Chapter 5. Testing and Implementing Digital Tax Administration

Sanjeev Gupta, Michael Keen, Alpa Shah, and Genevieve Verdier
Published Date:
November 2017
  • ShareShare

Related Material

Show Summary Details
Jingnan (CeCilia) Chen Shaun Grimshaw and Gareth D. Myles

The benefits of digital technology are well documented, leaving no doubt that it can also ease tax compliance, reduce tax collection costs, and increase administrative efficiency.1 Yet detailed analysis of consequences is crucial to these efforts.

Behavioral economics has shown that even small changes, or “nudges,” can significantly affect actions.2 This is particularly true of taxation, in which compliance is determined by a complex mixture of financial, social, moral, and psychological factors. The behavioral implications of any implementation of digital technology for tax administration need to be scrutinized to avoid unintended consequences. Innovations that initially appear innocuous and beneficial may well introduce nudge behavior in detrimental directions.

Pre-population of tax returns is a leading and current digital innovation, pioneered by the Danish revenue service in 1988 and followed in a number of countries.3 Pre-population is often accepted without question as a way to significantly reduce transaction costs in tax payment, but it is also a significant nudge that, psychologically, transfers “ownership” of errors from the taxpayer to the revenue agency. Tax administrations are also encouraging online submission of tax returns. Evidence shows the consequences of onscreen prompts, but little research is directly related to the tax environment.4 Taxation is complex, and individuals filing online may make errors. The service must always be designed to minimize errors. Unlike participants in most online activities, those in tax payment are unwilling, and compelling evidence suggests that a significant percentage are prepared to cheat if they perceive it to be beneficial. Digital developments must be designed to ensure that they do not provide additional incentive or motivation for cheating.

Digital technology in tax administration not only offers lower transaction costs, but also allows innovation in tax policy. A tax system will not function effectively if it imposes requirements that administration cannot meet. For example, the marginal rate of income tax cannot be determined by family income if the administrative system records only individual income. Nor can a consumption tax depend on the quantity of consumption if the system does not record purchaser identity. The policy implications of digitalization are inherently linked with advances in other areas of science and technology. What we can do with digital technology depends on the level of our understanding of what creates individual differences. As this knowledge progresses, our perspective on the foundations of tax policy will also have to change. Where this may lead is currently unknown, but speculative ideas are presented in the following discussion.

The United Kingdom provides an example of advancing digitalization. In its March Budget of 2015 the government outlined a vision for a digitized and online tax system, dubbed “Making Tax Digital” by HM Revenue and Customs (HMRC). It looks forward to simplification of tax payment for individuals and firms with information from third parties used to pre-populate returns, and ultimately envisages real-time taxation, eliminating the need for annual tax returns.

HMRC has used information technology to store and process tax data for a considerable time, and online submission of returns is gradually becoming the norm. However, the online submission system is little more than a digitized version of the paper return, with little or no added functionality. It produces an automatic tax calculation, but has no interaction with the taxpayer. The step remaining is the full exploitation of interactive online systems that integrate reporting, recording, advising, submission, and payment.

The reluctance of many revenue agencies to advance digitalization is clearly understandable given the potential costs of mistakes. Foremost among these are the risk to revenue, damage to reputation, and potential reduction of tax morale. The digitalization of tax administration is technically complex given the volume of activity the system will have to accommodate and the importance of security and absence of errors. The required quality standards will be achieved only through extensive technical and functional testing. Any system inadequately tested will quickly fall into disrepute, with potentially significant financial and reputational costs.5

The necessity for technical testing seems self-evident. What this chapter argues is that a system needs an accompanying and equally intensive program of behavioral testing. This is because any system carries with it behavioral implications, and the design will determine how taxpayers react. These reactions will include whether the system was intuitive to use or attractive to look at, as well as the extent to which taxpayers are compliant with the tax code and their attitudes to tax collection.

Clearly, compliance is affected by a complex set of economic, psychological, and social factors, and behavioral economics has demonstrated how small nudges can lead to large behavioral change.6 And a move from traditional paper-based filing to an online system, with pre-population and real-time activity, as rather more than a small nudge, could significantly impact compliance. This is why a digitized system needs behavioral testing, which is based on the idea that small details can matter. The methodology of experimental economics described in Box 5.1 is perfect for this.7

The results of two experiments undertaken at the United Kingdom’s Tax Administration Research Centre that investigated different aspects of making taxation electronic form a major part of the chapter.8 The first experiment was designed to explore taxpayer response to incorrect or incomplete pre-population. The results of the experiment are described alongside other research investigating pre-population.

The second experiment considered the impact of online assistance during completion of a tax return relative to traditional paper or phone assistance. Providing guidance to taxpayers in paper form is the long-established standard, and it is not known how behavior will change if paper guidance is transferred to online guidance or what effect online “pop-up” boxes offering assistance will have. As a byproduct, the experiment also provided insight into the nature of errors in tax returns. Moving beyond this, the online environment permits greater interaction with the taxpayer during return completion, which can allow inclusion of nudges and prompts in the return. Experimental methods can be used to maximize the effectiveness of these.

The introduction of pre-population and integrated online tax services are small steps at the start of the digital revolution in tax administration, and many countries have already gone much further than the United Kingdom. What impact will these innovations have on tax policy? In particular, will they open the possibility for refinement of existing tax instruments or introduction of new instruments?

The economic theory of taxation was developed amid pre-digital tax administration. The chapter takes the implications of the experiments as a starting point from which to explore the extent to which extant theory must be updated given new technology. The conclusion—based on speculation about technologies that may be developed—sees the elimination of one of the most basic tenets of tax theory.

The next section reviews the key principles and the methodology of experiments in economics, followed by a review of experiments on the consequences of the pre-population of tax returns. The chapter then considers how completion of returns is affected by the form of customer service, reflects on how economists conceptualize tax theory, and speculates on the consequences of digitalization and technology.


An important component of digital innovation is the use of third-party data to pre-populate the tax return. This is a first step to eliminating the need for an annual return. Under the U.K. system (and those of many other countries) taxpayers are required to enter data obtained from third parties into the tax return. Examples of such data include income from employment, income from property, and eligible expenses such as private pension contributions.

HMRC receives much of this information directly from the third party, so that the taxpayer is providing information that HMRC already holds. This might have some strategic advantage for the revenue service as an indicator of potential non-compliance, but it unnecessarily burdens taxpayers, who are required to store information and may need to seek information to complete the return. HMRC’s intention is to use the information they already hold to pre-populate the tax return so the taxpayer will not need to re-enter the data when filing.

Pre-population is appealing because it reduces the compliance costs of taxpayers and has potential to reduce errors and omissions. It saves the revenue agency time, because pre-populated data will not need checking against the records from third parties. Moving to pre-population is a small task relative to moving to real time, and the benefits to doing so are identifiable. But experiments have also revealed potential costs: if the pre-populated data are not correct, how do taxpayers respond?

The fundamental problem with pre-population is that it is possible for the revenue agency to include incorrect or incomplete information. The latter could arise, for example, when a taxpayer has multiple sources of income and the revenue agency does not receive reports from all third parties. How a taxpayer responds to a pre-populated form on which the information is not entirely correct then becomes a question in behavioral economics. The most positive outcome from the perspective of the revenue service is that the taxpayer will simply correct the information. It might be thought that this would definitely be the outcome if the revenue service had overstated the true level of income.

Box 5.1.Economic Experiments

Only a few years ago economics could, without risk of dissent, be labeled a nonexperimental science. A leading textbook noted that “It is rarely, if ever, possible to conduct controlled experiments with the economy. Thus, economics must be a non-laboratory science.”

This situation has changed completely and experiments are now accepted as part of the standard methodology of investigation in economics (Starmer 1999).

Experiments permit investigation of complex behavioral phenomena that may be hidden within economic data by the multiplicity of simultaneous environmental changes. Experiments can be deliberately designed for implementation in an experimental laboratory or in the field. Others, called natural experiments, are derived from exogenous changes in policy or situations that create treated and control groups whose behaviors can be contrasted. The ability to precisely control the environment is a key benefit of designed experiments. Replication is also possible, to compare results across time and cultures (such as the public good games discussed by Ledyard 1995). Natural experiments do not permit control or replication, but have the advantage of natural behavior and large sample size.

Experimental economists are generally agreed on a set of principles that govern the conduct of experiments: (1) salient financial incentives for experimental subjects to encourage considered participation, (2) absence of deception in the design and execution of the experiment, and (3) random assignment of subjects to treatment conditions to provide statistical validity.

A typical experiment lasts between 30 minutes and 2 hours and involves completion of one or more tasks and sometimes repetition of the same task. It may also involve pre- and post-testing of attitudes and opinions. The sample size is usually 60–300 participants depending on the task and the number of treatments. A treatment is a specific set of values for the experimental parameters, and experimenters are interested in how a change in parameters affects behavior. The number of subjects has to be large enough to ensure sufficient participation in each treatment to obtain statistical significance. The level of subject payments is set to reflect the opportunity cost of the time spent in the experimental session by subjects. All monetary amounts within an experiment are expressed in experimental currency units, the exchange rate of which to US dollars is set according to duration and number of repetitions or rounds.

