June 2017 Marco Hernandez, Julian Jamison, Ewa Korczyc, Nina Mazar and Roberto Sormani © 2017 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 15 14 13 12 This work is the product of the staff of the World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Execu- tive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immuni- ties of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following condi- tions: Translations - If you create a translation of this work, please add the following disclaimer along with the attribu- tion: This translation is an adaptation of an original work by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptation – If you create an adaptation of this work, please add the following disclaimer along with the attribu- tion: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adapta- tion are the sole responsibility of the authors of the adaptation and are not endorsed by The World Bank. CONTENTS 4 Acknowledgements 5 Overview 8 Background 10 Experimental Design 10 Methodology 11 Randomization 12 Treatments 16 Results 16 Which letters were most and least effective in increasing compliance among taxpayers in arrears? 18 Was there a difference in effect between soft- and hard-tone messages? 20 Do different groups of taxpayers react differently to the contents of letters? 23 Testing the Delivery Method 25 Cost-benefit analysis 27 Conclusions 28 References 30 Annex 1. Applying Behavioral Insights to Tax Communications in Poland: Evidence from a Personal Income Tax Letter Trial 32 Annex 2. Main Regression Model Results: Significance in Terms of the Regular Dunning Letter 33 Annex 3. Main Regression Model Results: Significance in Terms of the Standard Behavioral Letter 34 Annex 4. Treatment Effects (Versus the Standard Dunning Letter) of Different Letters by Taxpayer Age 35 Annex 5. Treatment Effects (Versus Standard Dunning Letter) of Different Letters by Initial Tax Liability 36 Annex 6. Does the Cut-Off Date Matter? 38 Annex 7. Control Letters and Experiment Letter Variant 60 Annex 8. Sample Execution Order ACKNOWLEDGEMENTS This report has been prepared at the request of the (Lead Economist), Marina Wes (Country Director), Government of Poland. We are exceedingly grateful and Lalita Moorty (Practice Manager). to the Ministry of Finance of Poland, the Tax Chamber in Zielona Góra, and the staff from tax chambers and The authors thank Anne Brockmeyer, Tasmia Rahman, offices across Poland. Emilia Skrok, Varun Gauri, and the World Bank semi- nar and conference participants for helpful comments. The report was prepared by a World Bank team com- prised of Marco Hernandez, Julian Jamison, Ewa Kor- The findings, interpretations, and conclusions expres- czyc, Nina Mazar, and Roberto Sormani, who worked sed in this paper do not necessarily reflect the views of under the guidance of Arup Banerji (Country Direc- the World Bank, its executive directors, or the govern- tor), Carlos Pinerua (Country Manager), Ivailo Izvorski ments they represent. 4 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION This paper summarizes the results of a randomized OVERVIEW controlled trial (RCT) that used letters to remind tax- payers in Poland to pay their taxes. These taxpayers had declared their personal income tax (PIT) for the 2015 fiscal year but had failed to pay what they owed by the deadline, April 30, 2016 (i.e., taxpayers in arrears). Mobilizing domestic revenues efficiently is a priori- The trial took place between May and August 2016 and ty for the Government of Poland, but it is not easy. covered a total of 149,925 individual taxpayers.1 There are numerous instruments that can be used to achieve this objective. Traditional measures to The trial had two objectives: (1) test the effect of boost government revenues include changes to the tax different behavioral messages on income tax comp- legislation and reforms in the area of tax administra- liance, and (2) test whether how the letters were deli- tion. Such measures can have a large fiscal impact, vered (regular versus registered mail) had an effect but are often politically challenging to design and on compliance. Taxpayers were randomly assigned negotiate, and can take time to implement. Beha- to receive the official enforcement letter (a ‘dunning vioral interventions often focus on adapting existing letter’) used by the Polish Tax Office or one of nine systems and processes and can thus be implemented letters that were adapted using behavioral design. relatively quickly and at a low cost. Overall, they are an additional tool in the policy toolkit that country Figure 1 summarizes the design of the experiment. authorities have to improve tax compliance, and Of the nine adapted letters, one was a standard beha- thus complement but do not substitute traditional vioral letter that was clear, concise, and conformed measures to establish effective tax collection systems to many of the best practices from the tax-compliance including changes in tax legislations and tax admi- literature. The other eight letters were identical to nistration reforms. Behavioral interventions can the first except that to the plain reminder each added also help the Tax Authority to align its strategy more a different persuasive message meant to prompt accurately to taxpayer behavior. The Polish authori- action by the recipients. The messages expressed eco- ties were interested in applying insights from beha- nomic, moral, pride, self-image, and reciprocity con- vioral economics to their communications with tax- cepts. The ranking of behavioral letters with respect payers to see if making small changes could promote to the payment rate and payment amount is presen- tax compliance. ted in Figure 2. Figure 1. Sample Size and Treatment Assignments Source: Data from the Polish authorities. 1 This trial built on lessons from a pilot experiment in 2015, when the World Bank and the UK Behavioural Insights Team supported the authorities in designing and conducting a trial in two regions, Lubuskie and Wielkopolskie, which compared the results from one behavioral letter with those from the dunning letter. EXPERIMENTAL EVIDENCE FROM POLAND 5 Figure 2. Key Results Payment Rate Payment Amount Source: Data from the Polish authorities. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. To test whether sending a letter by regular or by number of compliant taxpayers. Similarly, all be- registered mail affected tax compliance outcomes, havioral letters mobilized higher average PIT pay- an additional group of taxpayers was sent the original ments than the control group. As with the payment dunning letter by registered mail. rate, the letter that combined messages about the omission and deterrence was the most effective Five key messages emerged from this experiment: in mobilizing higher payments. The average payment amount per letter was PLN 347 more than the dun- Message 1: Behavioral letters significantly ning letter sent by regular mail. improved tax compliance relative to the dunning letter Message 2: ‘Hard-tone’ messages were more effective than ‘soft-tone’ messages Behavioral letters helped to achieve higher rates of pay- ment and higher payment amounts. Of taxpayers who Taxpayers reacted more to messages that had a har- received the dunning letter, 40.2 percent paid. Of those der tone than to those that had a milder, softer, tone. who received the behavioral letters, 42.9 to 48.6 percent In this report, the hard-tone messages are deterrent paid. messages that highlight sanctions for noncompliance or that frame nonpayment as an intentional and de- The most successful behavioral letter had a deter- liberate choice. By contrast, the soft-tone messages rence message and framed nonpayment not as over- highlight social incentives, such as public goods or sight but as an intentional and deliberate choice. social norms. Overall, hard-tone messages were more This letter yielded an increase in the payment rate of effective – taxpayers were more likely to pay the 8.4 percentage points (pp) above the dunning letter income tax due, pay higher amounts of taxes, and – a 20.8 percent increase in the number of compliant reduce their tax liabilities (their tax debt).2 For taxpayers. The least effective behavioral letter had example, the payment rate among those taxpayers a public goods message noting how not paying taxes who received a soft-tone message was 43.9 percent affected the provision of preschools, schools, roads, (9 percent higher than the payment rate of 40.2 per- and safety. While this was the least effective beha- cent in the control group who received the original vioral letter, it still significantly increased the pay- dunning letter), but the payment rate for those tax- ment rate over the dunning letter by 6.7 percent. payers that received a hard-tone message was The payment rate for taxpayers who were sent the 46.9 percent (18 percent higher than for the control standard behavioral letter was 46.3 percent – 6.1 pp group). With regard to payment amount, on average more than for recipients of the dunning letter. Such the hard-tone messages mobilized more (PLN 93) than a change represents a 15.2 percent increase in the soft-tone messages. 2 In fact, as shown further in the report, adding some of the social messages actually reduces the impact of the letter compared to the standard behavioral letter. 