships were found in both cases against the binary full compliance measure lends additional support to its validity as a measure of compliance. There was also a negative relationship found between a participant’s score and their compliance, in line with Duch & Solaz (2015), although the relationship was not significant.

Table 3: The directional effect of independent variables on each measure of compliance Measure4 In d ep en d en t Var iab

le Percentage _Compliance Binary: Fully _complied Binary: Partially _complied

Score ( - ) ( - ) ( - )

4_{Effects on percentage compliance estimated using OLS linear regressions, whilst effects on binary variables are}

(23)

Relative Performance ( - ) ( - ) ( - ) Income Equality ( - ) ( - ) ** ( - ) Tax Cheating ( - ) *** ( - ) *** ( - ) *p<0.1, **p<0.05, ***p<0.01 EFFECTIVENESS OF TREATMENT

To establish the effectiveness of the treatment, participants were asked in the follow-up questionnaire “On a scale of 1 to 10, to what extent do you feel the exercise was a good test of ability (compared to effort)?”. We would anticipate that, if the treatment were effective, the average score in the ability treatment would be higher than in the base treatment, which would in turn be higher than in the effort treatment.

The average response to the ability vs. effort question was highest in the ability treatment, but it was not significantly higher than the effort treatment at the 5% level. This could be in part due to the small sample size in the effort group – in a one-tailed t-test the average score was

significantly higher when the ability group was compared to the combined base and effort groups (t = - 1.70, p = 0.047).

In addition to shifting task perception, the treatments consolidated views on the task. Even in the case of the effort treatment, which did not significantly shift aggregate perceptions, there was less variation in the perception of the task than in the baseline case. This suggests that giving more information on the task caused views to converge. In a one-sided F-test the variance of the responses to the ability vs. effort questions is significantly lower in the combined treatments than in the base condition (F = 1.90, p = 0.030).

Table 4: Effect of treatment on task perception

Treatment

Overall Group 1: base Group 2: ability Group 3: effort

(24)

Standard Deviation 2.26 2.76 2.03 1.9

Overall, there is evidence that the treatments were salient in both cases, reflected by the reduced variance in task perception. Furthermore, the ability treatment significantly affected the

perception of the task.

H1: EFFECTS ON TAX COMPLIANCE

H1: Participants who attribute the source of income to ability will be more likely to comply with taxation.

Comparing the ability and effort treatments, the proportion fully declaring income is significantly higher in the effort treatment than in the ability treatment – contrary to the predictions of the model – implying that when income is more heavily dependent on ability individuals are actually less willing to pay taxes (z = - 2.15, p = 0.032). This reflects both an increase in full compliance relative to the base in the effort treatment, and a decline relative to the base in the ability treatment although neither difference is statistically significant.

However a probit regression of tax compliance on task perception found no significant result. This could be due to the amount of noise in the task perception measure, which given the small sample size means significant results are unlikely to be obtained. Additionally, a confounding variable could be partially offsetting the effects of the treatment – if individuals who have a higher tendency to compliance are also more likely to attribute tasks to ability, this would offset the effects of the treatment. There is supporting evidence for this; a probit model run on the base treatment only demonstrates a weak positive effect of task perception on the probability of compliance, although the result is not statistically significant.

Another explanation is that the treatment effect is caused by group composition, in particular the high prevalence of women in the effort group. However, whilst women have a slightly higher

(25)

compliance rate than men across all treatments, this result is not significant. Furthermore, the effect of treatment is still significant when gender is included in a probit model5.

ADDITIONAL HYPOTHESES

H2a: Participants who believe they are of lower ability relative to the group are less likely to comply with taxation in the ability treatment.

There is a negative relationship between perceived relative performance and tax compliance across all treatments, although it is not significant and the effect is primarily explained by the weak negative effect of score on tax compliance, in line with the findings of Duch & Solaz (2015).

Introducing an interaction term between ability and relative performance demonstrates a strong negative effect on compliance (z = - 2.40, p = 0.016), however this is entirely explained by the substantially lower compliance rate in the ability treatment. This provides no evidence that those who believe they have performed less well feel structurally disadvantaged by the tax system, even when income is based on ability.

Table 5: Effect of relative performance perception on tax compliance

Dependent variable: Tax Compliance

(1) (2) (3) Relative Performance - 0.0770 (0.0861) - 0.0318 (0.0894) - 0.0177 (0.117) Ability * Relative Performance -0.104** (0.0432) - 0.136 (0.178) Ability (dummy) 0.225 (1.21) Constant 0.513 (0.591) 0.554 (0.600) 0.454 (0.809) Observations 76 76 76 Pseudo R2 0.0076 0.0637

(26)

*p<0.1, **p<0.05, ***p<0.01

H2b: Participants will exhibit overconfidence in perceptions of their own performance.