Laboratory experiments permit “clean” comparison of the consequences of different treatments, but the laboratory is always an unnatural environment for experimental subjects and a simplified setup will always appear artificial. External validity requires results that hold for the general population facing a real decision problem in their natural environment. This makes the use of university students as subjects questionable for tax experiments (Choo, Fonseca, and Myles 2016; and Alm, Bloomquist, and McKee 2015). Using an appropriate subject pool (such as taxpayers for a tax compliance experiment) and moving the experiment online enhance external validity. It can be improved further by taking the laboratory into the field (a framed field experiment), but the control of the experimenter is reduced and the experiment will be context heavy.

Experimental investigations of tax compliance share common features (Alm 2012). Each experimental subject is given or earns income and then decides the amount to declare to the tax authority, which is subject to tax at a given rate. Meanwhile, there is a given probability of being audited. A subject who is audited and has unpaid tax will be fined proportionally to the level of unpaid tax. The results of experimental investigations into tax compliance to date suggest that there is no single design that is the best fit for all purposes, and that designs should be constructed in line with the research question under investigation.

However, as reported in the following paragraphs, a different behavior sometimes emerges in experiments. There are also two potential negative reactions to incorrect or incomplete pre-population. The taxpayer may accept the pre-populated values without comment, perhaps through having more faith in the revenue agency than in their own records, so the pre-populated value becomes established as the truth. In behavioral terms, this is a form of status quo bias or behavioral inertia. It can arise whether the pre-populated value is above or below the true value. The alternative negative reaction is more strategic. The pre-population of the tax return can be interpreted as a signal of the information held by the tax agency and, correspondingly, of what it does not know. Pre-populated values below the true level indicate the limited information of the revenue agency and can encourage deliberate evasion (by knowingly accepting an incorrect value) since they signal reduced likelihood of evasion being accepted.

The economic analysis of tax compliance has focused on explaining how taxpayers react to changes in the audit rate, level of punishment, and tax rate. Substantial theoretical literature models the decision process (see Hashimzade, Myles, and Tran-Nam 2013) and experimental literature tests these models. However, the effectiveness (or lack thereof) of pre-population of tax returns has not been a significant topic. This is possibly because pre-population has only recently become important to administration. The limited evidence is now reviewed.

Bruner and others (2015) studied the effect of pre-populating tax returns using undergraduate students at two US universities as experimental subjects. The experiment involved subjects earning income and making reports. The level of income for each subject was determined by undertaking a task at the start of the experiment and then remained constant for the subject throughout the experiment. The tax liability depended on earned income and claimed deductions. Tax returns could be audited and a punishment imposed for noncompliance.

The subjects had to make three entries into the tax return. Income was separated into “on-the-record” and “off-the-record” components. On-the-record income was known to the revenue agency through reports from third parties. Income “off the record” was not subject to third party reporting, so was unknown by the revenue agency (and this was known by the subject). Subjects could also make tax deductions that could be standard (such as deductions for spouses or children) or itemized. In some of the treatments, lines on the tax return were pre-populated, but were not in other treatments.

Subjects filed multiple tax returns in a sequence, each of which corresponded to a different profile of deductible expenses. In some cases, it was advantageous to file an itemized deduction, and in others it was not. A number of audit rates were also used, and fixed for each treatment and so did not respond to reported incomes. The baseline treatment had no pre-population, certainty of deductions, and the subjects received no off-the-record incomes. This was compared to a treatment with pre-population, certainty, and no off-the-record income. Further treatments introduced off-the-record income, uncertainty of deductions, and higher levels of off-the-record income for some subjects.

The results showed that compliance was extremely high for matched income. Most of the noncompliance that arose was from underreporting off-the-record income (only 81 percent of this income was reported) and from overstating deductions (claimed deductions were 112 percent of the allowable amount). Pre-population caused underreporting of off-the-record income to increase, and this effect was strongest when the pre-populated deduction exceeded the allowable amount. Furthermore, if the pre-populated amount incorrectly understated tax liability, then underreporting increased. This final result illustrates the danger revenue agencies face when using pre-population, and that it can signal the limited information of the agency.

Kotakorpi and Laamanen (2016) used data from a “natural experiment” in the mid-1990s in Finland. In the experiment, a subset of taxpayers had their tax forms partially pre-populated with data from third parties. These taxpayers were only required to file a return if the pre-populated information was incorrect or incomplete. They had the option to file if they wished to claim eligible deductions. All other taxpayers had to complete a standard tax return that was not pre-populated. The analysis explored how pre-population affected filing for five types of items: (1) pre-populated income (from primary and secondary employment), (2) non-pre-populated income (other earned income and capital income), (3) pre-populated deductions (mortgage interest deduction in 1997), (4) non-pre-populated deductions, and (5) reported wealth.

The most significant impact of pre-population was observed for non-pre-populated deductions: a partially pre-populated return led to a reduction in filed deductions compared to the control group. Overall, about 25 percent fewer taxpayers claimed non-pre-populated deductions. In contrast, claims for the pre-populated deductions increased. The reported level of non-pre-populated income and reported wealth also declined. The reporting of pre-populated items was not affected, nor was total taxable income. Kotakorpi and Laamanen (2016) observed that receiving a partially pre-populated tax return creates a tendency to report fewer of the items not pre-populated but more of those that are.

Fonseca and Grimshaw (2017) tested the effects of pre-population using a one-shot decision quasi-field experiment using U.K. taxpayers as subjects (because it is questionable whether students act in the same way as taxpayers in experiments).9 That is, the evidence is mixed on the extent to which results from student samples can be generalized to the wider population, with indication that students are more noncompliant than experienced taxpayers in an experimental setting (Alm, Bloomquist, and McKee 2015; Choo, Fonseca, and Myles 2016). Experimental subjects played the role of a fictitious taxpayer with two income streams (not subject to withholding) and tax-deductible expenses. Using two income streams allowed modeling of a revenue service with limited information, in which case the return was pre-populated with only one of the two streams.

Various forms of pre-population were assessed against a baseline treatment without pre-population. Pre-population was also combined with onscreen prompts intended to create barriers to noncompliance. The prompts included the need to click on a checkbox to unlock entries and warning messages about the audit probability. The experiment also included an expense item determined by the roll of a dice by the experimental subject. Experimenters did not observe the value of the roll, giving an unverifiable component to the experiment. The subjects were told they could be audited, but were not informed of the audit rule.10

The experiment used seven separate treatments:

  • BASE: The tax form was not pre-populated.

  • CORR: The two income streams were correctly pre-populated and the tax form revealed that the revenue agency held the correct information.

  • OVER: The revenue agency is shown as having information on three income streams (the form is pre-populated with one of the actual streams double-counted).

  • UNDER: Pre-populated with data on only one of the two income streams and this is the only stream known to the revenue agency.

Three variations of UNDER were also used:

  • UNDERGENERIC: Featured a checkbox that had to be clicked to unlock the pre-populated income field and clicked again to confirm any new value entered.

  • UNDERALWAYS: Featured the message: “Most people in your circumstances enter an income value of more than 40,000. Values below this amount are more likely to be audited. Click the tick box to confirm you wish to proceed.”

  • UNDERTRIGGER: The same message as UNDERALWAYS if the participant inputted a total self-employment income amount lower than 40,000.

Figure 5.1 illustrates the results: it displays the verifiable compliance rate for the experimental treatments. Part always remained unverifiable because the dice roll was not observed. The BASE treatment had a very high compliance rate, but only 70 percent of subjects fully reported income.

Figure 5.1.Average Verifiable Compliance Ratio, by Treatment

Source: Fonseca and Grimshaw 2017, p. 33.

The CORR treatment had a higher average compliance rate and a higher proportion of fully compliant subjects than BASE, though neither difference was statistically significant. The OVER treatment was slightly lower than BASE in both dimensions. A marked difference occurs with the UNDER treatment, which led to a large and significant fall in average compliance and in fully compliant subjects. There was a further fall for the UNDERGENERIC treatment. The latter two treatments show that subjects were willing to accept false low reports but unwilling to engage in a process (checking a box) to make a correction. The nudges used in UNDERALWAYS and UNDERTRIGGER restored some of the compliance.

Figure 5.2 details the results by showing separate compliance rates for each class of income. When self-employment income is correctly pre-populated the rate of compliance is high. This reveals that pre-population is successful for the revenue agency if it holds correct information.

Figure 5.2.Propensity for Verifiable Compliance, by Treatment

Source: Fonseca and Grimshaw 2017, p. 34.