6 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION Message 3: The effectiveness of some of the messages depended on taxpayers’ characteristics The effect of a particular message varied depending on the taxpayer’s age, gender, geographic location, and amount owed. For example, the public goods mes- sage helped to increase tax compliance among people aged 20–29 but had a negative effect on the compliance of those aged 50–64. Given the public goods listed in the behavioral letter (preschools, schools, roads, and safety), one possibility is that younger people responded positively because they are more likely to benefit from these than older people and are thus more motivated to pay up when reminded of the public goods benefits. If the heterogeneity observed is in fact due to the choice of public goods mentioned, higher compliance might be induced by adapting the types of public goods mentioned in letters to the age groups of recipients. Similarly, public good messages work better in rural areas, possibly because of a tighter link with the outcome of that kind of spending and the relevance to community. Message 4: Sending letters by regular mail (the cheaper option) proved to be just as effective as sending them via registered mail Analysis of the outcomes for the delivery-method test found no statistical difference in payment rates or payment amounts whether the dunning letter was hand-delivered by registered mail or sent by regular mail. This suggests that in Poland the tax admini- stration (and other agencies) could generate useful savings by replacing use of registered mail (PLN 14.96 per letter) to invoice taxpayers in arrears with delivery by regular mail (PLN 9.20 per letter). Message 5: Tax compliance among taxpayers in arrears can be tackled cost-effectively A cost-benefit analysis revealed that if the most effective letter had been sent to all 149,925 taxpayers in the sample, the tax authority would have gene- rated PLN 39,328,742 more in revenues than the control group – the additional revenues would have been 28 times larger than the cost of sending the letters, including staff time.3 The trial results make a compelling case for rigorous testing of small adap- tations of government communications with tax- payers. They also demonstrate that communication processes can be enhanced to promote tax comp- liance at low cost and without the need for new legislation. 3 The Polish Tax Department estimates that the cost of sending a dunning letter by regular mail, including staff time costs, is approximately PLN 9.20 sending it by registered mail would cost about PLN 14.96. EXPERIMENTAL EVIDENCE FROM POLAND 7 been found to enhance the effectiveness of public po- BACKGROUND licy because people think automatically and socially and often use mental models that are unconscious – that is, they use heuristics and shortcuts that do not always apply in a given context. They also tend to think in terms of stories or narratives rather than Mobilizing domestic revenues efficiently is a prio- data points. These observations can help policy- rity for the Government of Poland. In the past two makers better align their communication strategies years there has been heightened emphasis on expan- with the behavior of citizens.4 ding social programs in a fiscally sustainable way – i.e., financed largely by increased revenue. The go- Behavioral interventions to promote tax comp- vernment is therefore implementing fiscal measures liance have proven to be cost-effective since they aimed at expanding the revenue base and also is look- generally involve modifications to existing systems ing into its own processes and procedures which can and processes. In recent years the impact of beha- boost tax compliance. In the area of tax policy, in 2016 vioral interventions has been measured using rando- the Government introduced a new tax on financial mized controlled trials (RCTs), which are now widely institutions and increased the progressivity of the used in medicine, business, and international deve- PIT. In March 2017 it started the implementation of lopment. For example, the effectiveness of tax re- a comprehensive reform of the tax administration minders has been tested using RCTs in a number in order to integrate tax and customs offices. These of countries including Australia, Argentina, Austria, traditional measures are often politically challenging Chile, Costa Rica, Denmark, Germany, Guatemala, to design and negotiate and can take time to bring Israel, Peru, Switzerland, the United States, the Uni- tangible results. As they pursued tax policy and tax ted Kingdom, and Venezuela.5 A common feature of administration reforms, the Polish authorities deci- these RCTs is that they are based on data already ded to see whether applying insights from behavioral collected by the tax authorities, and as such they economics to their communications with taxpayers were implemented quickly and at low cost. Hence, – making small changes to regular processes – could recent experience from other countries suggests that promote tax compliance quickly and at low cost. behavioral interventions can be easily implemented, replicated and scaled-up, stimulating a process of In recent years, tax authorities in different coun- continuous learning and enhancement of admini- tries have begun to experiment with different types strative procedures. of communications (letters, emails, SMSs, web- sites) using insights from the behavioral science Recent research has concentrated on using diffe- literature to persuade taxpayers to pay what they rent behavioral messages to incentivize higher tax owe in taxes. International evidence suggests that compliance. Seminal research by the UK Behavioral behaviorally-informed reminders can affect tax beha- Insights Team (BIT) and the tax authority in the UK vior, at least in some contexts. Results from a number had originally shown the efficacy of sending taxpayers of trials have shown that applying behavioral insights letters with behavioral messages. Some of the beha- to tax communications can help to promote tax comp- vioral messages tested invoked social norms, such as liance and both raise tax revenues and reduce admi- how many people pay on time, or associated taxes nistrative costs. In these trials, tax authorities have with gaining or losing public goods. Relative to the tested a variety of reminder messages highlighting control group, the former increased payment of de- patriotic motives for paying taxes, social norms, pos- clared tax liabilities by up to 5.1 pp within 23 days of sible sanctions, or information from third parties. delivery, and the latter (whether framing loss or gain) increased payments by 1.6 pp (Hallsworth et al. 2014). The World Bank has also been exploring the appli- More recently, the World Bank worked closely with cations of behaviorally-informed policies. Its 2015 tax authorities in Guatemala to design better commu- World Development Report, “Mind, Society, and Beha- nication strategies using these and other behavioral vior,” noted that applying behavioral insights had insights. Letters sent to Guatemalans who had failed 4 The report can be found at: http://www.worldbank.org/en/publication/wdr2015. 5 See Blumenthal et al. 2001; Wenzel 2005; Torgler 2007; Kleven et al. 2011; Ariel 2012; Castro and Scartascini 2013; Del Carpio 2013; Ortega and Sanguinetti 2013; Pomeranz 2013; Dwenger et al. 2014; Hallsworth et al., 2014; Kettle et al. 2016, and Brockmeyer et al. (2016). 8 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION to declare their income taxes in 2014 showed that, to around PLN45 billion). The majority of PIT is paid in 11 weeks, the letter highlighting a social norm in- in the form of monthly advances by the employers creased the average amount paid per taxpayer by (around 85 percent of total PIT, i.e. PLN 39 billion). $13.97 (210 percent), and the deliberate choice letter Towards the end of the year, the employee is obliged (omission vs. commission) brought in $17.95 more to submit a PIT statement to the tax office for the final (269 percent) than not sending a reminder letter. PIT resettlement. At this stage the outstanding liabi- Remarkably, compliance by those who received these lity is paid or the tax office returns the paid tax due two letters was also high in the next fiscal year (Kettle to the use of tax breaks or tax credits. In 2015, the to- et al. 2016, Brockmeyer et al. 2016) document similar tal tax liability to be paid by tax payers in end April findings using email reminders in Costa Rica. Another (for the fiscal year 2014) was around PLN 6 billion, field experiment in Argentina sent behavioral letters out of which around 30 percent is not collected by to payers of property tax with messages related either the deadline. The analysis of historical taxpayer comp- to levels of enforcement (deterrence), social norms, liance rates found that while the tax base has expan- or provision of public goods, and found deterrence ded since 2011, more and more taxpayers do not pay to be the most effective. While the latter two had no their taxes until after the statutory deadline, and the effects, the deterrence message increased compliance share of those who do not pay at all (10 months after by nearly 5 pp relative to the control group (Castro the deadline) has also been heading up. At the same and Scartascini 2013). Clearly, the effectiveness of time, the vast majority of taxpayers in arrears have behavioral letters can vary. One important reminder a very small tax liability (Figure 3 and 4). Such condi- from the behavioral science literature is that con- tions seemed conducive to behavioral