There is strong evidence of overconfidence across the sample as a whole, with an average

relative performance score of 6.67, significantly higher than the 5.5 score which would represent a belief of median achievement (t = 6.09, p = 0.0000). Furthermore, perceptions of personal performance are not significantly related to score (t = 1.28, p = 0.203) nor significantly affected by treatment effects or task perception.

H3: Participants who believe effort is a large determinant of income will perform better on the task and have higher tax compliance rates.

There is no significant relationship between declared task perception and final scores in the task (t = - 0.11 p = 0.909). Additionally, there is no significant difference in average score between the two treatment groups (t = 0.39 p = 0.698). However, there is a significant difference between the scores in the treatment groups and the baseline group (t = 2.66 p = 0.009), which could suggest that it is increased information about the task that affects effort as opposed to the perceptions of the task itself.

However, this does not correspond to lower compliance in the base group compared to the effort groups, so does not verify Kirchler et al. (2009)’s finding that compliance is lower under low effort conditions.

6. DISCUSSION

The key result of interest is the finding that, contrary to predictions, participants who were told that their achievement is based on their effort were more likely to declare their income fully than those who were told their achievement was based on their ability. This goes against the idea that individuals feel more entitled to income that they believe they have earned due to their effort. The robustness of this finding is reflected in the consistency across all measures of compliance – those in the effort condition were the most likely to fully comply, the least likely to not comply

(27)

at all, and amongst those who partially complied their average percentage compliance was the highest. However they were not more likely to attribute the task to effort than the base condition, which makes determining the mechanism behind the increased compliance less clear. One possibility is that mentioning “effort” caused other associations, which changed compliance behaviour. For example, if “effort” is associated with civic responsibility the use of the word could cause participants to be more focused on ideas of responsibility and subsequently be more compliant, even if it does not change their perception of the task. Such a mechanism is not dissimilar from the famous priming study by Bargh et al. (1996) who found that participants exposed to words associated with the elderly subsequently walked more slowly, although the results of such experiments are still considered controversial (e.g. Doyen et al., 2012). In the context of this experiment, this is a potential area for further research – if the causality is correct then using the word “effort” even when it is not related to the nature of the task would also increase compliance. If the change in compliance is not due to a change in task perception this leaves the initial question unanswered, however it does still suggest that changes to the language used around tax can make a significant difference to compliance.

In terms of the second hypothesis – how overconfidence could lead to lower compliance in the ability treatment – whilst there was strong evidence of overconfidence there was no evidence that beliefs regarding personal performance affect compliance, which undermines the mechanism outlined in the hypothesis. Furthermore it suggests that those who believe they performed badly do not consider themselves unfairly penalised, even if they believe the task to be ability

dependent, as this would contribute to lower compliance (Duch & Solaz, 2015). A possible area for further research could be whether this changes where a task is more obviously discriminatory – for example in a language-based task – although this was a question this paper deliberately sought to avoid.

Finally, scores on the task were significantly higher in the treatment groups than in the base condition, but did not vary significantly between treatments. This could suggest that the

provision of more information regarding the task had a positive effect on effort. The conjecture that increased effort on the task increases compliance is not supported however, as whilst compliance is higher than the base in the effort treatment it is lower than the base in the ability treatment.

(28)

LIMITATIONS

There are limitations to the experimental method that necessitate caution when drawing

conclusions. The first is the small sample size, particularly in the effort treatment. This, in many cases, reduced the potential to draw significant conclusions from the data. Other potential threats to internal validity are issues of fatigue and incentives. Fatigue and a lack of effort on the part of participants may have meant instructions were not read thoroughly, which is particularly

important as this is where a large part of the treatment occurred. By the end of the experiment participants may also have been reluctant to read and consider the questionnaire thoroughly. Whilst questions were posed to ensure the instructions were read, and which reiterated the treatment, these factors (and particularly in the case of the questionnaire) cannot be entirely eliminated. However, the fact that responses to the questionnaire demonstrated the expected correlations between variables and with compliance where predictions could be made suggests that responses were not wholly arbitrary. Any issues of a lack of effort may have been

exacerbated by the small probability of being selected for payment.

The paper also faces the many potential problems faced by tax compliance experiments in general regarding external validity, most notably issues of replicating the importance,

consequences and complications of real tax decisions. For example this paper, along with most on the subject, simplifies the problem by using only one source of income. However this ignores the different ways in which taxes are evaded, for example through overstating expenses (Kim, 2002).