An unexpected finding in the OVER treatment is that it is compliance with expenses that responds to the incorrect pre-population. What seems to be happening is that subjects realized the overstatement of income, but were reluctant to change the pre-populated value. Instead, they engaged in compensating behavior through over-claiming for expenses. Compliance for self-employment income in the UNDER treatments declined significantly. The results emphasize the willingness of subjects to accept mistakenly low pre-populated values and the benefit of nudges to restore some degree of compliance. It is perhaps not surprising that the triggered nudge was most effective since, psychologically, this creates an impression of monitoring of actions.

A surprising finding of this experiment was that pre-population with over-estimated income levels had little effect on behavior. The values were corrected in some cases, but in others the subjects accepted the incorrect values even though this resulted in an excessive tax payment. This behavior reflects an acceptance of the authority of the revenue agency, with the thought process “if they say this is correct then I have to believe it.”

In contrast, pre-populated values that understated correct income had a significant impact. This arose because subjects were happy to accept the incorrectly low values. As discussed earlier, this can be explained in that low values were seen as a sign of revenue agency ignorance, which subjects were willing to exploit. The introduction of barriers to editing pre-populated fields may worsen noncompliance if the pre-populated values are incorrect. Finally, behavioral prompts help overcome incorrect pre-population only if they are responsive to behavior in the filing process. The appearance of a pop-up box in response to a lower-than-expected income report conveys the impression that the system is taking notice and encourages increased compliance.

The focus of Fonseca and Grimshaw (2017) is slightly different to that of Bruner and others (2015), so the two studies are complementary. Fonseca and Grimshaw study a one-shot decision with one set of parameters, where only the pre-populated value in one of the entries is varied, whereas Bruner and others look at a wider set of parameters and a more complex filing decision. Bruner and others consider several audit rates, which are invariant to behavior and known with certainty, while Fonseca and Grimshaw consider an unknown audit rate, which depends on filing behavior. Bruner and others consider a more complex environment with itemized and non-itemized deductions, as well as on-the-record and off-the-record incomes. That both studies find that pre-populating tax returns with values that underestimate taxpayers’ liabilities leads to higher non-compliance lends greater robustness to both sets of results.

These investigations into pre-population are informative of the consequences of further digitalization of the tax process. We know there are taxpayers who are noncompliant under the current system. Pre-population may help reduce some of the noncompliance that arises from error. But even this is not guaranteed, because the experiments reveal a reluctance to change incorrect pre-population.

Only if the revenue agency is correct are errors sure to be reduced. OECD (2008) reports apparently high accuracy of pre-populated returns (about 70 percent in Denmark and Sweden needing no adjustment). But, critically, it also observes “these reporting arrangements do not include details of income from self-employment and rental properties” (OECD 2008, 8). This is important because these are the income sources that are open to noncompliance and the hardest to pre-populate. Some of the deliberate noncompliance may be deterred by the pre-populated value acting as a minimum which a noncompliant taxpayer will not wish to correct downward for fear of signaling their noncompliance. However, taxpayers with a propensity to be noncompliant will take the pre-populated information as a signal and use it to refine their noncompliance strategy. Pre-population with an incorrect value acts as a signal of the limited information of the revenue agency which a noncompliant taxpayer will wish to exploit. The experiments agree that revenue agencies run a significant compliance risk from understated entries.

On the other hand, the results provide clear encouragement for strategic behavior by the revenue agency. The following comments should be prefaced by saying that it is not expected that any revenue agency should ever adopt these strategies. But, the strategic implications of pre-population cannot be ignored. Because understating increases noncompliance and some taxpayers are reluctant to reduce overstated entries, there is a strategic incentive for the government to overstate. Clearly, this would be counter to all rules of good governance, and if a revenue agency were discovered to have acted in this way, it would reduce trust.

Even more sinister is the nature of the motive a revenue agency has to deliberately understate. The revenue service could understate an item about which it is certain to test the willingness of the taxpayer to make the required correction. A failure to correct could then be used as an indicator that an audit is required. More disturbingly, understatement in pre-populated values could be used to lure a taxpayer into noncompliance—with punishment to follow.

The experiments give valuable information on how taxpayers will respond to mistakes in pre-population. The revenue service could act strategically, but if it is simply trying to be as straightforward as possible, then pre-population should be undertaken to the best of the agency’s ability using all available information. The potential noncompliance implications will have to be accepted as the price paid for easing the tax affairs of compliant taxpayers. Behavioral prompts can work, but have to be carefully designed and tested. Implementation should also account for the evident reluctance of taxpayers to correct errors in pre-populated entries.

What is not clear is whether pre-population with imperfect information will ultimately increase or reduce tax revenues received; none of the experiments is sufficiently precise to answer that question. A fair expectation is that refinement of the system over time by increased integration of systems would improve accuracy and ultimately eliminate the noncompliance effects.

Taxpayer Guidance: Pop-Up Or Paper?

National Audit Office (2016) scrutinizes the impact of HMRC customer service on personal taxpayers.11 The office estimated that 17.5 million taxpayers used HMRC’s information and advice services in 2015. The report found that the quality of service of taxpayers may affect tax compliance. The move to online tax accounts will shift the emphasis of service from traditional paper to online guidance. The experiment investigates whether HMRC tax guidance affects tax compliance and, if it does, by how much. The effect of a support line handled by tax advisors on tax compliance is also explored. In addition, as a by-product, the experiment gives insight into the possibility of errors when completing a return.

Revenue service tax guidance is often the starting point of the taxpayer journey. The contents, as well as the delivery form of the contents, largely determine the ease of comprehension and thereby the need for additional help. Consequently, the quality of the tax guidance (such as the ease of comprehension) may affect demand for further contact with the revenue service and ultimately influence the overall tax compliance level. If the cost of seeking help exceeds the benefits of completing a fully compliant tax return, taxpayers may simply resort to their own best endeavors to complete a compliant tax return or potentially even behave in a deliberately noncompliant manner (for example, see Graetz and Wilde 1985; Clotfelter 1983).12 Many reforms in tax administration strive for better service quality to attain greater compliance.

Traditionally, enforcement effort, intensity of audits, and fines and penalties have been the tax authority’s primary tools to promote voluntary tax compliance. More recently—and lagging the actions of revenue services—academic research has begun to focus on the impact of the provision of tax information and assistance services on overall tax morale and compliance. Alm and others (2010) demonstrated that taxpayers respond positively to service programs in an experimental setting. Specifically, customer-friendly tax administration increased average compliance by 27 percent.13

Two main reasons were proposed to account for the results. First, by relieving the burden of complying with tax regulations, the tax authority can affect “soft” tax compliance factors, such as the perception of fairness and trust. Secondly, the tax authority is able to reduce “hard” tax compliance factors, such as the actual compliance costs of the taxpayers. Another experiment, by McKee, Siladke, and Vossler (2011), found additional evidence in support of the arguments above by showing that a helpful information service drastically reduced tax evasion. A further experiment, by Vossler and McKee (2013), looked into the effectiveness of a taxpayer service program in enhancing tax reporting, with the emphasis on the accuracy of tax filing. They found that even an imperfect service helped increase the likelihood of filing and filing accuracy.

Finding appropriate tax filing information and applying it costs taxpayers in time and effort. There may be barriers to the degree of cost subjects are willing to bear to find an appropriate rule, and factors that reduce search costs may therefore lead to greater compliance.

Based on this line of reasoning, we predict that holding the delivery form constant, people are less likely to make mistakes when tax guidance is succinct and precise rather than long and detailed. On the other hand, holding the contents of the guidance constant, people may be less likely to make filing mistakes when the guidance is provided as an online pop-up. As such, relevant information is immediately at hand, requiring no searching through pages of printed materials.

Combining these observations, precise guidance delivered in an online pop-up may be more customer-friendly and may encourage greater tax compliance than the long and detailed guidance printed on paper. The experiment we report is designed to test these observations to enhance the online filing experience.

Experimental Design

The experiment features a one-shot tax filing decision. The primary focus of the research was the investigation of the effect of various treatments on the values reported in a tax return for a moderately complicated taxpayer profile. Typically, there are benefits of experimental designs with repeated actions as they allow for learning by subjects. However, such benefits are typically greatest when time allowed for decisions is short. The design presented here does not have such advantages as there is a requirement for the decision to be complex to force subjects to examine the tax materials they are presented with to be able to file a compliant return. This is different than many other laboratory tax experiments in which the filing decision is very simple.

In the experiment, subjects were given the profile—receipts and expenditures—of a particular taxpayer. The experiment focused on whether the expenditures were allowable business expenses. Uncertainty arose in whether an item in the profile was allowable, what proportion of a particular expenditure was allowable, and into which field in the tax return the subject should enter a value deemed allowable. The information services were provided to remove these uncertainties for subjects who wished to be compliant.

Self-employment was chosen as the basis for the experimental tax profile due to the level of relevance to the self-assessment population as a whole.14 Given the complex nature of the tax return that needs to be completed (SA100), it can be assumed that many self-employed need some level of support. The number of accountants and tax advisors offering assistance suggests that many self-employed seek professional help.