interventions text matters greatly. Interventions that previously in the form of letters to remind taxpayers to pay. In this worked in one setting may fail in others. Thus, while context, in 2015, with the support of the World Bank evidence from other countries can provide insights and the UK Behavioural Insights Team, the Polish Tax to inform the design of behavioral interventions, there Administration carried out its first RCT trial. The expe- was still a need to experiment to see whether beha- riment targeted PIT taxpayers in arrears in two regions vioral insights can help promote tax compliance of the country and tested the impact of two reminder in Poland. letters on tax compliance outcomes. The results of this pilot trial are summarized in Annex 1. In this context, the Polish tax authorities decided to start testing if behavioral insights can strengthen This report summarizes the results of a nation- tax collection and promote higher tax compliance -wide RCT conducted in 2016 that used letters in Poland. Due to methodological considerations to remind taxpayers in Poland to pay their taxes. the authorities decided to focus its experiment on the The trial took place between May and August 2016 and personal income tax (PIT) and implemented a pilot covered 149,925 individual taxpayers, that is, the uni- experiment in 2015 to test the impact of behavioral verse of taxpayers in arrears in Poland. These tax- insights in promoting tax compliance. PIT revenues payers had declared their PIT for the 2015 fiscal year constitute around 17.3 percent of all tax revenues and but had failed to pay what they owed by the deadline correspond to around 2.5 percent of GDP (equivalent of April 30, 2016. Figure 3. Share of Taxes by Payment Date Figure 4. Share of Taxes Due Collected by the End of May Source: Polish Ministry of Finance. EXPERIMENTAL EVIDENCE FROM POLAND 9 liability (PLN 50 and above) and had not paid their EXPERIMENTAL income tax for 2015 in full by May 13, 2016. DESIGN Two types of impact were tested: • Impact of the content of the letters • Impact of the delivery method (registered vs. regu- lar mail) This RCT trial had two objectives: (1) to increase There were three outcomes related to PIT payments and identify key features of an effective • payment rate communication strategy; and (2) to test whether • payment amount the delivery method (registered vs. regular mail) had • tax liability. an impact on tax compliance. Outcome indicators were measured at three intervals: Taxpayers in arrears were defined as those who had 4, 8, and 12 weeks after the trial launch. filed their PIT declarations on time (by April 30, 2016) but had failed to pay their tax liability since. The trial Table 1 shows the trial schedule: involved all taxpayers that had non-negligible net tax Table 1. Key Dates for the Experiment Date Step May 13, 2016 Cut-off date for identifying the sample May 13–22, 2016 Randomization and preparation of letters May 23, 2016 Roll-out: all letters are sent to taxpayers June 13, 2016 End of the 1st monitoring round: until this date, there were no other attempts to contact the taxpayers covered by the trial (i.e., the tax authority did not follow up with taxpayers regarding their liability); under regular proceedings, tax offices undertake “soft execution” measures, which implies that they would contact taxpayers in arrears by phone, e-mail, text message, among other forms of communication. July 4, 2016 End of 2nd monitoring round August 8, 2016 End of 3rd monitoring round METHODOLOGY harder to interpret the findings of the experiment because other interventions occurred, such as tax The test measured three outcomes using anonymized office enforcement activities. For example, a number tax records for the 149,925 participants in the trial, of taxpayers from all treatment arms that failed updated as of June 13, 2016 (four weeks after the Tax to pay by June 13 were sent registered dunning let- Authority sent the letters). This sample covered all ters. Meanwhile executive proceedings were initiated taxpayers in arrears in Poland with liability above for taxpayers who had been sent the registered dun- 50 PLN; excluded were taxpayers with no liability, ning letter when the trial began. These factors could those for whom data were missing for the covariates have created confounding effects that might have values, and outliers.6 affected the validity of the estimates. By setting June 13 as the cut-off, the estimates of the treatment The cut-off date was chosen because until June 13, effects are not contaminated by other interventions 2016, the tax administration did not intend to take by the tax authority. Nonetheless, Annex 6 shows out- any other enforcement activities. After that date it is comes based on payment to the latest date available 6 Outliers were identified as observations that, in the OLS regression of payment amount on treatments and all covariates, had a residual of more than 2.5 standard errors. 10 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION as evidence suggesting the treatments had sustained impact of the delivery method on the payment rate effects. is estimated. Specification of the regression model is the same as in the previous comparison. The Polish Tax Office collected outcome variables regularly and automatically as administrative data. The outcomes of interest are these: RANDOMIZATION • Payment (binary): dummy variable equal to one if This was a randomized controlled trial. Randomiza- the taxpayer paid a non-zero amount of tax by the tion as a method of experimental control has been given date, and zero otherwise. extensively used in both human clinical trials and • Payment amount (continuous): the total amount biological experiments and in the social sciences paid by the taxpayer in PLN unconditional on pay- for tests ranging from financial education and tax ment. repayment to organ donation. Conceptually, this is • Log payment amount (continuous): the (log) amount a method for randomly allocating to one or more paid unconditional on payment. intervention conditions. Thus, it is possible to com- • Log of the outstanding tax liability (continuous): pare the effectiveness of different interventions the total (log) amount of the taxpayer’s liability by while minimizing both known and unknown factors the given date. that may influence the outcome being investigated. As a result, randomization produces comparable The primary estimation consists of an intention-to- groups and eliminates the source of bias in treat- -treat (ITT) analysis of the impact of the letters based ment assignments. on the following regression model: Analysis of historical data regarding the payment Yi = Tib + X i g + ui patterns of Polish taxpayers revealed important characteristics affecting the probability of payment. where Yi is the payment outcome, Ti is a treatment Based on administrative data for 2011–14 it was vector (T1= Dunning Letter (Control Group); T2= Baseline possible to identify the key determinants of non- Behavioral Letter, T3= Social Norm; T4= Public Good Posi- compliance by Polish PIT taxpayers (Figure 5). tive; T5= Public Good Negative, T6= Deterrence, T7= Deter- A number of factors had a statistically important rence + Execution Order; T8= Omission, T9= Omission + De- impact on compliance. For example, low tax liability terrence, T10= Omission Taxpayer-Perspective), X i is a vec- decreased the likelihood of payment by 13.5 percent tor of control variables comprising the characteristics and PIT-36L declaration decreased it by 17.6 per- of taxpayers (amount of tax liability, declaring child- ren, type of tax form, gender, age, total taxable in- come), and ui is the error term. Regression models Figure 5. Determinants of Noncompliance are evaluated by ordinary least squares (OLS) multiple by Polish Taxpayers, Percent regression. In the OLS regressions heteroscedasticity in ui is controlled for using robust standard errors. To investigate the impact of the contents of letters, first the behavioral letters are compared to the control group, which was sent a dunning letter (via regular mail). In this case, stars on the figures (*) indicate a significant difference from the dunning letter.7 Different behavioral letters are then compared to the baseline behavioral letter to see which behavioral messages were most effective in encouraging tax- payers to comply. Statistical significance against the baseline behavioral letter is indicated by daggers (‡). The analysis mainly compares separate letters and their performance compared to the control group or the baseline behavioral letter. To investigate the impact on tax compliance of the delivery method (registered or regular mail), one treat- Source: Polish authorities. Note: Analysis refers to 2011–14 period. Bars above the axis indicate ment group received a standard dunning letter deli- determinants that lower tax compliance; bars below indicate vered by registered mail. In the separate model the determinants that improve tax compliance. 