Could educational communication successfully tackle tax evasion? : evidence from a randomized survey experiment in Greece

(1)

1 University of Amsterdam,

Master in Economics (Public Policy Track)

Could Educational Communication Successfully Tackle Tax Evasion? Evidence from a Randomized Survey Experiment in Greece

by Christos Politis (11679131) July 2018

Supervisor: Dirk Veestraeten Second reader: Naomi Leefmans

Abstract

The purpose of the paper is to empirically determine whether educational communication could successfully tackle tax evasion in Greece. A randomized online survey experiment was designed to examine the persuasive effect of an educational video including information about tax rates, public spending, the societal cost of tax evasion and the detection risk. The results provide evidence that the video positively influenced participants’ attitude to their own tax evasion and their perception on fairness of the tax system. However, the video reduced the taxpayers’ perceived probability of undergoing a tax inspection. These findings are interesting since many researchers have established a connection between attitudes and behavior.

(2)

2

Statement of Originality

This document is written by Student Christos Politis who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

1. Introduction

Tax evasion represents a huge problem in many countries around the globe, which limits their capacity to raise revenues and implement important economic and social policies. In the European Union, the revenue loss due to tax evasion was estimated to have exceeded €860 billion in 2009, an amount similar to its total healthcare expenditures (Murphy, 2012). In search of a solution, greater enforcement efforts (e.g. increasing the number of audits) may not always be cost-effective or even be interpreted as overly oppressive (Hasseldine et al., 2007). One alternative way to reduce tax evasion might be to promote the intrinsic motivation for paying taxes through the use of persuasive communications (Leder et al., 2010; Adams, 2010).

Previous work has revealed that many people have insufficient knowledge about tax regulations and unrealistic perceptions about the tax system (Roberts et al., 1994; Eriksen and Fallan, 1996). Thus, during the past few decades, many researchers have explored the persuasive effect of educational communications (Violete, 1989; Wartick, 1994; Blumenthal et al., 2001; Bott et al., 2017). The intuition behind is that enhancing the level of general fiscal knowledge (i.e. knowledge about tax rates, changes in tax laws, the usage of tax payments, etc) might improve tax compliance as a result of a more positive perception about taxation (Jackson and Milliron, 1986). However, the results reported regarding the success of such educational communications are mixed. One potential explanation for this inconsistency among the results is the different research designs (Hasseldine et al., 2007; Adams, 2010).

In search of what makes educational communication effective, the current study firstly reviews the main approaches used in the existing literature in terms of: educational content, delivery mechanism, target groups and wording choices. Afterwards, based on all the best practices, a new education communication was created by means of a video. The title of the video is “Introduction to Public Economics” and it includes information about tax rates, public spending, the societal cost of tax evasion and the detection risk.

(6)

6 The paper proceeds by designing and implementing a randomized online survey experiment on a sample of 112 adult citizens. Based on their demographic characteristics, all participants were randomly assigned to either control or treatment group. The educational video was shown only to the treatment group, while participants in both groups were asked to fill in a questionnaire about their attitudes, beliefs and preferences on tax evasion and other tax-related issues. Finally, all the observations were analyzed using the Principal Component Analysis (PCA), the Mean Comparison technique and Ordinary Least Squares (OLS) regressions.

This experiment differs from prior research in three ways: (i) the use of an educational communication that was built according to practices indicated as the “best” by the existing literature, rather than arbitrary choices of the author; (ii) the implementation in Greece, a country where no similar experiment ever carried out and where tax evasion has often been called the “national pastime”; (iii) the effectiveness of the video was measured in a variety of ways, including also two newly proposed: the motivation to request for a VAT receipt and the support for policies against tax evasion.

The results indicated that the educational video was successful in influencing participants to have a stricter attitude on their own tax evasion and a more positive perception of the tax system fairness. However, it failed to significantly impact on people’s policy preferences, motivation to ask for VAT receipts and attitudes towards other people's tax evasion. Lastly, the video reduced the taxpayers’ perceived probability of undergoing a tax inspection. These results are interesting since prior research suggests that an individual’s actual tax evading behavior can be influenced by his/her tax attitudes and his/her perceptions of the risk to be detected (Carnes and Englebrecht, 1995; Eriksen and Fallan, 1996; Adams, 2010).

The remaining parts of this paper are organized as follows. Section 2 reviews existing literature. Section 3 describes the design and implementation of the online survey experiment. Section 4 presents the data analysis and the results found, and Section 5 concludes.

(7)

7

2. Literature Review

In the first major review of the literature on tax evasion, Jackson and Milliron (1986) mentioned education as one of the 14 main determinants of tax evasion. Many studies have examined the impact of the taxpayer’s general level of education on tax evasion (Song and Yarbrough, 1978; Witte and Woodbury, 1985; Kinsey and Grasmick, 1993), while others focused on the taxpayer’s specific fiscal knowledge (Eriksen and Fallan, 1996; Chen Loo et al., 2009). Amongst the second category of studies, the most prevalent research method is that of carrying out experiments that measure the effectiveness of educational communication (Hasseldine et al., 2007). In order to precisely define the term “educational communication”, one should first take a step back and examine a wider term, called “persuasive communication”.

2.1. Persuasive communication

Persuasive communication is “any message that is intended to shape, reinforce, or change the responses of another or others” (Miller, 1980). In the field of tackling tax evasion, persuasive communication is used as an alternative approach for encouraging compliant tax reporting, and it can be implemented in three ways: threats of sanctions, appeal to moral principles and educational communications (Hasseldine et al., 2007) (see Figure 2.1).

Figure 2.1 The Three Dimensions of Persuasive Communication

“Threats of sanctions” are explicit audit threats and warnings of formal and informal personal consequences (Hasseldine et al., 2007; Slemrod, 2016). Formal consequences include fines and other monetary sanctions or even a jail sentence,

Persuasive communication

Threats of sanctions Appeal to moral principles

Educational communications

(8)

8 while informal consequences include means of social disapproval, like reporting the names of people who are evading taxes in a local newspaper (Schwartz and Orleans, 1967).

“Appeal to moral principles” or to conscience, are straightforward normative statements which indicate why one should be compliant; that is, not to engage in tax evading behavior (Hasseldine et al., 2007; Oceja et al., 2015). A typical example of such a persuasive message can be found in Kaplan et al. (1997). They stated that the “voluntary compliance approach is consistent with the American ideal that democracy is a government of the people and by the people”.