Table 5.1 details the profile used throughout the experiments, with the profile itself shown in Annex 5.1. Values are given in experimental currency units (ECU), as is typical in experimental economics. This is primarily to preserve framing effects over different subject pools, as the exchange rate for ECUs to actual cash can be varied to allow for different levels of compensation, but also to frame the experiment with real-world values.

Table 5.1.Tax Profile Details and Correct Allowance
CategoryDetailAmountCorrect Allowance
IncomeFitness classes25,200Not applicable
ExpensesCar purchase1,5000
Running car (8,000 business miles out of 10,000 total miles) to/from place of work2,5000
Church hall hire5,7605,760
Advertising flyers175175
Gym membership1,2000
Annual household bills (one day a month working from home)7,500246.58
Mobile phone (15 percent of total usage was for business purposes)42063
Source: Authors’ calculations.

From Table 5.1, net income (income less total expenses) is 6,145 ECU (subjects were informed of this figure as part of the system). The compliant level of deductions is 6,244.58 ECU, leaving a taxable income of 18,955.42 ECU. The payment of the subject for participation in the experiment was based on the subjects’ post-tax balance (net income less tax payment). The experiment used a tax rate of 20 percent so the compliant tax payment was 3,791.08 ECU and a post-tax balance of 2,353.92 ECU.15 This value of the post-tax balance corresponded to the subject earning £7.06 for the completion of a compliant return (for a total of £12.06 once the show-up fee was included). The maximum earnings possible from the task, obtained by an over-declaration of expenses to give a taxable liability of zero, was £18.43 (£23.43 with the inclusion of the show-up fee). The minimum level of earnings, from over-declaration of expenses leading to a large fine, is £0 (£5 with the inclusion of the show-up fee).

The focus of the experiment was to examine the effect of guidance on those who file for themselves. The experimental treatments vary the contents of the guidance as well as the delivery form of the guidance to examine compliance behavior. All the guidance contents were direct from HMRC materials. In the experiment, the long form of guidance (LONG) refers to the set of downloadable and printable PDF help sheets available on the U.K. government website.16 The short form of guidance (SHORT) refers to the information contained in the pop-up boxes on the HMRC online tax return. The items covered in both forms of guidance are mostly identical. However, there are notable differences in the information provided and the delivery form between the paper and the online guidance. First, for the same item, the long-form guidance is generally more detailed than the short. Second, how the information is delivered also differs. The short-form guidance appears as pop-up information boxes right next to the item in the tax form. With the exogenous variation implemented in the experiment, we are able to disentangle the differential effects of guidance contents and delivery form on voluntary tax compliance.

The first component of the experimental software was a set of instructions that explained the task. The instructions included details of the calculation of tax payable as 20 percent of the tax liability defined as the difference between declared income and expenses, and of the random chance of audit (set at 50 percent) and the calculation of fines for unpaid taxes, based on payment of the unpaid tax plus an additional 100 percent of the unpaid tax. Numerical details were presented for a number of examples of different filing decisions, based on a simple profile rather than the actual profile presented to subjects. The instructions detailed the incentive scheme to participants, in particular, the payment of a fixed £5 show-up fee and the conversion of any balance in the experimental system at the end of the session to pounds at a rate of 1,000 ECU to £3. The instructions also detailed the presence of assistance with the tax-filing decision based on the treatment.

The tax filing components consisted of three screens. The first screen allowed subjects to enter values for a number of expense fields. The value of the subject’s income, as shown on the profile, was pre-populated and uneditable. The second, the tax filing screen, showed participants their tax calculation based on the value of expenses they had entered and the default income level. Subjects were invited to either alter their tax declaration, which would return them to the previous submitted screen, or to submit their tax return. Upon submission of the tax return, subjects were shown the third and final page of the tax filing component of the system. On this page subjects were informed of their tax payment, whether they had been selected for audit, and in the case of any audit, what the result of the audit was and any additional taxes or penalties to be paid. Finally, subjects were directed to complete an online questionnaire as part of the software that asked them questions about their motivations for choices in the experiment as well as gathering demographic details.

Experimental Treatments

Original treatments

The initial set of experiments focused on three treatments in terms of the effect on compliance of assistance materials without the use of phone or online help. The treatments were decomposed into two parts. The first part addressed the content of the materials, in terms of the form of guidance: LONG used HMRC printed materials, while SHORT used HMRC guidance from the self-assessment tax filing website. The second component addressed the delivery form of the assistance. Assistance was either provided to subjects in print, referred to as PAPER, or provided through the pop-up information box, referred to as ONLINE. The three treatments detailed in Table 5.2 were undertaken in the first stage.

Table 5.2.Stage 1 Treatments
Treatment NameDescription
LONG_PAPERHMRC long form guidance delivered on paper
LONG_ONLINEHMRC long form guidance delivered as online pop-up box
SHORT_ONLINEHMRC short form guidance delivered as online pop-up box
Source: Authors’ calculations.Note: HRMC = Her Majesty’s Revenue and Customs.

From Table 5.2, it can be seen that no SHORT_PAPER treatment was conducted. Although it was felt that while this treatment may have added some insight, the results that would have been obtained would probably not be worth the cost of running the treatment. A further comment on this omission is presented after the results.

Additional treatments with a support line

Two additional treatments were run in which subjects were offered the opportunity for additional guidance through a support line. In all cases, the SHORT_ ONLINE guidance was used. In one set of treatments the laboratory computers were preinstalled with Skype and with a link to call a tax advisor. Subjects were told in the instructions and on the tax form that they could call through Skype if they required assistance. In a further set of sessions, telephones were installed in the laboratory with a fixed number to dial. Subjects were told they could use the phone to gain additional guidance on the instructions and on the tax form. They were also given a note with the direct number to call in case they were unfamiliar with the direct call mechanism.

Students who had previously undertaken the experiment in the first round of experiments were asked if they would wish to serve as paid advisors in the experiment. Ten advisors were recruited and attended a training session where they were given a document detailing the process of how to handle a call from a subject. Having worked through the process, advisors then undertook a series of practice calls with one another to complete their testing. The advisors were then recruited for each of the sessions requiring advisors ready to respond to calls for guidance. The advisors followed scripts with standard answers.

Experimental sessions

Sessions were conducted in the experimental laboratory at the University of Exeter. For the majority of the experiment, participants were undergraduate students at the university. A final session with advisors available was run using professional services staff recruited from the university. In a typical session, there were on average 20 (for original treatments) or 10 (for the additional treatments) subjects per session. In total, 266 subjects participated.


Table 5.3 summarizes the overall tax filing error rate by different treatments for stage 1. The error rate here is calculated as the percentage of the population who fail to declare the correct amount of allowable expenses (the correct amounts of allowable expenses for each of the items are outlined in Table 5.1). We include both underpayment of taxes (claiming more expenses or making positive errors) and overpayment of taxes (claiming lower expenses or making negative errors) in calculating the overall error rate. Across all treatments, around 98 percent of the population make errors in their tax filing. And most people made an error on the positive side, that is, they over-claim expenses and underpay taxes. However, still about 9 percent under-claim expenses and overpay taxes. Tables 4.4 and 4.5 detail the magnitude of those errors.

Table 5.3.Overall Error Rate by Treatment
TreatmentObservationsOverall Error Rate (percent)Population Overpaying Taxes (percent)
Source: Authors’ calculations.

Table 5.4 shows that the average amount of underpayment accounts for about 27 percent of the total taxes to pay. In comparison, the average overpayment (Table 5.5) amounts to about 17 percent of total taxes to pay. While some subjects (9.5 percent of the sample) under-claim on the amount of expenses they are entitled to (and thereby overpay their tax due), the majority of subjects over-claim in that amount, leading to underpayment of tax due on average.

Table 5.4.Underpayment by Treatment
TreatmentObservationsAverage UnderpaymentAs Percent Tax to Pay
Source: Authors’ calculations.
Table 5.5.Overpayment by Treatment
TreatmentObservationsAverage OverpaymentAs Percent Tax to Pay
Source: Authors’ calculations.

Next, we compare the average tax underpayment among the three treatments. As Figure 5.3 shows, people tend to underpay by the least amount in the SHORT_ONLINE treatment and by the largest amount in the LONG_ ONLINE treatment. The difference between these two values is statistically significant. This suggests that it is the content of the short-form guidance that causes a higher level of compliance, since we are holding the delivery form constant. Although on average people underpay taxes less in the LONG_PAPER than in the LONG_ONLINE treatment, the difference is not statistically significant. This implies the surprising conclusion that whether the information of the guidance is delivered using the pop-up information boxes or printed paper does not seem to cause a significant change in compliance behavior. Ordinary least squares regression analysis also confirms the above findings.

Figure 5.3.Average Underpayment of Taxes by Treatment

Source: Authors’ calculations.