7 The number of stars indicates significance difference from the control group: *=0.1, **=0.05, ***=0.01; daggers indicate significance difference from the standard behavioral letter †=0.1, ‡=0.05, ‡‡=0.01. EXPERIMENTAL EVIDENCE FROM POLAND 11 cent. However, being female increased payment siveness. For this experiment, randomization was probability by 1.9 percent and declaring children conducted at the individual level and stratification increased it by 9.9 percent. was used for initial liability, tax code submitted, elec- tronic submission, gender, parental status, marital To ensure that the differences in taxpayer characte- status, voivodship, and the age group. Given the high ristics do not affect the experiment, the procedure dimensionality of the stratification, balance on each chosen was stratified randomization, which prevents of these variables could not be perfect, but as the imbalance between treatment groups for known balance tests show, the sample was balanced across factors that influence prognosis or treatment respon- the treatment groups.8 Figure 6. Balance Tests for the Sample Source: Polish authorities. Note: The difference between each treatment group and the control group in terms of covariates expressed in standard deviations. TREATMENTS The sample covered 144,053 taxpayers in arrears each of whom had a tax liability of more than PLN 50. Tax- The aim of this experiment was to check how different payers in this sample were randomly assigned to one letter content affects compliance measures: payment of 10 treatment groups, each of which received a letter rate, payment amount, and outstanding tax liability. delivered by regular mail (Figure 7). Figure 7. Sample Size and Treatment Assignments Source: Polish authorities. 8 Full randomization was not possible because the covariates were derived from a different data sample (all taxpayers vs. the late taxpayers in the trial sample). Other differences were tax liability (sample included only taxpayers who owed more than 50PLN at the date of the experiment), and outliers were removed. 12 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION Control Group: Dunning Letter (n=6,091): This is the and noncompliant behavior. It also provides contact original letter that the Polish Tax Office had been information for the relevant tax authority so tax- sending to remind taxpayers to pay their taxes. payers can notify or seek clarification quickly without The dunning letter is normally sent by registered mail needing to look up the phone number themselves but in the experiment was sent by regular mail. (a possible cognitive barrier). It thus can be seen as The letter has a formal tone and sets out the legal a plain reminder that addresses nonpayment due to basis for this type of communication.9 It states the tax forgetfulness or oversight. liability and asks the taxpayer to calculate the accrued interest; it provides general guidelines for the calcu- The other letters consist of the behavioral baseline lation but does not provide an example. The letter has letter augmented with a brief section that includes a deterrent message, highlighting that “failure to per- a behavioral message. In addition to the plain re- form [payment] by the indicated deadline shall result minders, the behavioral messages are meant to in referral of the case to execution proceedings, there- address different motives for nonpayment – econo- by generating costs of execution proceedings to be mic, moral, and so on. Here, for explanatory clarity, covered first.” the behavioral messages are provided in a box before the discussion of each. (Annex 7 contains the comp- The letter comes across as very formal, and the language lete letters in the original Polish and translated into is convoluted, bureaucratic, and legalistic. It might be English.) very difficult for people with average reading skills to understand. Similarly, the explanation of how to Social Norm calculate the interest rate calculation is difficult to follow. According to our records, [8] out of 10 residents Behavioral Baseline Letter (n=15,232): A substantial in [REGION OF THE TAXPAYER] have already literature is evolving on the application of behavioral paid their income tax for 2015. You are part of science to social policy (e.g., Sunstein 2015). Such a minority that has not yet fulfilled that duty. terms as defaults, status quo bias, peer effects, simp- licity, salience, immediate gratification, and recipro- city have entered the dialog of policy-makers. Pro- The Social Norm Letter (n=15,474), is based on the ob- minent in the list of applications is tax compliance: servation that people generally have a natural prefe- see Hallsworth et al. (2014) for an overview of early rence to do what their peers are doing. For instance, work in the United Kingdom and elsewhere. Gerber and Rogers (2009) found that informing citi- zens that 71 percent of their compatriots had voted The standard dunning letter was therefore rewritten in the previous election increased voter turnout. to conform to many of the best practices from pre- For taxes, Hallsworth et al. (2014) describe the suc- vious research: cessful use of social norms to improve compliance in the UK, as Kettle et al. (2016) did for Guatemala, • It begins with a very clear description of the pur- where although the true rate of payment is only pose of the letter and a “call-to-action” at the top, 64.5 percent, saying that in the letter increased both explaining what the taxpayer must do and giving rates of payments and average amounts paid. One a deadline: “Please pay your income tax by June 3, additional aspect of the letter used here is that it gives 2016.” the actual rates by region rather than just nation- • It minimizes formal legal language to the extent wide; some research findings indicate that the closer possible, partly so that recipients will understand the reference or comparison group, the stronger the better but partly just to encourage them to read it. effects of this type of norm (Goldstein et al. 2008). • It specifies very concrete next steps regarding what to do and how to do it. Public Good Positive • In a separate table, it lists the interest due on each day until the deadline so that taxpayers do not need to calculate it themselves and have additional Are you aware that 37.79% of your personal incentive to pay the liability before the deadline. income tax goes to your municipality? From this income, your municipality finances pre-schools, The letter now comes across as reader-friendly. schools, roads, and safety, benefiting everyone The sentences are much shorter, the language is in your municipality including yourself and your simple, and the messages are very clear. The letter family. Don’t be an irresponsible inhabitant of is highly prescriptive in terms of what is being asked your municipality and pay your delinquent taxes! and clearly outlines the consequences of compliant 9 The Resolution of the Ministry of Finance from December 30, 2015 specifies procedures for creditors of monetary claims (Journal of Laws of 2015, item 2367). EXPERIMENTAL EVIDENCE FROM POLAND 13 The Public Good Positive Letter (n=15,424) is based on the assumption that people who knew what expendi- paying them. As part of the execution procedures, tures are financed from tax revenues might be more we can, for example, block your bank account, likely to pay their overdue taxes. Therefore, one objec- salary, and, in addition, you will have to cover tive of this letter was simply to inform citizens how all execution expenses that arise. some of their taxes are spent and to remind them that many services they use are available only be- cause of tax revenues. This in itself is not necessarily The Deterrence Letter (n=15,442) uses the behavioral behavioral, but several other elements incorporate letter with the addition of the deterrence message. behavioral science knowledge. The accurate and un- This message serves three goals. The first sentence usually precise figure of 37.79% focuses attention on is meant to evoke a negative self-conscious emotion what comes next, unlike a generic claim about taxes of guilt, which has been proven to be a powerful providing benefits. It also suggests that the govern- mediator to motivate moral action (Hoffman 1982a, ment is tracking everything closely. The link to muni- 1982b; H. B. Lewis 1971; M. Lewis 1993). The second cipal services brings everything closer to recipient sentence is intended to create a sense that the deter- and family. The whole effort stimulates feelings of rence threat is serious, and the third sentence gives reciprocity: one should do something in return for all concrete examples of the punishment actions. The the benefits being received. Finally, the last sentence examples are likely to evoke mental imagery that engages a perceived identity for the recipients as enhances realism and therefore behavioral intentions potentially responsible taxpayers rather than indi- (see, e.g., Miller and Marks 1997; Yoo and Kim 2014). viduals who are letting down their communities and Thus, together the last two sentences are meant neighbors; this pivots the message toward the sense to create a sense of fear of possible consequences, of self, not just the outward action. which can be an important cause of law-abiding or norm-respecting behavior (Haidt 2003). Public Good Negative Deterrence + Execution Order Are you aware that 37.79% of your personal income tax goes to your municipality? Without Not paying taxes places an unfair burden on all this income, your municipality cannot finance the taxpayers who have honestly fulfilled their pre-schools, schools, roads, and safety, damaging duty. We are therefore determined, more than everyone in your municipality including your- ever, to collect taxes from those who avoid pay- self and your family. Don’t be an irresponsible ing them. As part of the execution procedures, inhabitant of your municipality and pay your we can, for example, block your bank account delinquent taxes! or salary and in addition you will have to cover all execution expenses that arise. The Public Good Negative Letter (n=15,350) is based on We attach a sample Execution Order Form that the same assumption as the previous letter but we send to taxpayers who have not paid the taxes the behavioral framing is different. The motivation due. behind this “negative” letter was to harness all the stimulative aspects of the positive public good letter and in addition to frame a loss. It is well-established The Deterrence + Execution Order Letter (n=15,292) rein- that decision makers respond more strongly to per- forces the threat of punishment for noncompliance ceived losses related to a status quo than to perceived and makes the punishment more palpable by provi- gains. In this case the implicit threat is that without ding the actual Execution Order Form that non-comp- sufficient tax revenue, all the municipal benefits cur- liers receive.10 rently enjoyed may be lost. Omission vs. Commission Deterrence So far, we have thought of your payment delay Not paying taxes places an unfair burden on all to be accidental. However, if you disregard this other the taxpayers, who have honestly fulfilled notice, we will consider it an intentional choice their duty. We are therefore determined, more of yours and think of you as a dishonest tax- than ever, to collect taxes from those, who avoid payer. 