“Educational communications” are informational messages which aim to combat potentially unrealistic conceptions people might have about tax evasion, or even the tax system and the public sector in general. For instance, they can include information about the tax progressivity, the distributional effect of public spending, the societal costs of tax evasion, the number of audits taking place and the existing fines (Violete, 1989; Eriksen and Fallan, 1996; Roberts, 1994; Blumenthal et al., 2001). Interestingly, there are a number of cases that different categories of persuasive communication refer to similar topics, like audits and public spending. However, the specific content and phrasing are distinct; that is, educational communications always use an informative and more impersonal writing style (see Table 2.1).

In the existing literature, the persuasive effect of more than one category/dimension of persuasive communication is often tested in a particular study (Slemrod, 2016). For example, in the seminal study of persuasive communications conducted by Schwartz and Orleans (1967), one group of participants was exposed to the communication of existing sanctions (i.e. educational communication), while a second group was exposed to the communication of an appeal to conscience. Since the paper at hand focuses exclusively on the educational communications, the literature review in the following pages is concerned only with the educational part of those studies.

(9)

9 Table 2.1 Disambiguation of the Persuasive Communication’s Dimensions

Topic Communication Example from the literature

The increased number of audits Threats for sanctions

“The Inland Revenue is substantially increasing the number of enquiries into the tax returns of people who have previously reported a turnover of just below £15,000 in successive years. Your 2001 return may be one of those chosen for enquiry.” (Hasseldine et al., 2007)

Educational communication

“The 1992 IRS Annual Report reflects that criminal investigations of tax crimes are on rice. In 1992 alone, the Criminal Investigation Division indicates about 3,500 individuals for tax evasion…” (Kaplan et al., 1997)

The usage of tax payments

Appeal to conscience

“Your tax payment contributes to the funding of publicly financed services in education, health and other important sectors.” (Bott et al., 2017)

Educational communication

“Over 30 percent of state taxes go to support education. Another 18 percent is spent on health care and support for the elderly and the needy.” (Blumenthal et al., 2001)

2.2. The Effectiveness of Educational Communications

Regarding the question of whether educational communications are able to reduce tax evasion, the existing literature has not arrived at a clear-cut answer. There are some studies that find a positive and statistically significant treatment effect (Schwartz and Orleans, 1967; Violete, 1989; Roberts, 1994; Wartick, 1994; Eriksen and Fallan, 1996), while others find no effect at all (Blumenthal et al., 2001; Dwenger at al., 2016; Bott et al., 2017; Bergolo et al., 2017) or even a negative effect (McGraw and Scholz, 1991).

One potential explanation for this disagreement among the results is the different nature of the educational messages created by each author. For example, Roberts (1994) created educational communications by means of a video including information about tax rates and the distribution of tax burdens, while Blumenthal et

(10)

10 al. (2001) sent letters to taxpayers informing them about how the tax revenues are allocated amongst the various state services. Features like the aspects of fiscal knowledge included in each message (i.e. the specific information shared) and the delivery mechanism selected (e.g. letter or video), are supposed to be important parameters of the success of an educational communication (Richardson and Sawyer, 2001; Hasseldine et al., 2007). Similarly, the characteristics of the sample researched (i.e. the target group) and any direct or indirect wording differences, might also be crucial (Hasseldine et al., 2007).

An additional potential explanation might be the very nature of tax evasion. Tax evasion is a so-called “hidden phenomenon”, which means that people do it only in secret and it is not officially recorded. Consequently, most of the experiments conducted in this field had to use other variables as a proxy for the people's actual tax evading behavior. For example, some authors measured the people’s tax ethics (McGraw and Scholz, 1991) or the perceived tax fairness (Maroney et al., 2002), while others measured the people’s tax evading intentions (Kaplan et al., 1997) or their actual tax reporting (Bergolo et al., 2017). This diversity between the methods used in different studies also possibly contributed to the ambivalence of the results.

The rest of Chapter 2 is concerned with these two main propositions. In search of what makes an educational communication effective, their specific characteristics will be elaborated base on the existing literature (Subchapters 2.3.-2.6.). Afterwards, the methods available on measuring the persuasive outcomes will be thoroughly analyzed (Subchapter 2.7.).

2.3. Educational Content

During the examination of existing literature, one can find mainly eight aspects of fiscal knowledge being included in educational communications: tax rates, public spending, societal cost of tax evasion, sanctions, other introductory information, tax laws, tax evading opportunities and any other positive attribute of the tax system or of the public sector (see figure 2.2).

(11)

11 Figure 2.2 The Various Aspects of Tax Knowledge that an Educational

Communication may Include

2.3.1. Tax Rates

Previous work has revealed that many people have unrealistic perceptions on the tax system, for instance of the taxes they actually pay and the concept of progressive taxation (Eriksen and Fallan, 1996). Also, it is found that people’s intrinsic motivation to pay taxes can be low due to this misperception; that is, when people blame the government for tax rates that are too high or too small share in public goods and services (Leder et al., 2010).

Educational communication on tax rates usually includes information that may improve people’s perceptions of the tax system’s fairness. For instance, the actual tax rates, the tax burden and its distributional effects, the progressivity of income tax rates, the intuition behind the structure of the tax system or behind imposing/changing any specific tax (e.g. property tax) and any differences between the domestic tax system and those of other countries (Roberts, 1994; Maroney et al., 2002; Wartick, 1994; McGraw and Scholz, 1991).

When building educational messages about tax rates, one could possibly strengthen the persuasive effect even more, by taking into account all the dimensions of tax fairness. Jackson and Milliron (1986) pointed out: (i) the exchange equity; that is, the

Educational content Tax rates Public spending Societal cost of tax evasion Sanctions

Tax Laws and Tax Evasion Opportunities Introductory information Other Positive Attributes

(12)

12 benefits received for the tax dollars given, (ii) the vertical equity; that is, the equity across different income groups, (iii) the horizontal equity; that is, the equity within a taxpayer’s income group.

2.3.2. Public Spending

According to a study conducted by Kirchler (1997), people feel that the utility received from public goods and services is at level lower than desirable for all the areas of public spending (public health, infrastructure, education, etc). On the other hand, Leder et al. (2010) measured the perceived fiscal exchange and found that people were highly misinformed about the actual percentage of taxes distributed as public goods and services; people thought that it was only between 30% and 35%1.