In order to encourage participants to consider their decisions as they would a tax problem, this paper used loaded terminology such as “tax” and “income”. This was based on the evidence that tax terminology generally produces slightly higher compliance rates than neutral terminology, which better replicates real-world behaviour (Durham et al., 2014). It was particularly important in this paper given the focus on norms and behavioural aspects of compliance, which are context-dependent. Whilst Alm (1991) questions the generalizability of laboratory results when non-neutral terminology is used, more recently Alm (2015) finds that the behaviour of laboratory participants in tax compliance experiments is in-line with behaviour observed in the field. The laboratory experiments used to draw these conclusions used loaded tax language, and the

(29)

findings were consistent for student participants and non-students. This implies that the general problems faced by tax compliance experiments do not significantly affect external validity. One way in which this paper differs from many papers on the topic is in not auditing

declarations, so there was no risk involved in evasion. The conventional interpretation is that this would simply decrease compliance rates in all cases, so should not significantly affect the

interpretation of results between treatments. However, the use of punishment can also be seen to crowd-out pro-social norms (e.g. Gneezy et al., 2011). If the lack of auditing and fines changes the norms under which the compliance decision is made this may have a differential effect on compliance decisions dependent on treatment, which would cause problems for the external validity of the study.

7. CONCLUSION

Our results indicate that participants who were told their achievement was due to the effort they had put in were more likely to declare their income for taxation, however it is not possible to conclude that this was due to a change in their beliefs regarding the root of income. Instead this research presents a starting point for further questions, in particular regarding the potential effect of priming on compliance decisions, but also how different earnings tasks which are more explicitly determined by ability and effort could affect compliance.

REFERENCES

Allingham, M. (2014) Distributive Justice, London, Routledge.

Allingham, M. & Sandmo, A. (1972). Income tax evasion: a theoretical analysis. Journal of Public Economics, 1(3-4), 323-338.

Alm, J. (1991). A perspective on the experimental analysis of taxpayer reporting. The Accounting Review, 66(3), 577–593.

Alm, J. & Torgler, B. (2006). Culture differences and tax morale in the United States and in Europe. Journal of Economic Psychology, 27(2), 224-246.

(30)

Alm, J., Bloomquist, K., & McKee, M. (2015). On the external validity of laboratory tax compliance experiments. Economic Inquiry, 53(2), 1170-1186.

Alm, J., Jackson, B., & McKee, M. (1993). Fiscal exchange, collective decision institutions, and tax compliance. Journal of Economic Behavior & Organization, 22(3), 285-303.

Alm, J., McClelland, G. H., & Schulze, W. D. (1992). Why do people pay taxes?. Journal of public Economics, 48(1), 21-38.

Alm, J., McClelland, G., & Schulze, W. (1999). Changing the Social Norm of Tax Compliance by Voting. Kyklos, 52(2), 141-171.

Alm, J., Sanchez, I., & DE Juan, A. (1995). Economic and Noneconomic Factors in Tax Compliance. Kyklos, 48(1), 1-18.

Andreoni, J., Erard, B., & Feinstein, J. (1998). Tax compliance. Journal of Economic Literature. Retrieved from http://www.jstor.org/stable/2565123

Baldry, J. (1986). Tax evasion is not a gamble. Economics Letters, 22(4), 333-335.

Bargh, J., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of Personality and Social

Psychology, 71(2), 230-244.

Bazart, C. & Bonein, A. (2014). Reciprocal relationships in tax compliance decisions. Journal of Economic Psychology, 40, 83-102.

Becker, W., Büchner, H.-J., & Sleeking, S. (1987). The impact of public transfer expenditures on tax evasion. An experimental approach. Journal of Public Economics, 34(2), 243–252. Blackwell, C. (2002, March). A meta-analysis of tax compliance experiments. In Annual

Conference on Public Choice in San Diego.

Bordignon, Massimo (1993), ‘A Fairness Approach to Income Tax Evasion’, Journal of Public Economics, 52(3) 345-362.

Bosco, L. & Mittone, L. (1997). Tax Evasion and Moral Constraints: some Experimental Evidence. Kyklos, 50(3), 297-324.

(31)

Cadsby, C., Maynes, E., & Trivedi, V. (2006). Tax compliance and obedience to authority at home and in the lab: A new experimental approach. Experimental Economics, 9(4), 343-359.

Christensen, J., & Kapoor, S. (2004). Tax avoidance, tax competition and globalisation: Making tax justice a focus for global activism. Accountancy Business and the Public Interest, 3(2), 1-16.