Table 5.6 reports the regression results. The dependent variable is the tax filing error (the difference between the subjects’ correct allowance and claimed allowance, which can be positive or negative). The control group for the regressions is the LONG_ONLINE. The regressor SHORT_ONLINE is a dummy variable, equal to 1 if it is the SHORT_ONLINE treatment and 0 otherwise. Likewise, LONG_PAPER is also a dummy variable, equal to 1 if it is the LONG_ PAPER treatment and 0 otherwise. From (1), we can see that people in the SHORT_ONLINE treatment claim 239 ECU (or 6 percent) less than people in the LONG_ONLINE treatment. This is a rather large effect, especially considering the number of people in the treatment. In comparison, how the information is delivered also has positive impact on tax compliance, but the effect is fairly small and insignificant. From (2), the SHORT_ONLINE treatment effect persists while controlling for gender and age of the subjects. We observe no effect of gender or age on filing errors.

Table 5.6.Ordinary Least Squares on Tax Filing Errors with Treatment Effects








Number of Observations236229
Note: Robust standard errors are reported in parentheses. * indicates significance level at 10 percent, ** at 5 percent, *** at 1 percent.

We conducted similar analyses on the data from the additional treatments with either Skype or telephone support. We find that people in SHORT_ ONLINE treatment with support behave similar to the original SHORT_ ONLINE treatment. The main explanation is that only 10 percent (3 out of 30) subjects used the support line. From the post experimental survey, over 65 percent of subjects attributed their reason for not calling to the provision of sufficient information.

Our result, that appropriate guidance can increase the degree of tax compliance, is in line with previous studies into the effect of tax assistance on compliance behavior. As noted, many of these studies were conducted using students as subjects, so the subjects typically have little experience of the taxation system. This has raised questions about generalizability. The results from the student sample used in this study could, therefore, be best considered as applying to a set of taxpayers new to the self-assessment system, though only 34 percent of the sample responded positively to a question asking how likely they thought they would be self-employed in the future.

Our study differs from other investigations of the effect of customer service on tax compliance in that previous studies have been based on abstract settings, with a focus on the effect of simple information, revealing either a single correct value or a narrowing of the correct values appropriate for a specific field. In the experiment presented here, the goal was to investigate the effect of customer service using real-world examples of complexity in filing and the associated actual tax guidance. This design allows us to better examine our subjects’ decisions in the context of the real materials, but comes at the cost of the loss of some degree of environmental control.

The main result, based on comparison of the SHORT_ONLINE and LONG_ONLINE treatments, may overlook an important effect arising from the design of the experiment. The SHORT_ONLINE treatment reflects an operational reality: we used text from the HMRC online system so the information had been tailored for each of the fields in the online tax return. This design may lead to low search costs for tax filers for simple issues, in particular, questions of positive inclusion such as whether an expense should be filed in that particular field. The LONG_ONLINE treatment is artificial, however, in that the long-form guidance was simply pasted into the online pop-up. There is no such tailoring, therefore, and the tax filer was left to search through the full information.

On the other hand, such searching may have led the tax filer to discover issues of negative inclusion. For example, looking to see if an expense should be included in a particular field it might lead to the discovery that it should be filed in a different field. The results are consistent, nonetheless, with a reduction in search cost through guidance items with positive inclusion.

A fourth treatment using the short-form guidance on chapter materials may have shed further light on the issues by allowing further comparisons. But such a design again would have been artificial in that there is no such current operational reality. It was not clear how the linking of the short form guidance in chapter format to the tax form could best be performed to match that inherent in the SHORT_ONLINE treatment.

The post-experimental survey provides subjective evidence about the unintentional tax evasion. Among subjects, asked how they approached this experiment, 58 percent indicated that “I want to get my return right.” Another 30 percent suggested that “I don’t mind small errors.” Only 12 percent said “I did not mind having errors on my form if it benefited me financially.” A closer look at the error patterns suggests that the majority of subjects made an effort to determine the correct declaration values (Annex 4.2 provides detailed analysis of the error pattern). However, despite their efforts, they failed to get the tax return right.

More than 65 percent of the sample, meanwhile, indicated that the guidance provided sufficient information for them to complete the task (from the additional treatments). It may be of interest to further examine the gap between the high error rate and the level of overconfidence among taxpayers. Additionally, questions remain as to the characteristics of the contents that are the driving force of the behavior change and more detailed studies should be carried out with those questions in mind.

Digitalization and Tax Policy

The focus of the discussion here has been the impact of digitalization on tax administration in the near future. This is a necessary prelude to an analysis of policy, because administration determines what is possible, and practical value is limited in constructing a policy that is not administratively feasible.17 Furthermore, if administrative limitations result in a distorted version of a policy being implemented, then the outcome may be worse than from using a less desirable but implementable policy. The key observation is that technology does not just affect administration, but can transform what is possible in tax policy and, eventually, perhaps will even change how to conceptualize tax theory.

The first part of this section considers what policy innovations are possible in the near future as taxation moves online. The second takes a more fundamental perspective on tax policy and technology and speculates on what may ultimately occur.18

Policy Innovation

The focus of HMRC’s Making Tax Digital for individual taxation is the personal tax account that will provide real-time data on incomes, deductible expenses, tax payments, and tax credits. The simplest implication is that this moves the burden of the tax calculation from the individual (or the employer under pay-as-you-earn (PAYE) onto the revenue agency. It also removes the need for the current PAYE system, because the personal tax account could be linked directly to the individual’s bank account for regular payment of taxes. This would reduce the administrative burden on employers, but may not be advantageous for the revenue agency.19 A more significant advance from automating the calculation at the revenue agency is to permit greater complexity in the structure of marginal tax rates. Present systems that use a limited number of bands and marginal rates have no justification in tax theory, but reflect only computational convenience. They cannot provide the targeted incentives that many tax analysts consider justified (Mirrlees and others 2011) and can also create perverse incentives when combined with other features of the tax and benefit system.

If the system approaches anywhere close to using real-time data collection, then it becomes possible to extend a PAYE-like system to all taxpayers, including the self-employed. This can be implemented provided receipts of the self-employed can be matched to payments from third parties to confirm their tax status. This would remove the need for ex post payment of large sums of tax, easing the burden of payment and smoothing cash flow for the individual. Default would be less likely, which would reduce the need for the revenue agency to engage in chasing defaulters and in debt collection.

In addition, a system that tracked incomes and expenditures would ensure correct treatment of allowable expenses and, consequently, reduce errors. As revealed by the experiment reported in the previous section, such errors are commonplace in the completion of returns. For the United Kingdom, the latest figures for 2014-15 show that errors (£3.2 billion) and failure to take reasonable care (£5.5 billion) constituted almost a quarter of the £36 billion tax gap (HMRC 2016). This illustrates the potential benefits of an intelligent system that removes the possibility of error or ability not to take care. A truly advanced system could adjust tax charges according to the flow and the timing of income in recognition of the lumpiness of income for many self-employed.

The discussion of the experimental evidence made frequent reference to the compliance impact of pre-population and customer service. Noncompliance is a significant issue for all tax agencies, so it is important to consider if digitalization might prove beneficial in this respect. The noncompliant population (for income tax, similar comments apply to other taxes) can be broken broadly into deliberate cheats who report a false income level, moonlighters who report income from one or more jobs legitimately but have other income from additional employment that is not reported, and “ghosts” who simply do not appear in the system.20 Our discussion has mostly focused on the impact of pre-population and customer service upon the first two groups. If digitalization increases the information received by the revenue agency from third parties, then it will necessarily lead to an eventual reduction in noncompliance. We say “eventual” because the experiments have revealed that pre-population that is incomplete or incorrect can act as a signal of limited information and encourage noncompliance.

Ghosts are the most difficult group for a tax agency to monitor and control. HMRC is unlikely to be the only tax agency that has very limited data on the extent of the ghost problem. Since they are, by definition, outside the official system, there is no tax record to even form the basis of an investigation. It is with this group that digitalization holds the most promise for increasing compliance. The growth of digital records coupled with the linking of records will ensure that it becomes increasingly difficult to avoid leaving a digital fingerprint somewhere in the system. There may be no tax records, but it is hard to avoid a birth, school, medical, or welfare record, or in some cases a criminal record. If all systems were linked then the absence of a tax record for an individual could be easily flagged by the system and act as a marker for investigation. This may not directly affect the motivation behind becoming a ghost, but it reduces the probability that a ghost can continue unobserved.

The discussion so far has identified some minor revisions to the operation of the U.K. system. We now explore the potential for digitalization to permit fundamental changes to the way in which individuals are taxed.