Educational communication on public spending could include information about the wide range and variety of public goods and services, the use of taxes for financing them, social security and welfare, its distributional effects, the level of public spending and the percentage of it devoted to each sector (health, education, defense, etc.) and any differences in public spending across countries through the prism of social objectives. This information could potentially strengthen people’s perception of fiscal exchange and help them to understand the importance of paying taxes (Leder et al., 2010).

In addition, one strategy that could possibly increase the persuasive effect of this information is to focus on public goods and services that have been shown to be important to most taxpayers. In other words, to avoid mentioning issues that may be considered undesirable/unpopular by some taxpayers like public spending for military purposes (Falsetta et al., 2015).

2.3.3. Consequences to Society

Educational communication on the societal cost of tax evasion usually includes information about the revenue loss (and thus the fewer funds available to the

1

The study by Leder et al. (2010) was conducted in Italy, where no information about the actual percentage is available. However, in other countries like Austria, about 50 percent of the tax payments

(13)

13 government), the public goods and services that could be provided if evasion cease, the larger budget deficit that necessitates higher tax rates and larger interest payments, the widening of income inequality and social injustice and the creation of imbalances in the economy as a result of the distortion of competition between businesses (Kaplan et al., 1997; Ernst and Young, 2016).

According to Kaplan et al. (1997), information about the societal cost of tax evasion (as well as the information about tax rates and public spending) could encourage compliant behavior of individuals with a higher capacity for moral reasoning; that is, people who “tend to focus beyond self and consider the consequences of their actions to others and society”. Empirically, Kaplan et al. (1997) found that knowledge about the fairness of the tax system and the societal cost of tax evasion can influence the tax attitudes (although not the tax intentions)2 of those taxpayers who utilize relatively high moral reasoning. On the other hand, taxpayers with relatively low moral reasoning were not significantly influenced.

2.3.4. Sanctions

Taxpayers with a lower capacity for moral reasoning; that is, people who “tend to focus on the self and are primarily motivated by avoiding punishment and acquiring individual benefits” (Kaplan et al., 1997), are supposed to be unaffected by educational communications related to tax rates, public spending and the societal cost of tax evasion. Empirically, Kaplan et al. (1997) found that these individuals can be influenced by legal sanctions communications.3

Broadly speaking, educational communication on the penalties for tax evasion usually includes information about formal and/or informal sanctions, the numbers and kinds of audits taking place per year, the average detection rate, the available information sources and specific methods used for detecting possibly evaded taxes and compliance rewards in the form of social recognition. Authors like Schwartz and

2_{Intention is the person’s “plans on performing the behavior” while attitudes are the person’s} “evaluations, either positive or negative, of performing the behavior” (Adams, 2010). See subchapter

2.7. 3

Legal (or formal) sanctions include fines and other monetary penalties or even a jail sentence, while informal sanctions include means of social disapproval, like reporting the names of people who are

(14)

14 Orleans (1967), Violete (1989), Kaplan et al. (1997), Dwenger et al. (2016), Bergolo et al. (2017) included similar information in their educational messages and the results found were mixed.

Some potential explanations about this ambiguity are as follows. Firstly, the penalties for tax evasion or the likelihood of detection are different from case to case, depending on how much evasion is attempted, what kind of evasion (e.g. hiding income or overestimating expenses) and so on (Slemrod, 2016). As a result, a policy maker would find it difficult to come up with an educational message including information relevant to each taxpayer. Secondly, according to Violete (1989), only the legal sanctions information can lead to significantly greater tax compliance, while the informal sanctions information cannot. Lastly, some sanction messages might have produced a backlash effect resulting in stronger tax evading attitudes (Roberts, 1994). This opposite-than-expected effect could possibly be attributed to evidence from the Resistance literature, which will be elaborately examined in Subchapter 2.6.

2.3.5. Tax Laws and Tax Evasion Opportunities

In one study, Eriksen and Fallan (1996) examined the effectiveness of a rules-oriented tax course which mainly included information on how to calculate tax liabilities. Their results indicated a significant influence on the students’ perceptions of the tax system fairness. However, caution is needed as this kind of information can have a side effect. Tax-payers with more knowledge of the tax laws might be more inclined to evade (Jackson and Milliron, 1986), due to a better understanding of the existing tax evasion opportunities.

McGraw and Scholz (1991) experimented on the impact of communication of tax evasion opportunities. They created a video including information about aggressive strategies for reducing tax liability and found that the taxpayers who viewed the video were more likely to try minimizing their tax liabilities.

(15)

15

2.3.6. Introductory Information

Kuziemko et al. (2015) expressed the opinion that educational communication should include simple and easy-to-understand information. Information treatments should aim on providing a large “educational shock” to individuals, rather than a PhD-level discussion. One could incorporate in every educational message some introductory information about the tax system. Such information could be: the definition of direct and indirect taxes, an overview of the goals and the functioning of the tax system, the definition of tax evasion, a short description of the governmental bodies responsible for collecting revenues and detecting tax evasion and the approximate level of tax evasion domestically.

2.3.7. Other Positive Attributes

Often, people who are disappointed with one part of the government system (e.g. the quality of education or the high tuition fees), tend to hold a similarly negative stance on other parts of the government system (e.g. taxation) (Ahmed and Braithwaite, 2005). As a result, an educational communication aiming to boost tax compliance could refer to any issue, regardless of its relevance to the tax system or any financial issues, which could potentially enhance the relationship between the government and its own citizens.

2.4. Delivery Mechanism

Until today, most of the experiments that tried to influence tax compliance through the use of educational messages relied on the use of letters as the main delivery mechanism (Bergolo et al., 2017; Bott et al., 2017; Dwenger at al., 2016; Blumenthal et al., 2001). However, alternative approaches were also used: 30-second televised announcements (Roberts, 1994), 30-minute videotapes (McGraw and Scholz, 1991), university courses (Eriksen and Fallan, 1996), simple text reading (Kaplan et al., 1997; Wartick, 1994; Violete, 1989), etc.

As indicated by Ortega and Scartascini (2015), the effectiveness of persuasive communication can be significantly different due to the specific delivery method selected. In their study, about 20,000 taxpayers were contacted by the state tax agency

(16)

16 either by letter, email or personal visit by a tax inspector. According to their findings, personal visits by a tax inspector are much more effective than a letter or an email.

However, Roberts (1994) suggested that a good delivery mechanism should also be able to reach large segments of the taxpaying population. Methods like personal visits by a tax inspector or university courses might be too costly; while other methods, like letters and emails, might be hardly understood by poorly educated or older people. Instead, a videotaped message offers the potential to communicate with people who may not be reached otherwise (Roberts, 1994).