10 The Execution Order is defined in the Resolution of the Ministry of Finance from May 16, 2014, (J. L. of 2014, item 650). The example of the Execution Order can be found here: http://www.mf.gov.pl/c/document_library/get_file?uuid=665fa5c2-0834-40fb-94f8-4395c687671e&groupId=764034 ; please also see Annex 8. 14 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION The Omission vs. Commission Letter (n=15,249) mes- Omission + Deterrence sage is that dishonesty is a deliberate choice. Most real decisions have a status quo alternative: doing nothing or maintaining one’s current or previous de- So far, we have thought of your payment delay cision. Numerous studies have found that indivi- to be accidental. However, if you disregard this duals tend disproportionately to stick with the status notice, we will consider it an intentional choice quo (Samuelson and Zeckhauser 1988 and Anderson of yours and we will treat you as a dishonest 2003), for two reasons: (1) the losses from acting may taxpayer. As part of the execution procedures, loom larger than the gains (Tversky and Kahneman we can, for example, block your bank account, 1981); and (2) moral violations tend to be judged less salary, and, in addition, you will have to cover harshly when the violation results from inaction all execution expenses that arise. rather than action (DeScioli et al. 2012). Individuals may use the ambiguity and uncertainty associated with acts of omission in order to minimize future The Omission + Deterrence Letter (n=15,238) is a variant psychological costs arising from the threat to self- of the behavioral letter. Both omission and deter- -image of acting dishonestly (Hallsworth 2013; Mazar rence interventions were included to test whether et al. 2008). there are additive effects if both are used together. That is, would combining the two interventions be This message is intended to overcome the status more effective than using each separately? quo bias toward doing nothing – by framing the dishonest behavior as a deliberate choice – an action Omission Taxpayer-Perspective in itself. By notifying the reader that failure to comply will be treated as a deliberate choice to be a dishonest taxpayer, this messages aims to elimi- So far, you might have thought of your payment nate omission as an excuse for noncompliance, thus delay to be accidental. However, if you disregard increasing both the cognitive dissonance around this notice, you should consider it an intentional the taxpayer’s self-image of an honest person and choice of yours and think of yourself as a dis- the perceived cost of paying later. The wording also honest taxpayer. gives taxpayers an exemption for not previously de- claring, which introduces an element of reciprocity, because the implication is that the taxpayers have The Omission Taxpayer-Perspective Letter (n=15,261) been granted a favor or shown good will. The text is a slight but important variation to the previous let- is also worded to give the impression that taxpayer ter. In particular, the perspective is changed from the behavior is being closely monitored, which may tax authority to the taxpayer. Recent research on moral heighten the perceived threat of subsequent actions judgment and decision making suggests that in ad- against noncompliance. Thus it acts as a mild deter- dition to the deterrence threat (whether that is mild rent. A similar approach was shown to be effective or strong), individuals care about their moral self- in Guatemala (Kettle et al. 2016). -image – they want to retain a positive view of them- selves, and a threat to their moral self-image can be a significant motivator of honest behavior even beyond the deterrence threat (Mazar et al. 2008). This inter- vention was meant to test the strength of this in the context of tax payments (see also Shu et al., 2012). EXPERIMENTAL EVIDENCE FROM POLAND 15 Statistical significance versus the behavioral letter: RESULTS The standard behavioral letter was very effective at mobilizing taxpayers to pay their overdue PIT. With a payment rate of 46.3 percent, it outperformed several letters with specific behavioral messages. For example, the payment rate for the negative mes- The presentation of the trial results from the trial sage on public goods was 3.4 pp. below the standard is organized in terms of the following questions: behavioral letter, implying that 7.3 percent fewer re- cipients paid their income tax than in recipients of • Which letters were most and least effective in increasing the standard behavioral letter. The difference in the compliance among taxpayers in arrears? payment rates is statistically significant at 5 percent • Was there a difference in effect between soft- and hard- level. Similarly, the recipients of the social norm, -tone messages? positive message on public goods, and deterrence • Do different groups of taxpayers react differently to the letters were less likely to pay up than those receiving contents of letters? the standard behavioral letter. One possible explanation for the relatively poor per- WHICH LETTERS formance of public goods letters may relate to the low WERE MOST AND LEAST levels of satisfaction with certain Polish public servi- EFFECTIVE IN INCREASING ces. According to OECD data from 2013, confidence/ satisfaction s with the national government was only COMPLIANCE AMONG 25 percent, much lower than the OECD average of TAXPAYERS IN ARREARS? 42 percent (OECD 2015).11. These low levels may be due to characteristic peculiar to public services, or they Payment Rate (Figure 8): may simply be a broad reflection of moving away from the collectivist outlook in a post-communist setting. Statistical significance versus the control group: All the behavioral letters (with a payment rate of (with pay- ment rates of 42.9 to 48.6 percent) were more effective Figure 8. Payment Rates of Letters Tested at bringing in payments than the control group dun- ning letter (40.2 percent). • The letter combining omission and deterrence mes- sages was the most effective, with a payment rate of 48.6 percent, an 8.37 pp increase over the dunning letter sent via regular mail and a 20.8 percent in- crease in the number of compliant taxpayers. • The worst-performing of the behavioral letters in terms of raising the payment rate was the nega- tive message on public goods letter, but statistically its 42.9 percent payment rate was still significantly better than the 40.2 percent rate of the dunning letter. • The payment rate for taxpayers sent the standard behavioral letter was 46.3 percent – 6.12 pp more Source: Ministry of Finance data. than for recipients of the dunning letter, and Notes: Error bars indicate 95% confidence intervals. Stars indicate 15.22 percent increase in the number of compliant a significant difference from the control group (dunning letter) taxpayers. – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. 11 According to the 2015 OECD Government at a Glance Report for Poland, the percentage of citizens satisfied or confident about public services was 25 percent for the national government, 36 percent for the judicial system, 43 percent for health care, and 59 percent for education. 