Cohen, G. A. (1989). On the Currency of Egalitarian Justice. Ethics, 99(4), 906–944. http://doi.org/10.1086/293126

Daude, C., Gutierrez, H., & Melguizo, A. (2012). What Drives Tax Morale? (No. 315). OECD Publishing. 104(4), 771.

Doran, M. (2009). Tax penalties and tax compliance. Harvard Journal on Legislation, 46(1), 111–162.

Doyen, S., Klein, O., Pichon, C., & Cleeremans, A. (2012). Behavioral Priming: It's All in the Mind, but Whose Mind?. Plos ONE, 7(1), e29081.

Duch, R.M., & Solaz, H. (2015). Why we Cheat: Experimental Evidence on Tax Compliance, Working Paper. Centre for Experimental Social Sciences, Nuffield College, University of Oxford. Retrieved from

http://politics.as.nyu.edu/docs/IO/38899/Why-we-Cheat-Template-Dec2014.pdf

Durham, Y., Manly, T., & Ritsema, C. (2014). The effects of income source, context, and income level on tax compliance decisions in a dynamic experiment. Journal of Economic

Psychology, 40, 220-233.

Dworkin, R. (1981). What is equality? Part 2: Equality of resources. Philosophy and Public Affairs 10 (4):283 - 345.

Elster, J. (1989). Social Norms and Economic Theory. Journal of Economic Perspectives, 3(4), 99-117.

Falkinger, J. (1995). Tax evasion, consumption of public goods and fairness. Journal of Economic Psychology, 16(1), 63-72.

(32)

Feld, L.P. & Frey, B.S. (2010). Tax evasion and the psychological tax contract. In Alm, J., MartinezVazquez, J. & Torgler, B. (eds.) Developing Alternative Frameworks for Explaining Tax Compliance (pp. 74-94). London and New York: Routledge.

Frey, B. & Torgler, B. (2007). Tax morale and conditional cooperation. Journal of Comparative Economics, 35(1), 136-159.

Friedland, N., Maital, S., & Rutenberg, A. (1978). A simulation study of income tax evasion. Journal of Public Economics, 10(1), 107-116.

Gneezy, U., Meier, S., & Rey-Biel, P. (2011). When and Why Incentives (Don't) Work to Modify Behavior. Journal of Economic Perspectives, 25(4), 191-210.

International Revenue Service. (2016). Tax Gap Estimates for Tax Years 2008–2010.

Kettle, S., Hernandez, M., Ruda, S., & Sanders, M. (2016). Behavioural Interventions in Tax Compliance : Evidence from Guatemala. Policy Research Working Paper Series 7690, The World Bank.

Kim, C. (2002). Does fairness matter in tax reporting behavior?. Journal of Economic Psychology, 23(6), 771-785.

Kirchler, E., Muehlbacher, S., Hoelzl, E., & Webley, P. (2009). Effort and Aspirations in Tax Evasion: Experimental Evidence. Applied Psychology, 58(3), 488-507.

Leicester, A., Levell, P., & Rasul, I. (2012). Tax and benefit policy: insights from behavioural economics. IFS Commentary C125. http://doi.org/10.1920/co.ifs.2012.0124

Mankiw, N., Weinzierl, M., & Yagan, D. (2009). Optimal Taxation in Theory and Practice. Journal of Economic Perspectives, 23(4), 147-174.

Mirrlees, J. (1971). An Exploration in the Theory of Optimum Income Taxation. The Review of Economic Studies, 38(2), 175.

Mittone, L. 2006. “Dynamic behaviour in tax evasion: An experimental approach.” The Journal of Socio-Economics 35, 813-835.

Muehlbacher, S., E. Kirchler and H. Schwarzenberger (2011), “Voluntary versus Enforced Tax Compliance: Empirical Evidence for the “Slippery Slope” Framework”, European Journal of Law and Economics 32, 89–97.

(33)

Mueller, C. & Dweck, C. (1998). Praise for intelligence can undermine children's motivation and performance. Journal of Personality And Social Psychology, 75(1), 33-52.

Naylor, R. (1989). Strikes, Free Riders, and Social Customs. The Quarterly Journal of Economics,

Piketty, T., & Saez, E. (2013). Optimal labor income taxation. Handbook of Public Economics, 5, 392–474. http://doi.org/10.1016/B978-0-444-53759-1.00007-8

Raven, J. (1958). Standard progressive matrices. London: H.K. Lewis.

Rawls, J. (1971). A theory of justice. Cambridge, MA: Belknap Press of Harvard University Press.

Scheffler, S. (2005). Choice, circumstance, and the value of equality. Politics, Philosophy & Economics, 4(1), 5-28.