Last but not least, a crucial part of the delivery mechanism could be considered the "originator" of the message. According to Roberts (1994), people might be skeptical of information and arguments offered by the tax authorities themselves. Instead, they might be more willing to consider factual information presented by a neutral originator. Collaboration with universities or other educational bodies could increase the credibility to the messages, which is necessary to influence taxpayers.

2.5. Target Group

Apart from the content and the delivery mechanism, the characteristics of the targeted population could also affect the success of persuasive communication (Hasseldine et al., 2007). For example, there is evidence that fiscal-knowledge treatments could be more successful with younger taxpayers than with people who have more experience with the tax system. The first years of one’s taxpaying life is said to be a critical period in the formation of tax attitudes. As a result, the younger taxpayers’ attitudes may be less resistant to change (Roberts, 1994). In addition, pensioners and salaried people have only little opportunity for noncompliance, while this is not true for the self-employed (Hasseldine et al., 2007). Thus, actual educational campaigns, as well as relevant experiments, should preferably focus on taxpaying groups of special interest (Kaplan et al., 1997).

Finally, the taxpayers’ receptivity might differ among countries, where the magnitude and nature of tax evasion, the tax morale and the institutional environment also differ (Slemrod, 2016). For example, Blumenthal et al. (2001) and Wartick (1994)

(17)

17 conducted their experiment in the USA, while Bott et al. (2017) in Norway, Dwenger at al. (2016) in Germany and Bergolo et al. (2017) in Uruguay. According to Slemrod (2016), some interventions work better than others due to the specific characteristics of the countries they took place in.

2.6. Specific Wording

There are few educational communications that share the same topic, or even the same piece of information, but still they express/phrase it in a fundamentally different way. For example, information about the societal cost of tax evasion can be given both in terms of current revenue loss; and revenues that can be collected if tax evasion is eliminated (see Table 2.2). These differences in wording are considered to be possibly affecting whether compliance occurs or not (Hasseldine et al., 2007).

Table 2.2 Examples of Wording Differences

Topic Specific focus Example from the literature

The societal cost of tax

evasion

on losses

“Revenues lost through tax evasion mean fewer funds to support education, medical research, human services…” (Kaplan et al., 1997)

on gains

“If those who currently evade their tax obligations were evade 10% less, the additional revenue collected would enable all of the following: to supply 42,000 portable computers to school children; to build 4 high schools…” (Bergolo et al., 2017)

People’s actual tax

reporting

in positive terms

“The great majority report information about their income and assets in Norway correctly and completely.” (Blumenthal et al., 2001)

in negative terms

“…a small number of taxpayers who deliberately cheat owe the bulk of unpaid taxes.” (Bott et al., 2017)

According to Hallsworth (2014), one could enhance the effectiveness of persuasive communication by paying more attention to evidence from behavioral science, such as

(18)

18 framing. Thus, the rest of this Subchapter will offer a short review of some behavioral-related theories.

2.6.1. Framing Literature

Framing refers to “selecting and highlighting some elements of reality and suppressing others, in a way that constructs a story about a social problem, its causes, its moral nature and its possible remedies” (Entman, 1995).

Levin et al. (1998) described three categories of framing manipulations. Firstly, according to “risky choice framing”, people are more likely to take risks when options emphasize on the likelihood of avoiding losses rather than realizing gains. Secondly, according to “attribute framing”, attributes are evaluated more favorably when expressed in more positive terms than in negative terms. For example, Levin and Gaeth (1988) showed that a sample of people had better perceptions of the quality of meat when it was labeled as “75% lean” rather than “25% fat”. Thirdly, according to “goal framing”, when a message focuses on losses, it tends to have a greater impact on people’s behavior than a comparable message that focuses on gains. A typical example is that of breast self-examination. Meyerowitz and Chaiken (1987) showed that women were more motivated to perform the test when presented with information focusing on the negative consequences of not performing it, rather than when focusing on the benefits of performing it.

As a result, based on the framing literature, one could say that Kaplan et al. (1997) and Blumenthal et al. (2001) made better wording choices than Bergolo et al. (2017) and Bott et al. (2017) respectively (see Table 2.2).

2.6.2. Resistance Literature

Even though the strength of framing effects has been documented in a variety of studies (Fatemi et al., 2008), Levin et al. (1998) raised concerns that under some circumstances, these effects could be mitigated. If there is a considerable distance between one’s prior beliefs and the position advocated in the message, then the included information might be viewed as too extreme for serious consideration. Even

(19)

19 worse, the message might be considered as propagandistic and the original counter belief might be strengthened (Eagly and Chaiken, 1993). Also, according to the resistance literature, in cases that the persuasive communication is conceived as highly threatening, people will hold attitudes in a direction opposite than those advocated in the message (the so-called “boomerang effect”) (Eagly and Chaiken, 1993).

2.6.3. Public Opinion is Group-Centric

The public opinion on government policy is said to be group-centric; that is, “strongly influenced by the attitudes citizens possess toward the social groups perceived as the beneficiaries (or victims) of the policy” (Nelson and Kinder, 1996, p. 1055). For example, in one study, Bullock and Fernald (2005) found that when the elimination of the dividend tax was framed as being advantageous to the wealthy rather than to the general public, people were less willing to support it. They also found that agreeing to the elimination of the dividend tax was negatively correlated with attributing wealth to corruption and positively correlated with considering wealthy as deserving.

2.6.4. Cognitive or Affective Information

According to Roberts (1994), educational communication (and especially a video) can be built either in a cognitive or an affective approach. The first is more direct and informational as it includes pie charts, bar graphs and a narration. The second is based on role-play situations involving taxpayers and includes the use of empathy and humor. Although psychology research has indicated the importance of both approaches, Roberts (1994) found that the cognitive televised announcements were more effective in influencing attitudes towards the tax system fairness.

2.7. Measuring the Impact

In addition to what makes educational communication successful, one should also decide on the criteria of measuring the success. Given the aim of tackling tax evasion, the measure of an individual’s actual tax-evading behavior would have been the first-best choice. For example, in the context of a local church tax in Germany, Dwenger et al. (2016) had the opportunity to estimate the taxpayers’ true tax liability and thus,

(20)

20 using administrative records of taxes paid, to distinguish those who were not compliant. However, this is the exception rather than the rule. In most cases, researchers could not observe the actual tax evading behavior. As a result they had to use other alternative approaches.