16 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION The results of the trial also show that despite being sages was the most effective: the average amount already effective in mobilizing payments, the stan- paid per letter was PLN 1,469 – PLN 347 above the dard behavioral letter can be improved. For example, dunning letter sent via regular mail. In logarithm the inclusion of messages that eliminate omission terms, this letter increases taxpayer payments by as an excuse for noncompliance (as in Omission, Omis- 25.9 percent. sion + Deterrence, and Omission Taxpayer-Perspective • Although the negative message on public goods was letters) significantly improves the results. Specifically, again the worst-performing letter, it was still signi- the omission letters perform significantly better than ficantly better than the dunning letter, bringing the behavioral letter (47.4 percent vs. 46.3 percent). in an average amount per letter that was PLN 122 The effect of omission framed from the taxpayer’s higher than the control group. point of view is smaller but still statistically significant. • The standard behavioral letter was second-best While including the deterrence message does not im- in mobilizing higher payments, bringing in an ave- prove on the standard behavioral letter, there is a signi- rage of PLN 1,371, which was PLN 249 higher than ficant interaction when it is paired with the omission the dunning letter. In logarithms this difference letter that produces a 2.3 pp improvement in the pay- it implies a 19.3 percent increase over the dunning ment rate over the standard behavioral letter. regular mail letter in the amount paid. Payment Amount (Figure 9): Statistical significance versus the baseline behavioral letter: Most of the other behavioral letters yield sta- The impact of the treatments on the payment amount tistically similar results as the baseline behavioral is estimated using the effect in terms of (1) the PLN letter in terms of payment amount. Only the omission amount paid, and (2) the logarithm (log) of the amount + deterrence letter attracts payments that are close to paid by the taxpayer. Both models yield similar results. PLN 100 higher. The other omissions letters, which offer an excuse for noncompliance (omission and the Statistical significance versus the control group: dunning omission taxpayer perspective) do not yield any letter: Compared to the control group, each behavioral significant improvement on payment amounts over letter letters mobilized higher average payments for the standard behavioral message, despite increasing its entire sample (including those who did not pay the compliance rate. This suggests that the omission as well as those who did). While in the control group message encourages payment mostly among tax- the average payment per letter was PLN 1,123, for payers with a smaller liability; combining it with the behavioral letters the average ranged from PLN 1,244 threat (omission + deterrence) is very effective for (10.8 percent higher) to PLN 1,469 (30.9 percent higher). taxpayers who owe more. Statistically, public good Both are statistically significant. letters, both positive and negative frame, are signi- ficantly worse than the baseline behavioral letter. • As with the analysis of the payment rate, the letter The average payment per letter was PLN 81 for the that combined the omission and deterrence mes- negative and PLN 127 for the positive public letters. Figure 9. Payment Amounts Brought in In PLN In Log Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. EXPERIMENTAL EVIDENCE FROM POLAND 17 Outstanding Tax Liability (Figure 10): Figure 10. Tax Liability after Response to Test Statistical significance versus the control group: dun- ning letter: All behavioral letters were also effective at reducing the tax liability (i.e., debt) among tax- payers in arrears. The largest reduction, 19.3 percent, was in response to the omission + deterrence letter; at the other end of the spectrum, compared to the control group the negative public good letter reduced the debt outstanding by only 5.6 percent. The baseline behavioral letter lowered the tax liability on average by 14.6 percent as compared to the control group. Statistical significance versus the behavioral letter: Among behavioral letters statistically only omission + deterrence letter was significantly more effective (by 4.8 percent) than the baseline behavior letter in reducing the tax liability. Among those who ended up with higher tax liability were taxpayers who re- Source: Ministry of Finance data. ceived the public good negative message (higher by Notes: Error bars indicate 95% confidence intervals. Stars indicate 9 percent) and the social norms message (higher by a significant difference from the control group (dunning letter) 6 percent) compared to recipients of the behavioral – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. letter. WAS THERE A DIFFERENCE IN EFFECT BETWEEN were analyzed separately. All behavioral messages used in the experiment can be grouped in two catego- SOFT- AND HARD-TONE ries: (1) soft-tone and (2) hard-tone. The soft-tone MESSAGES? messages correspond to social incentives, while the hard-tone messages put emphasis on deterrence Besides the analysis of the impact of individual letters messages. Table 2 shows the composition of the on outcome measures, distinctive groups of letters groups. Table 2. Classification of Letters by Content Letter Contents Tone Social Norm According to our records, [8] out of 10 residents in [REGION OF THE Soft-tone TAXPAYER] have already paid their income tax for 2015. You are part of a minority that has not yet fulfilled that duty. Public Good Positive Are you aware that 37.79% of your personal income tax goes to your Soft-tone municipality? From this income, your municipality finances preschools, schools, roads, and safety, benefiting everyone in your municipality including yourself and your family. Don’t be an irresponsible inhabitant of your municipality and pay your delinquent taxes! Public Good Are you aware that 37.79% of your personal income tax goes to your Soft-tone Negative municipality? Without this income, your municipality cannot finance preschools, schools, roads, and safety, damaging everyone in your municipality including yourself and your family. Don’t be an irresponsible inhabitant of your municipality and pay your delinquent taxes! Deterrence Not paying taxes places an unfair burden on all other taxpayers, Hard-tone who have honestly fulfilled their duty. We are therefore determined, more than ever, to collect taxes from those, who avoid paying them. As part of the execution procedures, we can, for example, block your bank account or salary, and, in addition, you will have to cover all execution expenses that arise. 18 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION Letter Contents Tone Deterrence Same as deterrence message above plus: Hard-tone + Execution Order We attach a sample Execution Order Form which we send to taxpayers that have not paid their taxes due. Omission So far, we have thought of your payment delay to be accidental. Hard-tone However, if you disregard this notice, we will consider it an intentional choice of yours and think of you as a dishonest taxpayer. Omission Same as omission message above plus: Hard-tone + Deterrence As part of the execution procedures, we can, for example, block your bank account, salary, and, in addition, you will have to cover all execution expenses that arise. Omission So far, you might have thought of your payment delay to be Hard-tone Taxpayer-Perspective accidental. However, if you disregard this notice, you should consider it an intentional choice of yours and think of yourself as a dishonest taxpayer. Overall, the hard-tone letters were more effective than Figure 11. Payment Rate by Letter Tone soft-tone letters in mobilizing payments. Not only did more taxpayers pay income tax due, they also paid higher amounts and their debt was reduced more. Payment rate (Figure 11): Compared to the dunning letter, the soft messages generally increase the pay- ment rate – 43.9 vs. 40.2 percent, a 9 percent increase in the number of compliant taxpayers. Hard-tone messages also perform significantly better than the standard dunning letter. Comparing soft-tone and hard-tone messages with the standard behavioral letter yields the following conclusions: • On average, soft-tone messages perform signifi- cantly worse than the standard behavioral letter Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate (43.8 vs. 46.3 percent). a significant difference from the control group (dunning letter) • Hard messages together slightly outperform the – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference standard behavioral letter, but the differences are from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. very small and not statistically significant. Payment amount (Figure 12): Both the standard beha- clearly better than the dunning letter, they were vioral and the hard-tone letters mobilized similar significantly worse than both the baseline behavioral amounts per letter – amounts significantly higher and the hard-tone messages and lowered the average than the average amount paid by recipients of the payment amount relative to the baseline behavioral dunning letter. While soft-tone messages were also letter by PLN 93 per letter. EXPERIMENTAL EVIDENCE FROM POLAND 19 Figure 12. Payment Amount Results by Message Tone In PLN In Log Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. Outstanding tax liability (Figure 13): The analysis of DO DIFFERENT GROUPS the outstanding tax liability yields similar results. Each type of behavioral letter significantly decreases OF TAXPAYERS REACT tax liability compared to the dunning letter control DIFFERENTLY TO THE group, but the soft-tone letters performing signifi- CONTENTS OF LETTERS? cantly worse than the hard-tone messages and the baseline behavioral letter. Heterogeneity (Figure 14 – Figure 17): The results de- scribed thus far reflect the average effects of treat- ment on taxpayer compliance outcomes and implicitly Figure 13. Outstanding Tax Liability assume the effect is similar for all taxpayers. But what by Message Tone is also very important is to investigate the degree to which the impact of a given treatment varies for specific geographic or sociodemographic groups of interest. Further analysis of the