Scott J. Boylan & Geoffrey B. Sprinkle (2001) Experimental Evidence on the Relation between Tax Rates and Compliance: The Effect of Earned vs. Endowed Income. The Journal of the American Taxation Association: Spring 2001, Vol. 23, No. 1, pp. 75-90.

Sheshinski, E. (1972). The Optimal Linear Income-tax. The Review of Economic Studies, 39(3), 297.

Spicer, M., Becker, L.A. (1980). Fiscal Inequity and Tax Evasion: An Experimental Appoach. National Tax Journal 33(2), 171-5.

Svenson, O. (1981). Are we all less risky and more skillful than our fellow drivers? Acta Psychologica, 47(2), 143-148. http://dx.doi.org/10.1016/0001-6918(81)90005-6 Tekeli, R. (2011). The Determinants of Tax Morale: the Effects of Cultural Differences and

Politics. Policy Research Institute, Ministry of Finance of Japan, PRI Discussion Paper

Series, (11A-1).

Torgler, B., & Schneider, F. (2004). Does Culture Influence Tax Morale? Evidence from Different European Countries. Science, 0–38. http://doi.org/10.1016/j.joep.2005.09.002. Tuomala, M. (1990). Optimal income tax and redistribution. Oxford: Clarendon Press.

(34)

APPENDIX

TABLE 6: DESCRIPTION OF VAR IABLES Variable Name Description

Gender A binary variable taking the value of 1 if the participant is female Score Test score in terms of experimental units (1 correct answer = 10

experimental units) Percentage

Compliance

A continuous variable indicating the percentage of earned income declared Percentage

Compliance Reduced

A continuous variable indicating the percentage of earned income declared, excluding 0% and 100% declarations

Fully Complied A binary variable taking the value of 1 if the participant declared all their experimental income

Partially Complied

A binary variable taking the value of 1 if the participant declared a non-zero income

Task Perception Numerical response to the question “On a scale of 1 to 10, to what extent do you feel the exercise was a good test of ability (compared to effort)?” Relative

Performance

Numerical response to the question “How do you think you performed relative to other participants on this task?”, where responses were presented

(35)

as discrete deciles ranging from “Bottom 10%” to “Top 10%”. Income

Equality

Numerical response between 1 and 10 inclusive, where 1 indicated “Incomes should be made more equal” and 10 indicated “We need larger income differences as incentives for individual effort”

Tax Cheating

Numerical response between 1 and 10 inclusive, where 1 indicated

“Cheating on taxes if you have a chance is never justified” and 10 indicated “Cheating on taxes if you have a chance is always justified”

TABLE 7: SUMMARY STATISTICS BY TREATMENT Treatment

Overall Group 1: base Group 2: ability Group 3: effort Average age 25.12 (5.26) 24.05 (2.19) 25.63 (6.36) 25.38 (5.40) Male 34 11 19 4 Female 42 11 19 12 Female % 55% 50% 50% 75% Score 148.95 (29.28) 135.45 (24.44) 153.42 (30.34) 156.88 (28.22) Percentage Compliance 70% (38.78) 73% (39.81) 61% (40.64) 85% (27.69) Percentage Compliance reduced 55% 51% 54% 66% Fully Complied 50% 59% 37% 69% Partially Complied 86% 86% 82% 94% Task Perception 6.80 (2.26) 6.23 (2.76) 7.24 (2.03) 6.56 (1.90) Relative Performance 6.67 (1.68) 6.55 (1.99) 6.68 (1.65) 6.81 (1.33) Income 4.8 5.18 4.71 4.5

(36)

Equality (2.10) (2.17) (2.13) (2.00) Tax Cheating 2.87 (1.96) 2.86 (1.86) 2.92 (1.98) 2.75 (2.18)

Figures in parentheses indicate standard errors

EXPERIMENT INSTRUCTIONS

NB: the text in bold varied by treatment. Bold text has been used here for clarity, but in the instructions viewed by participants the text was not distinguishable from the rest of the text. Specific text for each treatment can be seen below.

Instructions Baseline

Treatment

You will receive 10 experimental units for every question answered correctly, and there will be one example question before the task starts. Ability

Treatment

Your achievement in these questions is strongly determined by your ability. You will receive 10 experimental units for every question answered

correctly, and there will be one example question before the task starts. Effort Treatment Your achievement in these questions is strongly determined by the effort you

put in. You will receive 10 experimental units for every question answered correctly, and there will be one example question before the task starts.

Instructions

Please read the following instructions carefully. Once you have finished reading the instructions you will be asked to answer two simple questions to ensure you have read them in full.

Ability vs. hard work : how does our perception of success affect tax compliance?