Before proceeding to this, it is worth noting that the first impact of successful digitalization may be implementation of the current system as intended. Noncompliance coupled with auditing and punishment result in the effective tax system being significantly different from the intended system. Hashimzade and others (2015) show that the effect of noncompliance is to create a group of taxpayers who pay very low effective rates of tax (the noncompliant who are not detected), a group who pay the correct rate of tax (the compliant), and a final group who pay high effective rates of tax (the noncompliant who are caught and eventually pay the correct tax plus a fine). It is very difficult to conceive of any scenario in which this would be the intended outcome of tax policy design. Hence, digitalization which reduces noncompliance can ensure the intended system is more closely implemented.

The significant issues of whether digitalization can support any major changes in tax policy and whether policy can be improved by fundamental change are now addressed. Linking currently separate data systems can allow for policy innovation in addition to the better administration already noted. Under current arrangements a revenue service receives a flow of data about a taxpayer’s income that is simply stored until the time at which the annual tax return is compiled. The revenue service may hold other data—such as residential address, or sex—but this is can be of limited value for tax design. What we have in mind here are potentially valid reasons for differentiating personal taxes according to individual characteristics. Arguments have been advanced that a lower marginal rate of tax on people aged over 65 will encourage them to remain productive in the workforce, and that lower average rates can help overcome disincentives to labor force participation for females with young children or others who face high fixed costs of work.21 Both variations from the standard tax schedule are possible without digitalization, but would be administratively easier if operated through a personal tax account using data that are already held in administrative databases that could be linked to tax data. The benefits of digitalization for capital taxes, corporate taxes, and value-added tax are explored in Chapter 2.

A further benefit of linking datasets is that it makes possible the seamless integration of taxation and benefits. Many examples have been presented of how the interaction of the tax system and the benefit system can result in the creation of perverse incentives. This is particularly a problem in systems, such as in the United Kingdom, that apply tax at the individual level but allocate benefits at the family level. The benefit of digitalization and linking of datasets is that the system can be administered in close to real time with an online environment easing the input of updated information.22 This cannot remove all the conflicts caused by individual/family distinctions but can lead to some alleviation. Pushing this further, the current system of tax credits could be developed into a fully fledged negative income tax system, with all benefits collapsed into a negative income tax that was fully and automatically responsive to changes in circumstances. As examples, linking tax data with educational data could deliver a reduction in tax in the month before a child starts senior school to assist a family with related school expenditures, while linking with health data could automatically adjust the tax level without further eligibility testing. In brief, linkage would make possible the automatic implementation of a range of targeted assistance without the need for testing of eligibility.

The argument that a consumption tax is preferable to an income tax because it does not distort the saving decision has a long history in public finance. Meade (1978) was a forceful proponent of the idea and the arguments were reinforced in Mirrlees and others (2012). A flat consumption tax can be implemented by a uniform and comprehensive value-added tax.23 A non-uniform value-added tax permits some progression in the consumption tax if budget shares for commodities are correlated with individual characteristics. However, if the correlation is weak the progression will be poorly targeted. Meade (1978) demonstrated that a progressive consumption tax could be implemented if income and contributions to eligible saving instruments were recorded. The consumption tax could then be levied on the difference between income and eligible savings with, potentially, any chosen degree of progressivity. The drawback with this approach is that it requires annual assessment to determine the tax base, and so runs into the problems of payment difficulty and default that withholding schemes such UK PAYE are designed to avoid.

The limits to what can be achieved by digitalization are met when the implementation of a consumption tax is considered. For people in employment, the flow of income is fairly smooth and predictable so a consumption tax can be implemented (approximately) using a withholding tax based on either actual saving in recorded assets (accepting the lumpiness in tax payments this may cause) or a presumptive level of saving (which would smooth tax payment). An annual adjustment would be required unless actual saving was sufficiently smooth (or equal to presumptive saving if this method were used), so annual interaction with the revenue service would remain necessary.

Further progress meets with a fundamental difficulty even if comprehensive data on purchases were linked to income data. This difficulty is that many significant household purchases are very lumpy (such as the purchase of a house or car) even when the resulting flow of consumption is smooth. Implementing a consumption tax based on observed purchases would tax expenditures but not the flow of consumption. This is why the arguments made in Chapter 2 applied only to perishable consumption goods. To tax the latter would require imputation of the flow of consumption since it is not directly observed. The housing services tax proposed in Mirrlees and others (2011) proposed using housing rents to measure consumption flow, and similar methods could be used for other goods, so the problems are not insurmountable. The main point is that digitalization in itself does little to assist with this practical difficulty.

The Foundations of Tax Policy

The fundamental question for tax policy is why do we want to tax? The answer determines what we would want to tax if there were no limitations on the design of the system. This determines the nature of the ideal tax system. How we are able to tax is determined by the available technology for tax administration. The theory of optimal tax design studies the nature of the tax system that emerges as the best attainable approximation of the ideal tax system.24 The underlying premise of the economic theory of tax policy is that individuals have unalterable personal characteristics, some of which are unobservable, but make observable market transactions. The ideal tax system for equity purposes would be based on the immutable personal characteristics that generate differences in economic potential between people. Remarkably, using these characteristics as the tax base is also the most efficient way to tax: there is no change in behavior that can reduce the tax burden and, hence, there is no deadweight loss.

By definition, an unobservable personal characteristic cannot be used directly as the tax base. The imperfect tax system that is implemented has instead to be based on the observable personal characteristics—some of which may not be relevant for determining economic potential—and observable market transactions. Using transactions for the tax base causes two sources of deviation from the ideal. First, the observed transactions may be imperfectly correlated with unobservable personal characteristics. Second, an incentive can be created to change transactions to reduce tax liability, giving rise to a deadweight loss. These ideas were first clearly expressed by Mirrlees (1971) in his seminal study of income taxation and have become the foundation of tax theory. The models of tax theory focus on differences in endowments and preferences and explore the nature of the optimum tax systems that emerge. One general conclusion of the theory is that we should compensate for differences in endowments but not in choices.25 Another way to express this is that the role of the tax system is to achieve the equalization of economic potential. What people choose to do with their economic potential should not affect the design of the tax system. For example, if two individuals have the same level of labor market skill, but one chooses to work while the other does not, then there is no justification for redistributing income between the two. Hence, it is argued that it is economic potential that matters, not choices.

When this view of the world is pushed into a practical interpretation some difficulties start to emerge. The model assumes that economic potential is a fundamental and unalterable characteristic. This cannot be the case since potential reflects both ability and training, so it seems natural to search for something deeper that determines ability.26 It is here that we currently run into difficulties because of our incomplete understanding of what makes one person a talented musician and another a talented swimmer. Superficially, it is possible to point at various physical traits but the real question is what determines these traits. At present our conceptualization of the unobservable personal characteristics as an endowment of “ability” reflects our current ignorance. Only when we reach a position where the true underlying source of differences are understood can we proceed with the implementation of an ideal tax.

The Future

Current modeling in tax design is founded on the assumption that there are unalterable personal characteristics that determine economic potential. When some, or all, of these characteristics are unobservable the tax system has to tax observed market transactions as a proxy. The question is, will technological advances make currently unobservable characteristics observable and, hence, allow the implementation of novel taxes?

To implement the ideal tax system we need to determine what the relevant characteristics are and how these characteristics determine economic potential. These two requirements are equally important, and the first alone is not sufficient. For example, suppose we conclude that what matters for economic potential is an individual’s genetic code. Current technology allows us to read the genetic code at reasonably modest cost. What we must also possess for this reading to be of any value for tax purposes is knowledge of the link between the genetic code and economic potential. Such knowledge—with the possible exception of some weak correlations—is almost entirely absent at present. Without it we cannot use our current knowledge of the genetic code to progress any deeper with our tax theory.

Putting aside current limits on knowledge and technology, it is interesting to engage in speculation about potential consequences of technology. For the sake of argument assume individual economic potential is determined by genetics alone.27 It is possible that research will eventually unlock the genetic code and identify the mapping from genetics to economic potential. In the context of the discussion of tax theory above, this will make genetic makeup the personal characteristic on which the ideal tax system should be based. The interpretation of this reasoning is that behind the veil of ignorance all individuals are a genetic blank canvas. We know what outcome will be achieved by each set of genetics and this determines how we should redistribute. Crossing the veil of ignorance then assigns a genetic structure of each individual and the tax policy is then implemented. As observed in Logue and Slemrod (2008), the tax system will then impose redistribution from those with genetics linked to economic success to those with less successful genetics to a degree that is merited by social perceptions of equity.28 This may be extreme but it is where we are led by following our existing representation of the optimal tax problem.

However, we have not reached the end of the story. The difficulty for the application of tax theory is that—even with current technology—genetics can be changed. The development of CRISPR and other gene splicing techniques already allows the replacement of sections of the genetic code. There can be no doubt that these techniques will advance in the future and become more accurate, even to the point where the genetic code is entirely a matter of choice. Although the legal system in the United States and many other countries does not currently allow the modification of the human genome, this is a position that will prove very difficult to sustain. It might seem an extreme claim, but if the human genome can be modified, then it is no longer an unchangeable characteristic. We can conceive of parents selecting the genetics of children based on a range of factors from among which tax implications cannot be excluded. Basing the tax system on genetics in a situation in which genetics can be modified then creates a new and disturbing direction in which tax policy can have a distortionary impact on behavior. We would lose any notion of there being an ideal and non-distortionary tax system and have to face the consequences of taxation potentially influencing the genetic mix of the population.