data by subgroup shows that in fact the treatment effects are hetero- geneous in terms of taxpayer characteristics – the observed intervention effects differ more from each other than might be expected due to random error (chance) alone. Compliance was investigated in relation to several characteristics: age, gender, family status (having versus not having children), geographic variation (urban versus rural), and the amount of tax liability. Overall, the results reported so far hold across the taxpayer groups in that the omission + deterrence letter was most successful and seems to be the best Source: Ministry of Finance data. nudging strategy. Nonetheless, some interventions do Notes: Error bars indicate 95% confidence intervals. Stars indicate work better for certain groups: a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate a significant difference from the standard behavioral letter - †= 0.1., ‡=0.05, ‡‡=0.01. AGE: Generally, all letters confirm that taxpayers in their mid-40s are most responsive, after which the difference gradually declines. However, the best-per- forming message (omission commission + deterrence) peaks higher among slightly older people, in their 50s, but the soft messages reduce compliance among older taxpayers (Annex 4). In terms of age, the response to the public good positive letter of two age groups 20 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION (20–29 and 50–64 years old) provides a good example Figure 14. Positive Public Good Letter of heterogeneity (Figure 14). The positive public good – Heterogeneity by Age Group, Percent messages had a positive effect among people aged Payment Rate 20–29 but a very negative effect among those aged 50–64 – although still better than the control group. The difference between the treatment effects in the two groups is 5.2 percentage points and is statistically significant at the 0.01 level. The heterogeneity in age groups is not entirely sur- prising given the public goods listed in the beha- vioral letter: preschools, schools, roads, and safety. Schools may be considered less relevant by people aged 50–64, who tend not to benefit from them directly; they may even have the opposite effect if people feel that their tax payments are going for Source: Ministry of Finance data. services that benefit others at the expense of services Notes: Error bars indicate 95% confidence intervals. Stars indicate that can benefit them. On the other hand, younger a significant difference from the control group (dunning letter) taxpayers may be more motivated to contribute to – *=0.1, **=0.05, ***=0.01; daggers indicate significant differences in the treatment effects (with respect to the standard behavioral education, since 85 percent of children in Poland letter) between the two groups: †= 0.1., ‡=0.05, ‡‡=0.01. attend public preschools (OECD 2013), and 59 percent of citizens indicated that they are satisfied or confident about the public education system (OECD 2015). If the heterogeneity observed is in fact GENDER: Women in general have a higher repay- due to the choice of public goods mentioned, higher ment rate and are more responsive to hard-tone compliance might be induced by varying the types messages, especially deterrence, omission commis- of public goods mentioned in letters sent to people sion, and omission commission + deterrence messa- in specific age groups. ges (Figure 15). Figure 15. Gender Heterogeneity, Percent Payment rate Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate significant differences in the treatment effects (with respect to the control group dunning letter) between the two groups: †= 0.1., ‡=0.05, ‡‡=0.01. FAMILY STATUS (having children): Parents (self- rence treatment, which has very similar payment -reported) are not very responsive to the messages rates. In general, hard-tone messages work on non- and, for them, the behavioral letter is the most effec- parents, but not on parents (Figure 16). tive, together with the omission commission + deter- EXPERIMENTAL EVIDENCE FROM POLAND 21 Figure 16. Family Status Heterogeneity, Percent Payment rate Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate significant differences in the treatment effects (with respect to the control group dunning letter) between the two groups: †= 0.1., ‡=0.05, ‡‡=0.01. LOCATION: Rural taxpayers repay more and the possibly because of a tighter link between the out- effects of all the treatments are quite similar. Inte- come of public good expenditures and its relevance to restingly, relative to the standard behavioral letter, the community (Figure 17). the public good message works better in rural areas, Figure 17. Location Heterogeneity, Percent Payment rate Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter) – *=0.1, **=0.05, ***=0.01; daggers indicate significant differences in the treatment effects (with respect to the control group dunning letter) between the two groups: †= 0.1., ‡=0.05, ‡‡=0.01. LIABILITY: Taxpayers who owe more seem to be The analysis of heterogeneity might be useful in con- more responsive to the omission + deterrence messa- templating the best nudging strategies for specific ge. (For specific results on liability-related hetero- segments of taxpayers. Further analysis of the data geneity, please see Annex 5). would be needed to understand the differences in tax- payer compliance according to their characteristics. 22 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION TESTING THE DELIVERY • For legal reasons a slight change of tone was intro- METHOD duced in the following sentence: Standard dunning letter delivered by registered There are two methods of delivering paper – commu- mail: “Failure to perform the said duty by indica- nication from the Polish tax office to taxpayers, regi- ted deadline shall result in referral of the case stered mail and regular mail. This part of the experi- to execution proceedings, thereby generating costs ment was designed to verify statistically if the method of execution proceedings to be covered first.” of delivery affects the payment rate, the payment amount, and the outstanding tax liability. The results Dunning letter delivered by regular mail: “Failure can inform the Polish authorities on the costs and to perform the said duty by indicated deadline may benefits of using both. result in referral of the case to execution pro- ceedings, thereby generating costs of execution In the registered mail delivery method, the Post Office proceedings to be covered first.” is obliged to deliver the letter in person within three working days after it is sent, though in fact it rarely The English and the original Polish versions of both takes that long. The taxpayer confirms receipt with letters can be found in Annex 7. a signature, and the tax office is notified that the letter was successfully delivered. If the taxpayer This trial had two arms: the standard dunning letter is not present at the time of delivery, the post office was sent to taxpayers by registered mail and the mo- leaves a notification and the taxpayer has seven days dified dunning letter by regular mail. from the first delivery attempt to pick up the letter at the post office.12 If the letter is not picked up within The sample covered 11,963 taxpayers in arrears each this period, the post office issues a second notification owing more than PLN 50. Taxpayers in the sample with another seven-day. After 14 days, a letter that were randomly assigned to either has not been picked up is returned to the tax office but is considered delivered because of the notifications. 1. a group that was sent the original dunning letter After that, all legal proceedings can be undertaken. by registered mail (n=5,872); or The process is highly formalized and registered 2. a group that was sent the modified dunning letter a dunning letter is a first step of execution pro- by regular mail (n=6,091). ceedings. The cost of sending the letter by registered mail is PLN 6.20 (PLN 14.96 including administra- Figure 18. Sample Size and Treatment tive costs). Assignments The regular mail delivery method is very simple; the post office delivers the letter to the mailbox of the taxpayer, normally within three working days after the sending date. The cost per letter is PLN 2.00 (PLN 9.20 including administrative costs). The content and looks of the dunning letter sent by Source: Polish authorities. regular mail and the standard dunning letter sent by registered mail were very similar. Small changes were introduced in the former to reflect the changed The analysis of the outcomes of the delivery method delivery method and the need to remove some phra- test found that the dunning letter sent by registered sing specific to the standard dunning letter: mail slightly outperformed the other letter, but the differences in outcomes are very small and statisti- • “Receipt confirmation” was removed as the delivery cally insignificant (Figure 19)13. Specifically: method changed. • Payment rate for the regular mail letter was • “Dunning Letter No: X/2016 dated…” was removed 40.2 percent, slightly below the 40.7 percent rate because the letter was no longer a formal dunning for the registered letter. letter and hence was not assigned a number. • Payment amount was PLN 1123 for the regular letter • “Dunning letter costs” was removed from the table and PLN 1159 for the registered letter, a difference of with information on liability and interest as the PLN36 per letter. not imposing any additional cost obligation on the • Similarly, differences in logs for both the payment taxpayer. amount and the outstanding tax liability were very • “Dunning letter” was replaced by “letter.” small and insignificant. 