According to the theory of reasoned action, the main determinant of actual behavior is intention; and the determinants of intention are attitudes and subjective norms (Adams, 2010). Intention is the person’s “plans on performing the behavior” while attitudes are the person’s “evaluations, either positive or negative, of performing the behavior” (Adams, 2010). Empirically, there is also evidence that attitudes and subjective norms are both correlated with intention and that intention is correlated with behavior (Adams, 2010). With these findings, Schwartz and Orleans (1967), McGraw and Scholz (1991), Roberts (1994) and Eriksen and Fallan (1996) examined the effects of educational communications on people’s attitudes towards taxation (i.e. tax attitudes); while Kaplan et al. (1997) and Violete (1989) examined the effects on people’s tax intentions.

In practice, Eriksen and Fallan (1996) defined tax attitudes as a multidimensional concept, consisted out of: (i) the attitude to other people's tax evasion; (ii) the attitude to one's own tax evasion (i.e. the tax ethics); (iii) and the perceived tax fairness. Additionally, Eriksen and Fallan (1996) measured the people’s attitude to other illegalities in order to examine whether educational communication influences only tax attitudes or attitudes in general (see table 2.3).

On the other hand, in order to measure the tax intentions, taxpayers were presented to a tax scenario and were asked to indicate the amounts they would evade (see table 2.4).

(21)

21 Table 2.3 The Four Dimensions of Tax Attitudes

Example of questions asked

Attitude to other people's tax evasion

Rate the degree to which you agree or disagree with the following statement: "One cannot criticize a person who declares a lower income than was the case on his/her tax return when there are so many others doing exactly the same." (Eriksen and Fallan, 1996)

Tax ethics

“How much of a moral obligation – that is, an obligation based on your own personal feelings of what’s right or wrong – do you feel to be completely honest in filling out your tax return?” (McGraw and Scholz, 1991)

Perceived tax fairness

“Overall, how high or low are the income taxes Americans pay compared to the benefits and services we receive from the federal government?” (Roberts, 1994)

Attitude to other illegalities

“How serious is the illegality of stealing a wallet containing NOK 1000?” (Eriksen and Fallan, 1996)

Table 2.4 Tax Scenario

Example from the literature

“The Smiths purchased a new computer system for use in their business. The old system was sold to a friend as a home computer for $1,600 in cash. The Smiths fully depreciated the old computer system on prior tax return, and they are aware that the $1,600 represents a taxable gain. Which of the following amounts of the $1,600 gain do you think is closest to the amount you would not report if you were in the same situation?” (Kaplan et al., 1997)

In addition to (or instead of) the tax attitudes and tax intentions, researchers like Schwartz and Orleans (1967), McGraw and Scholz (1991), Blumenthal et al. (2001), Bergolo et al. (2017) and Bott et al. (2017) used an alternative approach based on the people’s actual tax reporting. This is a “difference in difference” method where the change in taxes paid between two consecutive years (before and after the treatment) is compared with those exposed to the educational communication versus those who didn’t.

(22)

22 Lastly, Bergolo et al. (2017) used the perceived probability of detection as a dependent variable in their analysis. There is evidence that an individual’s tax compliance decisions can be influenced not just from the actual detection risk, but also from his/her perceptions of the probability to be detected (Carnes and Englebrecht, 1995).

3. Design of the experiment

The purpose of the paper is to extend the existing tax compliance research and empirically determine whether educational communication could successfully tackle tax evasion. In order to identify such a causal effect, a randomized online survey experiment was implemented in Greece, from the 25th of June until the 3rd of July 2018. The structure of the experiment is as follows: (i) a few demographic questions about age, gender, education, occupation and income; (ii) the educational communication shown only to the treatment group; (iii) questions on tax ethics, tax fairness and other measures; (iv) a couple of tax knowledge questions (see Figure 3.1). The educational treatment, questions asked, data collection, sample and other validity and reliability issues are discussed below.

3.1. Educational treatment

An education communication was created by means of a video, based on the best practices indicated by the existing literature as described in Chapter 2. It includes information about the tax rates, public spending, the societal cost of tax evasion and

Demographic characteristics Random assignment to groups Questionnaire Control group Treatment group Figure 3.1 Overview of the

(23)

23 the detection risk, plus some introductory information about the tax system and tax evasion itself. However, educational content about tax laws and tax evasion opportunities are not included as they are prone to fostering tax noncompliance (see Subchapter 2.3.).

The video was selected as the most appropriate delivery mechanism due to its very characteristics. A video is relatively cheap to build, less demanding in terms of educational level, more popular and attention catching amongst youngsters. Also, it can be reproduced both in television and internet (including social media) in a potential real policy intervention, reaching larger segments of the population (see Subchapter 2.4.).

The communication’s length was selected to be about 5 minutes. It was believed that a very long video, like the 30-minute videotapes created by McGraw and Scholz (1991), would create some bias to our results due to possible fatigue. When the survey participants become bored or tired, they might begin to underperform, providing less accurate data or skipping some parts of the video (i.e. partial compliance to the treatment). On the other hand, an overly short video, like the 30-second televised announcements created by Roberts (1994), would be very difficult to contain all the important information and to significantly affect people's tax compliance behavior.

The title of the video was “Introduction to Public Economics”. It was believed that such general title (i.e. which is not focusing on tax evasion) would mitigate the risk of being perceived as manipulative by the public. The main body of the video was built according to the cognitive approach that includes tables, bar graphs, maps and a narration. The specific wording techniques indicated in Subchapter 2.6 were also considered. At the last part, the logo of the “University of Amsterdam” was also displayed, aiming to enhance the message's credibility. A copy of the narrated text (translated in English) is provided in the Appendix I, while one can watch the full video on Youtube4.

(24)

24

3.2. Dependent Measures

Among all the approaches of measuring the impact of an educational communication that were used in previous researches (see Chapter 2.7.), the current study focuses only on two: the tax attitudes (in all its dimensions) and the perceived probability of detection. The other approaches could not be implemented, either due to shortage of data on the participants’ actual tax reporting, or concerns for raising survey fatigue; as it is relatively time-consuming for people to read tax scenarios before giving information on their tax intentions.

The following 5 hypotheses were formulated and tested:

H1:Attitudes to other people's tax evasion will be significantly stricter among those participants who receive an educational communication than those who did not.

H2:Attitudes to one's own tax evasion will be significantly stricter among those participants who receive an educational communication than those who did not.

H3:Attitudes towards tax fairness will be significantly more positive among those participants who receive an educational communication than those who did not.