The conclusion of this discussion is that technological advance may fundamentally impact our conception of how tax policy is formulated. We may reach a point at which there are no unalterable characteristics that determine economic potential. Instead, economic potential may be a matter of choice through genetic design. If this position is ever achieved the current foundations for optimal tax theory no longer apply. There will be no unchangeable characteristics, so there will be no ideal and non-distortionary tax system. The achievement of technology may just be to push the margin at which taxation is distortionary to another level.


The implementation of digital technology for tax administration has proceeded at different rates across countries. Some countries have been quick to adopt new technologies and others, including the United Kingdom and the United States, have been more cautious. Consequently, in these countries digital technology has had little to no impact on tax policy beyond the possibilities opened up for data mining to improve management information. The reluctance is understandable given the impact that unforeseen consequences may have on compliance and revenues.

The chapter has described how economic experiments can be used to test the impact of new digital platforms. The results of the experiments are not always as expected, but can be understood when interpreted using behavioral economics. The revenue service is an embodiment of authority, which explains the reluctance of experimental subjects to alter pre-populated values; but a conception of fairness will lead those same subjects to obtain compensation by exaggerating expenses. Digital systems should of course be tested exhaustively for technical functioning before implementation. We believe the results of the experiments provide strong grounds for advocating that digital systems also be thoroughly tested for behavioral impacts.

We have also looked ahead to speculate on how digitalization may impact tax policy. Digitalization has considerable promise for allowing the implementation of tax systems that would not be possible without it. This is particularly true when administrative data sets can be linked to fully exploit the potential of the information that is held. When we explore what future developments in technology can achieve, it becomes clear that some fundamental questions concerning the foundations of tax policy have to be resolved. Our current theory is based on current constraints on policy. In particular, existing tax theory judges potential tax systems by how they perform relative to the ideal system that would be used in the absence of constraints on the observation of economic potential. How we might want to tax if technology can relax these constraints requires a significant re-imagining of the theory.


    Alm, J.2012. “Measuring, Explaining, and Controlling Tax Evasion: Lessons from Theory, Experiments, and Field Studies.” International Tax and Public Finance19 (1): 5477.

    Alm, J., K. M.Bloomquist, and M.McKee. 2015. “On the External Validity of Laboratory Tax Compliance Experiments.” Economic Inquiry53 (2): 117086.

    Alm, J., T. L.Cherry, M.Jones, and M.McKee. 2010. “Taxpayer Information Assistance Services and Tax Compliance Behaviour.” Journal of Economic Psychology31 (4): 57786.

    Alm, J., T L.Cherry, M.Jones, and M.McKee. 2012. “Social Programs as Positive Inducements for Tax Participation.” Journal of Economic Behaviour and Organization84 (1): 8596.

    Alm, J., M. B.Cronshaw, and M.McKee. 1993. “Tax Compliance with Endogenous Audit Selection Rules.” KYKLOS46 (1): 2745.

    Alm, J., B. R.Jackson, and M.McKee. 1993. “Fiscal Exchange, Collective Decision Institutions, and Tax Compliance.” Journal of Economic Behaviour and Organization22 (3): 285303.

    Alm, J., and M.McKee. 2004. “Tax Compliance as a Coordination Game.” Journal of Economic Behaviour and Organization54: 297312.

    Alm, J., M.McKee, and WBeck. 1990. “Amazing Grace: Tax Amnesties and Compliance.” National Tax Journal43 (1): 2337.

    Banks, J., and P. A.Diamond. 2010. “The Base for Direct Taxation.” In Dimensions of Tax Design. Oxford: University Press.

    Bazart, C, and M.Pickhardt. 2011. “Fighting Income Tax Evasion with Positive Rewards.” Public Finance Review39 (1): 12449.

    Brimer, D., M.Jones, M.McKee, and C. A.Vossler. 2015. “Tax Reporting Behavior: Underreporting Opportunities and Prepopulated Tax Returns.” Working Paper 15–11, Department of Economics, Appalachian State University, Boone, NC.

    Choo, C. Y. L., M. A.Fonseca, and G. D.Myles. 2016. “Do Students Behave Like Real Taxpayers in the Lab? Evidence from a Real Effort Tax Compliance Experiment.” Journal of Economic Behaviour and Organisation124: 10214.

    Clotfelter, C. T.1983. “Tax Evasion and Tax Rates: An Analysis of Individual Returns.” The Review of Economics and Statistics65 (3): 36373.

    Collins, J. H., and D. R.Plumlee. 1991. “The Taxpayer s Labor and Reporting Decision: The Effect of Audit Schemes.” The Accounting Review66 (3): 55976.

    Coricelli, C, M.Joffily, C.Montmarquette, and M.-C.Villeval. 2010. “Cheating, Emotions, and Rationality: An Experiment on Tax Evasion.” Experimental Economics13 (2): 22647.

    Fonseca, M. A., and S. B.Grimshaw. Forthcoming. “Do Behavioral Nudges in Pre-Populated Tax Forms Affect Compliance? Experimental Evidence with Real Taxpayers.” Journal of Public Policy and Marketing.

    Fortin, B., G.Lacroix, and M. C.Villeval. 2007. “Tax Evasion and Social Interactions.” Journal of Public Economics91 (8): 2089112.

    Graetz, M. J., and L. L.Wilde. 1985. “The Economics of Tax Compliance: Fact and Fantasy.” National Tax Journal. 38: 35563.

    Hashimzade, N., G. D.Myles, F H.Page, and M.Rabien. 2015. “The Use of Agent-Based Modelling to Investigate Tax Compliance.” Economics of Governance16 (2): 14364.

    Hashimzade, N., G. D.Myles, and B.Tran-Nam. 2013. “Applications of Behavioural Economics to Tax Evasion.” Journal of Economic Surveys27 (5): 94177.

    HM Revenue Commission (HMRC). 2016. “Measuring Tax Gaps 2016 Edition: Tax Gap Estimates.” London.

    International Monetary Fund (IMF). 2015. “Current Challenges in Revenue Mobilization: Improving Tax Compliance.” IMF Staff Report, Washington, DC.

    Kotakorpi, K., and J. P.Laamanen. 2016. “Prefilled Income Tax Returns and Tax Compliance: Evidence from a Natural Experiment.” Tampere Economic Working Paper, School of Management, University of Tampere, Finland.

    Ledyard, J. O.1995. “Public Goods: A Survey of Experimental Research.” In Handbook of Experimental Economics, edited by J.Kagel and A.Roth. Princeton, NJ: Princeton University Press.

    Lipsey, R. G.1979. An Introduction to Positive Economics. London: Weidenfeld and Nicholson.

    Logue, K., and J.Slemrod. 2008. “Genes as Tags: The Tax Implications of Widely Available Genetic Information.” National Tax Journal61 (4): 84363.

    McKee, M., C.Siladke, and C. A.Vossler. 2011. “Behavioural Dynamics of Tax Compliance Under an Information Services Initiative.” MPRA Working Paper 38865, University Library of Munich, Munich.

    Meade, J. E.1978. The Structure and Reform of Direct Taxation. London: George Allen & Unwin.

    Mirrlees, J. A.1971. “An Exploration in the Theory of Optimum Income Taxation.” Review of Economic Studies38 (2): 175208.

    Mirrlees, J., S.Adam, TBesley, R.Blundell, S.Bond, R.Chote, M.Gammie, P.Johnson, G.Myles, and J.Poterba. 2011. Tax by Design. Oxford: Oxford University Press for the Institute for Fiscal Studies.

    Mirrlees, J., S.Adam, TBesley, R.Blundell, S.Bond, R.Chote, M.Gammie, P.Johnson, G.Myles, and J.Poterba. 2012. “The Mirrlees Review: A Proposal for Systematic Tax Reform.” National Tax Journal65 (3): 65584.

    National Audit Office. 2016. The Quality of Service for Personal Taxpayers. London: National Audit Office.

    Organisation for Economic Co-operation and Development (OECD). 2008. “Third Party Reporting Arrangements and Pre-Filled Tax Returns: The Danish and Swedish Approaches.” Paris.

    Shu, L. L., N.Mazar, F.Gino, D.Ariely, and M. H.Bazerman. 2012. “Signing at the Beginning Makes Ethics Salient and Decreases Dishonest Self-Reports in Comparison to Signing at the End.” Proceedings of the National Academy of Sciences109 (38): 15197200.

    Slemrod, J., and C.Gillitzer. 2013. Tax Systems. Cambridge, MA: MIT Press.

    Starmer, C. 1999. “Experiments in Economics: Should We Trust the Dismal Scientists in White Coats?Journal of Economic Methodology6 (1): 130.