H4:Attitudes to other illegalities will not differ among those participants who receive an educational communication than those who did not.5

H5:Perceived probability of detection will be significantly higher among those participants who receive an educational communication than those who did not.

In addition to the tax attitudes and the perceived probability of detection, two additional approaches are newly proposed and implemented. According to what is seen so far, educational messages aim to increase people's intrinsic motivation to comply with the tax laws and pay the full amount of their tax obligation. However, there are grounds to believe that people can contribute to the battle against tax evasion in more ways, and not only through tax compliance.

5

Consistent with Eriksen and Fallan (1996), the attitude to other illegalities is measured in order to examine whether educational communication influences only tax attitudes or attitudes in general (see

(25)

25 In most countries around the world, consumers have to pay the value added tax (VAT) on all goods and services they buy. The VAT is included in the price and thus collected by the seller. After the transaction is completed, the seller has to provide a VAT invoice or receipt to the consumer, and then the seller has to remit the VAT received to the country’s tax administration.

In practice, some sellers are deliberately not recording the transactions (they do not print VAT receipts) and thus, they are evading taxes. Moreover, although consumers have the right to ask for the VAT receipt (and thus to help the collection of taxes), they often disregard or overlook it. Potential explanations might be the low tax morale, and also a feeling of shame or laziness. If educational communication induces a better understanding of the seriousness of tax evasion, then the feeling of shame or laziness may be eased. Consumers would begin asking for VAT receipt and the tax evasion will be reduced.

An additional way to reduce tax evasion is altering people’s policy preferences. If people support government policies against tax evasion, or vote for political parties based on the party’s determination to tackle tax evasion, then politicians are expected to put greater effort on the battling against this social problem. In order to be elected, or to be re-elected, office seeking politicians might implement, or at least promise to implement, stricter measures against tax evaders. Therefore, examining the effect of educational communication on people’s policy preferences seems to be of primary interest too.

Finally, in the context of the present experiment, it is important to test whether the video indeed enhanced the participants' knowledge. As a result, based on the preceding discussion, the following 3 hypotheses were also formulated and tested:

H6:Motivation to request for a VAT receipt will be significantly higher among those participants who receive an educational communication than those who did not.

H7:Support for policies against tax evasion will be significantly higher among those participants who receive an educational communication than those who did not.

(26)

26 H8:Tax knowledge will be significantly improved among those participants who

receive an educational communication than those who did not.

To sum up, the current study measures the impact of an educational video on 8 variables: attitudes to other people's tax evasion (OTHER), tax ethics (ETHICS), perceived tax fairness (FAIRNESS), attitudes to other illegalities (CRIME), perceived probability of detection (DETECT), VAT receipt requesting (VAT), policy preferences (POLICY) and tax knowledge (KNOWLEDGE). For each of these variables, the participants were asked to answer a number of relevant questions. A copy of the questionnaire (translated in English) is provided in Appendix II.

3.3. Data Collection

To carry out the experiment, an online survey platform was created. Participants were invited to take part by clicking on an internet link, which was send to them through messages on social media or emails.

After reading an introductory page with instructions, the participants were asked to give their demographic characteristics. Prior research has found that some demographic characteristics (i.e. age, gender, education, occupation, and income) could be related to tax evasion (Jackson and Milliron, 1986; Kaplan et al., 1997). Based on these observables and according to the Minimization technique6, an algorithm randomly assigned each participant to either control or treatment group.

Also, several other steps were taken to ensure the reliability of the data collection. First of all, the participants were not informed about the real purposes of the study and the implementation of the experiment. The title of the online platform is “Opinion Survey: Tax Evasion in Greece", implying that the purpose of the study is simply opinion gathering. Such approach is imperative in order to avoid any experimental

6

According to Saghaei (2011), “With minimization the first subjects are enrolled randomly into one of groups. The subsequent subjects will be allocated to treatment groups after hypothetical allocation of each subject to every group, and then calculating an imbalance score. Using these imbalance scores, we

can decide to which group the new subject must be allocated, to have the minimum amount of imbalance, in terms of prognostic factors”. For more information, please refer to:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3317766/ while for an example of the method, please refer to: https://tinyurl.com/ya75eosr

(27)

27 (Hawthorne) effects; that is, an alteration of behavior by the subjects of the study, due to their awareness of taking part to an experiment (Adair, 1984).

Secondly, the introductory page included a text ensuring the anonymity and confidentiality of the answers of the participants. Given the sensitive nature of the compliance responses, participants should feel comfortable to submit truthful answers. Moreover, the introductory page included three notifications; (i) One can only participate in the survey once; (ii) You may be asked to watch a video related to the topic; (iii) It is necessary to use a computer, not a mobile phone.

The second and third notifications are parts of an extensive effort to reduce incidents of partial compliance; that is, cases that subjects assigned to the treatment group receive the treatment only partially. Running such an experiment online rather than in a laboratory, is less costly, yet one can have only limited control over the participants' behavior. Some participants might try to skip the video or watch a part of it, while others might not be very focused due to external disturbances like noise and conversations. These situations could introduce biases in the final results. For this reason, all participants were informed in advance that they may be asked to watch a video. They were also asked to take part in the survey only by computer, so as to increase the likelihood that people will watch the video at home, where the external disturbances are less. Moreover, in order to attract participants’ attention, the treatment was described as a video "containing information that many people don't know, about the tax system, public spending and tax evasion in Greece". Lastly, the subjects assigned to the treatment group were asked to confirm that they watched the whole video, before processing to the rest of the questionnaire.

Finally, it is important to mention that the participants were required to answer all the questions. However, they could exit the experiment at any time and thus, there it might be some survey attrition; that is, participants dropping out after being assigned to control and treatment group. This issue will be elaborately examined in the analysis part of the paper.

(28)

28

3.4. Sample

The sample of the present study includes 112 Greek adult citizens. It consists of 53 men and 59 women. A comparison of the sample to the Greek population (to the 2011 census) reveals that the participants in the sample are generally younger and better educated. Most of the participants are either salaried or retired, while most of them described their income as "sufficient to consume, but not much left for savings". Table 3.1 provides demographic information for the sample.

Table 3.1 Descriptive Statistics for Sample

Total 112 Gender Men 53 Women 59 Average age 34.733 Highest attained level of education Primary 1 Secondary 4 Tertiary 88 Master / PhD 19 Occupation Employed or retired 42 Self-employed 21 Student 38 Other 11 Income Insufficient 8 Hardly sufficient 35

Sufficient to consume, but not

much left for savings 52

I'm financially comfortable 17

Although the sample is not representative of the general population, this should not be a threat to the (internal) validity of the study. The larger and the more representative a sample is, the better for its external validity; that is, whether the research findings can be generalized to a larger population or not. However, when it comes to the internal validity (that is, whether causality can be established or not) of an

(29)

29 experiment, the sample does not have to be large nor representative. What is required is a successful random assignment to groups (Loewenstein, 1999). This issue will be elaborately examined in the analysis part of the paper at hand.

Finally, in order to better understand the sample researched, one should be aware of some basic facts about the country where the experiment took place, Greece. According to the World Economic Forum, Greece is amongst the countries where tax evasion is the most prominent (see Figure 3.2). Currently, its level estimated by DiaNeosis (a Greek think tank) is about 11 to 16 billion euros per year, which is equal to 6%-9% of the Greek Gross Domestic Product (GDP)7.

Figure 3.2 Executives' Assessment of how Important Tax Evasion is in their Country (the Lower the Measure the more Rampant is Tax Evasion)

Source: World Economic Forum (Global Competitiveness Report 2001/2002)

A study concluded that tax evasion in Greece is mainly due to a long-lasting mistrust between the Greek public sector and its citizens (Ballas and Tsoukas, 1998). Besides, during the last decade Greece suffered a great economic recession, losing about 25% of its GDP. A recent survey research by DiaNeosis (2018) suggests that; (i) the 98.2% of the population attributes the crisis to the inadequacy and corruption of the Greek

7_{For more information, please refer to: https://www.dianeosis.org/2016/06/tax_evasion_in_greece/} 0 1 2 3 4 5 6

(30)

30 governments; (ii) the 76.9% of the population thinks that tax evasion in Greece will never be eradicated8. Lastly, when it comes to the credibility of provision of information in Greece, the same survey suggests that the 80.7% of the population does not trust television, while about half (44.5%) does not trust internet.

4. Data Analysis and Results

According to the present survey, the participants assigned to the control group (that is, those who did not watch the video) are quite strict on other people's tax evasion, but rather lenient with their own tax evasion. Also, on average, they describe the tax system as unfair; have mixed opinions about the actual probability of detection; are motivated to ask VAT receipts; and consider tax evasion as an important political issue (see Figure 4.1). This Chapter discusses the effect of the educational video described in Chapter 3, on these attitudes and beliefs.

To begin with, the so-called Mean Comparison technique is to ensure that the assignment of subjects to groups (and thus, the demographic characteristics linked to tax compliance) is random. The null hypothesis that the percentage of men is the same in both groups (i.e. treatment and control group), was tested against the alternative hypothesis that the percentage of men is different. Similar tests were conducted for the rest of the demographic variables. This analysis, as illustrated in Table 4.1, indicates that P-values are always larger than 0.1. As a result, the null hypothesis cannot be rejected and it is believed that the randomization was successful. In other words, the Minimization algorithm (described in Chapter 3) which was responsible for randomly assigning each participant to either control or treatment group, is believed to be effective.

The experiment has a relatively high attrition9 rate of 55.7%, which includes those who remained online long enough to be assigned to either control or treatment group, but did not complete the questionnaire. As shown in Appendix III, the attrition is mainly random: 54.8% for the control group and 56.5% for the treatment group. Only

8

For more information, please refer to: https://www.dianeosis.org/2018/03/greeks-believe-2018/ 9_{In an experiment, there is “attrition” whenever participants are dropping out after being assigned to}

(31)

31 Figure 4.1 Examples of the Attitudes/Beliefs Held by the Control Group

0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10

a. It is a moral obligation for all citizens to

declare their exact income and pay the full amount of their taxes. (10= totally agree)

b. How serious would you consider the illegality

of not declaring an additional source of income, saving € 500 of taxes? (10= very serious)

c. Do you think that the distribution of tax burden

between high income citizens and low-income citizens is implemented in a fair way? (10= very fair)

d. How likely do you think it is to undergo a tax

inspection this year? (10= large likelihood)

e. The determination you believe a political party

will show against tax evasion, is it an important criterion for your voting decisions? (10= very

important)

f. When you encounter again the situation of not

getting a receipt, are you going to ask for one on your own initiative? (10= definitely yes)

(32)

32 Table 4.1 Comparison between the Means of the Demographic Variables,

for Those the Who Completed the Questionnaire

Control Treatment P-value

Total 56 56 Gender Men 27 26 0.853 Women 29 30 0.853 Average age 34.574 34.893 0.905 Highest attained level of education Primary 0 1 +∞ Secondary 3 1 +∞ Tertiary 44 44 1.000 Master / PhD 9 10 0.809 Occupation Employed or retired 20 22 0.703 Self-employed 11 10 0.816 Student 19 19 1.000 Other 6 5 0.769 Income Insufficient 5 3 0.542 Hardly sufficient 18 17 0.842 Sufficient to consume, but not much left for savings

25 27 0.710

I'm financially

comfortable 8 9 0.772

the percentages of the two higher income categories are found to be statistically different. However, the highest-income attritted people were mainly assigned to treatment group while the second-highest-income attritted people were mainly assigned to the control group. Thus, it does not seem that wealthy people systematically attrit from the experiment if they are assigned to the treatment group (see Appendix III).

The next step of the data analysis involves a Principal Component analysis (PCA); that is, a mathematical procedure that transforms a large set of variables into a smaller set, while still containing most of the initial information. This approach is essential

(33)

33 since the questionnaire included two or more questions for each of the dependent measures described in Chapter 3. Thus, two or more series of data had to be transformed each time into a single dependent variable. The PCA was implemented using always only the eigenvector with the highest eigenvalue.

The last part of the data analysis involves a series of OLS regressions, one for each of the 8 dependent variables. The regression equation includes; a dummy variable (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖) indicating whether the treatment was received or not; the set of control variables described in Chapter 3; an intercept (𝛽0) and an error (𝑢𝑖).

𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡𝑖 = 𝛽0+ 𝛽1∙ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 + ∑(𝛽1+𝑟 ∙ 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑟𝑖) + 𝑢𝑖 11

𝑟=1

The coefficient of interest is that of the first dummy variable (i.e. 𝛽₁) since it corresponds to the average treatment effect; that is the effect that the educational video had on the dependent variable. The null hypothesis that 𝛽1 is equal to zero was tested against the alternative hypothesis that 𝛽₁ is different from zero.