    Sunstein, C. R., and R. H.Thaler. 2009. Nudge: Improving Decisions about Health, Wealth and Happiness. New York: Penguin.

    Tan, R., and A.Yim. 2014., “Can Strategic Uncertainty Help Deter Tax Evasion? An Experiment on Auditing Rules.” Journal of Economic Psychology40: 16174.

    Vossler, C. A., and M.McKee. 2013. “Efficient Tax Reporting: The Effects of Taxpayer Information Services.” Working Paper 13–24, Department of Economics, Appalachian State University, Boone, NC.

Annex 5.1 Tax Profile for the Experiment

In this experiment, you will take the role of Tom, a self-employed fitness instructor. During this tax year, you have earned ECU 25,200 of income from running fitness classes. You are in the process of completing your tax return form, and need to decide what expenses to claim as tax allowances.

Your files show the following for this tax year.

Secondhandcar sales receipt 06-Apr-2014
CO2 emission165g/km
TotalECU 1,500
Year 2014–15
Personal journeys2,000
Travel between home and class8,000
  • 1. You bought a secondhand car to help you get to and from your classes.

  • 2. Here is the receipt for the purchase of your car and a summary of mileage, fuel, servicing expenses, and insurance costs.

  • 3. You run your fitness classes every evening in a local church hall, which you paid ECU 5,760 to hire.

  • 4. You paid ECU 175 for printing flyers to advertise your fitness classes.

  • 5. You paid ECU 1,200 for a gym membership to stay fit.

  • 6. Your household bills amounted to ECU 7,500 for annual rent, gas, electricity, water rates, and council tax. You spend about one day a month (12 days a year) working from your home (a studio flat) designing posters and leaflets about the classes, calling new members and dealing with the finances and administration.

  • 7. Your mobile phone bills were ECU 420; only 15 percent of total usage was for business purposes.

Annex 5.2 Analysis of Errors

Analysis of specific fields yields interesting results. Subjects appear to have taken time to complete the information given to them in the profile and the tax guidance, but not fully able to file a correct tax report. As a first example, the correct value to enter for phone costs was 238 ECU, because the flyer costs of 175 were appropriate for this category as was 15 percent of the 420 ECU mobile phone bill (63 ECU). Figure 5.2.1 shows that the majority of values entered reflect these numbers in some way.

Figure 5.2.1.Subjects Entering Particular Values for Phone Costs

Source: Authors’ calculations.

More subjects put 63 as the value in the LONG guidance-based treatments than in the SHORT guidance based treatment, where the response 238 was more popular suggesting that the correct field to enter the flyer costs into was more clear in the SHORT guidance. An offsetting value of 175 for the flyers can clearly be seen in the filings made for Other Expenses, shown in Figure 5.2.2.

Figure 5.2.2.Proportion of Subjects Entering Particular Values for Other Expenses

Source: Authors’ calculations.

A third example can be seen for values filed under rent. The correct value for this category was 6,007 ECU, comprised of 5,760 ECU for hire of the church hall and (12/365)*7,500 (247) as the appropriate value for use of the home for business purposes. Figure 5.2.3 shows the proportions of subjects filing particular values for rent by treatment.

Figure 5.2.3.Proportion of Subjects Entering Particular Values for Rent

Source: Authors’ calculations.

The pattern in Figure 5.2.3 for rent is similar to that shown in Figure 5.2.1 for phone costs in that the majority of values entered reflect a combination of the raw values and calculations, though some are wildly wrong, such as the value 13,260, which simply sums the value for church hall hire with the household rent bill. The higher proportion filing the correct value 6,007 in the SHORT guidance based treatment than for the LONG guidance treatment suggests that the mechanism for handling household rent was more apparent in the SHORT guidance. The figure of 6,385 arises as subjects (incorrectly) divide the household rent (7,500) by 12 and add that to the 5,760 figure for church hall hire.

A final example is shown in Figure 5.2.4 for travel expenses. The actual correct value of travel expenses was zero as the use of the car to drive to and from the same place of work does not qualify as a taxable expense. The values reported in this field are, however, informative of subject behavior. Subjects were informed of a purchase of a car for 1,500 ECU and running costs of 2,500 ECU, 80 percent of which were for business purposed. The range of values filed includes 2,000 ECU (80 percent of running costs); 2,500 (the full running costs); 3,500 (80 percent of running costs plus purchase cost); 3,600 from application of simplified costs; 4,000 (total cost of car); and 6,100 from simplified costs plus running costs.

Figure 5.2.4.Proportion of Subjects Entering Particular Values for Travel Expenses

Source: Authors’ calculations.

Once again, the figure suggests that subjects were working with the profile and the tax guidance but not quite able to get to the correct result. Notably, in all cases, the values used typically skew to over-claiming on expenses, as reflected in the previous results. It should also be noted however that this is designed into the profile, as there are items that subjects are required to exclude and therefore we cannot say from the results here that such over-claiming would apply more generally.

The authors thank Andy Morrison, Floria Hau, Andrea Scott, and Tim Bryant at the National Audit Office (United Kingdom) for their help, in particular, their guidance and expertise on the relevant tax issues and efforts to produce the profile used. They also thank Michael Keen for his extensive comments on an earlier draft.

The experiment reviewed in the section “Taxpayer Guidance: Popup or Paper?” was financed by a contract with the U.K. National Audit Office. The description of the results in this chapter are the authors’ own and do not necessarily reflect those of the audit office.

A tax return is pre-populated when the tax administration enters data into the return before sending it to the taxpayer. Tax returns are now pre-populated to varying degrees in more than 10 European Union countries, as well as in Australia and California (OECD 2008).

Shu and others (2012) report the outcome of an experiment designed to reduce cheating on exams.

A list of systems that either failed, ran over budget, or have not yet been delivered is available at

The chapter focus narrowly on how administrative systems directly affect compliance. Many more factors can affect compliance (see IMF 2015).

Tax compliance experiments have examined the effects of several policies: amnesties (Alm, McKee, and Beck 1990), audit schemes (Collins and Plumlee 1991; Alm, Cronshaw, and McKee 1993; Alm and McKee 2004; Tan and Yim 2014), publicizing information about audits and those audited (Coricelli and others 2010; Fortin, Lacroix, and Villeval 2007; Alm, Bloomquist, and McKee 2015), and positive inducements to encourage tax fling and compliance (Alm and others 2012; Bazart and Pickhardt 2011).

The center is operated in partnership with the University of Exeter and the Institute for Fiscal Studies. More information is available at

A quasi-feld experiment engages a relevant population (in this case, taxpayers) in a laboratory or online experiment.

The precise rule relating the probability of audit (denoted p) to the declared liability (denoted X) was: p = 10 percent for X ≤ 22,600 ECU, p = 6.6 percent for 22,600 ECU < X < 42,500 ECU, p = 3.3 percent for 42,500 ECU ≤ X.

This section describes an experiment on tax compliance behavior undertaken online and at the Finance and Economics Experiments Laboratory at Exeter at the University of Exeter between January and April 2016. The research was funded under contract by the National Audit Office.

The costs and benefits here refer to not only the monetary costs and benefits taxpayers may incur when seeking help from HMRC, but also to psychological costs and benefits.

The experimental design left subjects uncertain about the correct value of a tax deduction and a tax credit. In the basic treatment, decisions had to be taken with this uncertainty unresolved. The treatment with a “customer-friendly” revenue service introduced an information service that resolved this uncertainty.

Approximately 15 percent of the U.K. workforce are self-employed.

2,353.92 = 6,145.00 – 3,791.08.

A detailed development of this argument into a system-based perspective on taxation can be found in Slemrod and Gillitzer (2013).

See also the discussion of these issues in Chapter 2.

Under the present U.K. system of PAYE, employers provide an unpaid tax collection and enforcement service.

This discussion does not cover groups engaged in criminal fraud.

The U.K. Child Tax Credit and Working Tax Credit introduced in April 2003 were based on annual assessment. A change in family circumstances over the year could result in overpayment and a consequent demand from HMRC for repayment. In the first year of operation, approximately one-third of claimants were overpaid a total of £1.9 billion.

Flat here meaning the same marginal tax rate on all consumption with no exemption.

The following discussion does not consider equality of opportunity. This could also be a motive for a tax-transfer scheme, but a direct solution would always be preferable.

Banks and Diamond (2010) explores this argument.

Extensive literature debates whether “genius” stems from natural ability or from hard work. It seems natural to believe it takes both.

Considerable literature on genius debates the relative importance of ability and training in explaining exceptional performers. By focusing on potential, it is possible to sidestep this debate.

There are obviously many practical issues being glossed over. But it should be noted that we are not necessarily discussing a one-of lifetime transfer that would require knowledge of the future value of alternative genetics. Instead, the taxes could be annual and matched each year with the current value of genetics.

    Other Resources Citing This Publication