As an additional approach, the Mean Comparison technique was also implemented. The null hypothesis that the average value of each dependent variable is the same in both groups (i.e. treatment and control group), was tested against the alternative hypothesis that the average value of each dependent variable is different. Table 4.2 shows the results found.

In Table 4.2, the second column indicates the estimations about how much the average value of each dependent variable would differ between a control and a treatment population. For example, given that the respondents were asked to rate a 11-point rating scale from 0 to 10, an estimated coefficient of 1.3 would mean that watching the video increases the average value of the specific dependent variable (e.g. FAIRNESS) from 2.0 to about 3.3 (see Figure 4.2). The third and fourth columns of the Table 4.2 indicate whether the estimated coefficient is statistically significant or not, according to the regression and the mean comparison technique respectively. A P-value of 0.1 or smaller implies that the null-hypothesis can be rejected and thus, it is believed that the treatment effect is significant at a significance level of 10%.

(34)

34 Table 4.2 Summary of Treatment Effect results

𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕_𝒊 Estimated coefficient P-value (Regression) P-value (Mean comparison) OTHER 1.233 0.247 0.331 ETHICS 1.978 0.051 0.044 CRIME 0.197 0.489 0.593 FAIRNESS 1.337 0.017 0.011 DETECT -1.334 0.027 0.080 VAT 0.257 0.663 0.435 POLICY -0.249 0.662 0.801 KNOW 0.569 0.000 0.000

Figure 4.2 Example of the Different Attitudes/Beliefs between Groups

Finally, the current study found that the educational video had; a positive but non-significant effect on attitudes to other people's tax evasion (OTHER); a positive and significant effect on tax ethics (ETHICS); a positive and significant effect on the perceived fairness of the tax system (FAIRNESS); a slightly positive but non-significant effect on attitudes to other illegalities (CRIME); a negative and non-significant effect on the perceived probability of detection (DETECT); a slightly positive but

0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10 0% 20% 40% 60% 0 1 2 3 4 5 6 7 8 9 10

Do you think that the distribution of tax burdens between high income citizens and low-income citizens is implemented in a fair way? (10= very fair)

a. Control group b. Treatment group

(35)

35 non-significant effect on the motivation to request for a VAT receipt; a slightly negative but non-significant effect on the support for policies against tax evasion (POLICY); and a positive and significant effect on tax knowledge (KNOWLEDGE).

In other words, the results are mixed. First of all, as expected, the educational video significantly improved the participants’ tax knowledge, yet did not affect their attitudes to other illegalities. Secondly, it was successful in influencing participants to have a stricter attitude on their own tax evasion and a more positive perception of the tax system fairness (with an estimated coefficient of 1.978 and 1.337 respectively). However, inconsistent with H1, H6 and H7, the educational video failed to significantly impact on people’s policy preferences, motivation to ask for a VAT receipts and attitudes towards other people's tax evasion. Lastly, when it comes to the probability of detection, the results are opposite than expected. The perceived probability of detection is significantly lower among those participants who received an educational communication than those who did not (with an estimated coefficient of -1.334).

The lack of support to the hypothesis that an educational communication impacts on taxpayers’ attitudes towards other people's tax evasion could be due to a “floor (or ceiling) effect”. According to Kaplan et al. (1997), a floor effect is in place when persuasive communications have no significant effect on taxpayers because these taxpayers already hold a relatively strict opinion against tax evasion. As shown in Figure 4.1a, about 60% of the subjects assigned to the control group expressed their total agreement with the statement that “It is a moral obligation for all citizens to declare their exact income and pay the full amount of their taxes”, while in an 11-point rating scale from 0 to 10 (where 10 equals total agreement) the average value was 8.89. Thus, it seems that there was very little scope for a significant persuasive effect to be deployed.

The possible explanation of the opposite-than-expected effect on the perceived probability of detection is the previous overestimations by many people, about the number of audits taking place every year. As shown in Figure 4.1d, about 55% of the subjects assigned to the control group estimated the probability to “undergo a tax inspection this year” as medium or high. However, this perception is rather illusory in

(36)

36 a country where the adult population is more than 8.5 million and the tax inspections are only about 130,000 per year. Interestingly, Bergolo et al. (2017) found similar results for an audit-statistics letter sent to the owners of over 20,000 firms in Uruguay.

5. Conclusion and Discussion

This study used a video including information about tax rates, public spending, the societal cost of tax evasion and the detection risk to empirically determine whether an educational communication could successfully tackle tax evasion in Greece. The results indicated that the video was effective in improving tax ethics and tax fairness perceptions, while less effective in influencing people’s policy preferences, motivation to ask for a VAT receipt and attitudes towards other people's tax evasion. In addition, the video negatively affected (i.e. lowered) people’s perceptions of actual audit probability. These findings are interesting since many researchers have established a connection between attitudes and behavior (Eriksen and Fallan, 1996). Also, they are in line with findings by McGraw and Scholz (1991), Roberts (1994), Wartick (1994) and Maroney et al. (2002), which similarly indicate a positive effect on tax ethics and tax fairness.

Every interpretation of the current study’s findings should take account of the limitations of the implemented research procedure. Firstly, the sample consists of 112 citizens, who are relatively younger and better educated compared to the entire population of Greek taxpayers. In other words, the sample is not representative and thus, the results cannot easily be generalized. Secondly, the experiment was carried out on an online survey platform, where the control over partial compliance was limited. Thirdly, the participants' responses were recorded only just after receiving the treatment. Thus, the effect of the educational video was measured only in the short run. An additional limitation was the use of other variables as a proxy for the people's actual tax evading behavior. And finally, the results may have been affected by other external factors such as concurrent televised news reports about convicted tax evaders and proposed changes in pension laws, as well as political debates over the future of the country after exiting its bail-out program in August 2018.

(37)

37 In conclusion, the clearest result from this experiment is that educational messages can have a significant influence on, at least, some of the people’s attitudes and perceptions about taxation and tax evasion. This implies that it would be a step in the right direction for governments and tax administrators to begin considering implementing real educational campaigns. However, caution is needed when building such a campaign since some approaches might prove to be flawed, and even minor mistakes may lead to some opposite-than-expected results. In order to foresee and prevent these mistakes, it is recommended to first measure the current knowledge, attitudes and beliefs held by the targeting population, and then, based on these findings, to create a case-specific educational communication. For example, in the case of Greece, the current study found that many people had previously overestimated the number of audits taking place every year. As a result, a potential educational campaign against tax evasion should better avoid including relevant information.