12 The taxpayer may also authorize another person to pick it up. 13 While the results show no statistical difference between the two delivery methods, further analysis may be needed to understand better the results, for example with regards to the number of dunning letters sent by registered mail and not delivered to the taxpayer. EXPERIMENTAL EVIDENCE FROM POLAND 23 Figure 19. Outcomes for Delivery Method Payment rate Payment amount (PLN) Payment amount (log) Outstanding liability (log) Source: Ministry of Finance data. Notes: Error bars indicate 95% confidence intervals. Stars indicate a significant difference from the control group (dunning letter). 24 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION relative to the dunning letter control group. If the COST-BENEFIT omission + deterrence letter had been sent to all 149,925 taxpayers in the sample, it would have ge- ANALYSIS nerated an estimated PLN 39,328,742 in additional revenues compared to the dunning letter control group. In other words, the omission + deterrence let- ter would have been generated 28 times more addi- tional revenue than sending the letters would have The purpose of the cost-benefit analysis is to demon- cost. The baseline behavioral letter would have strate the potential fiscal gains of the most effec- brought in an additional PLN 27,581,129 relative to the tive compliance strategy vis-a !-vis the baseline, i.e. control letter – almost 20 times more than the cost 14 the dunning letter. The cost-benefit analysis of the of sending the regular letters. treatments is based on the results from the earlier estimations on the payment rate and the amount It is worthwhile to add that currently the tax offices paid. The Polish Tax Authority estimates that the are required to follow up with the taxpayer using cost of sending each reminder letter by regular mail, a registered mail dunning letter. Considering that including costs of postage and staff time associated the trial found no statistically significant difference with preparing the letter is PLN 9.20. The cost of sen- in payment rate and payment amount between the ding each standard dunning letter by registered mail two types of delivery, the difference in the costs is PLN 14.96. The calculations are presented in Table 3 (PLN 14.96 and PLN 9.20) represents a loss for the tax and Table 4. offices. For the whole sample of 2015 taxpayers in arrears, the additional cost of sending registered The conclusion is that changing the delivery method rather than regular letters would be about PLN 863,568 from registered to regular mail can reduce the cost – 62 percent more than using regular mail. to the Tax Authority, and the use of more powerful messages (omission + deterrence letter) would have These calculations represent an assessment of po- increased the payment rate and hence the effective- tential benefits related to using the most effective ness of the taxpayer notification system. approach of notifying taxpayers. However, as this has not been applied in reality (the most effective The total cost of sending any letter by regular mail behavioral letters was not sent out to all taxpayers), to all 149,925 taxpayers in the sample would have been the revenue generated by the trial itself was also PLN 1,379,310. Regarding the benefits, the best-per- estimated: it brought in an estimated PLN 90,154,155 forming letter (omission + deterrence) significantly at a cost of PLN 1,413,133, thus generating PLN 88,741,023 increased the payment rate and the amount paid for the budget. 14 The cost and benefit analysis presented in the report does not take into account the additional costs associated with handling the complaints filed with the tax authorities by recipients of certain behavioral letters (mainly hard tone). The Ministry of Finance does not have data on the total number of complaints in Poland and the costs associated with their handling, hence they are not part of the cost-benefit analysis. Data from the Tax Chamber in Warsaw show that only 7 complaints were filed formally, while more than 30 thousand letters were sent in Mazowieckie region, which suggests that the cost associated with complaints was not very large. However, the employees of the Tax Offices also reported to have received over 3,000 phone calls and emails regarding the content of behavioral communications. In addition, it should be also noted that there might be other tools that can improve tax compliance apart from behaviorally informed letters, but identifying them is beyond the scope of the report. EXPERIMENTAL EVIDENCE FROM POLAND 25 Table 3. Cost-benefit analysis: scenarios Average payment Payment Net No of amount per rate, Unit cost, Total Benefit, benefit, taxpayers letter, PLN percent PLN cost PLN PLN Scenario 1: Using the most effective letter with regards to all taxpayers Omission commission, admin, deterrence 149,925 1469.3 48.6 9.20 1,379,310 107,003,434 105,624,124 Control group (dunning letter) 149,925 1122.7 40.2 9.20 1,379,310 67,674,693 66,295,383 Dunning registered 149,925 1159.0 40.7 14.96 2,242,878 70,696,466 68,453,588 Difference compared to control group – 39,328,742 39,328,742 Difference compared to dunning registered (863,568) 36,306,969 37,170,537 Scenario 2: Using standard behavioral letter with regards to all taxpayers Standard behavioral 149,925 1371.4 46.3 9.20 1,379,310 95,255,822 93,876,512 Difference compared to control group – 27,581,129 27,581,129 Difference compared to dunning registered (863,568) 24,559,356 25,422,924 Source: World Bank staff based on Ministry of Finance data. Table 4. Cost-benefit of the trial Average Benefit Net payment Payment Total from Net benefit No of amount per rate, Unit cost, cost of the trial, benefit, to cost taxpayers letter, PLN percent PLN the trial PLN PLN ratio Control group (dunning letter) 6,091 1122.7 40.2 9.20 56,037 2,749,418 2,693,381 48.1 Dunning registered 5,872 1159.0 40.7 14.96 87,845 2,768,915 2,681,070 30.5 Public goods (negative) 15,350 1289.9 42.9 9.20 141,220 8,496,603 8,355,383 59.2 Social norms 15,474 1299.7 43.7 9.20 142,361 8,795,341 8,652,980 60.8 Deterrence 15,442 1361.6 44.8 9.20 142,066 9,410,073 9,268,007 65.2 Public goods (positive) 15,424 1244.4 44.8 9.20 141,901 8,595,010 8,453,109 59.6 Standard behavioral 15,232 1371.4 46.3 9.20 140,134 9,677,750 9,537,616 68.1 Omission commission, taxpayer 15,261 1350.7 46.6 9.20 140,401 9,596,605 9,456,203 67.4 Deterrence with executive order 15,292 1342.0 47.1 9.20 140,686 9,673,336 9,532,649 67.8 Omission commission, admin 15,249 1317.9 47.3 9.20 140,291 9,515,544 9,375,253 66.8 Omission commission, admin, deterrence 15,238 1469.3 48.6 9.20 140,190 10,875,560 10,735,370 76.6 Total 149,925 1,413,133 90,154,155 88,741,023 62.8 Source: World Bank staff based on Ministry of Finance data. 26 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION of heterogeneity across taxpayer characteristics also CONCLUSIONS suggests that calibrating reminder messages to cer- tain groups of taxpayers could effectively increase outcome measures. Moreover, since there are no statistically significant This trial confirms that sending a letter to taxpayers differences in delivery method outcomes, the option in arrears works as an effective reminder to promote of sending reminders by regular mail can simplify compliance with Poland’s PIT laws. The results sug- execution procedures in Poland without affecting gest that adapting a behavioral design in taxpayer revenue collection. communications could produce considerable gains. All behavioral letters induced higher payment rates The trial results provide a compelling case for rigo- and amounts and decreased the amount of tax liabili- rous testing small adaptations to communications ty relative to both the control group (modified dun- to Polish taxpayers and demonstrate that without ning letter sent by regular mail) and the original dun- the need for new legislation current communication ning letter. Hard-tone messages proved very effec- processes can be enhanced at low cost to promote tax tive in mobilizing higher payments, but the evidence compliance. EXPERIMENTAL EVIDENCE FROM POLAND 27 REFERENCES Anderson, C., Keltner, D., & John, O. P. 2003. “Emotional convergence between people over time”. Journal of Personality and Social Psychology, 84, 1054–1068. Ariel, B. 2012. “Deterrence and Moral Persuasion Effects on Corporate Tax Compliance: Findings From A Ran- domized Controlled Trial.” Criminology, 50, 27–69. Blumenthal, M., Christian, C. and Slemrod, J. 2001. “Do Normative Appeals Affect Tax Compliance? Evidence From a Controlled Experiment In Minnesota.” National Tax Journal, 54, 125–136. Brockmeyer, Anne; Hernandez, Marco; Kettle, Stewart; Smith, Spencer Douglas. 2016. “Casting the Tax Net Wider: Experimental Evidence from Costa Rica.” World Bank Policy Research Working Paper; No. WPS 7850. Washing- ton, D.C.: World Bank Group. Castro, L. and Scartascini, C. 2013. “Tax Compliance and Enforcement in the Pampas: Evidence from a Field Experiment.” Inter-American Development Bank Working Paper Series. Washington, D.C.: Inter-American Development Bank. Castro L., and C. Scartascini. 2015. “Tax Compliance at Enforcement in the Pampas: Evidence from a Field Experiment.” Journal of Economic Behavior and Organization 116(C): 65–82. DeScioli, P., K. Asao, and R. Kurzban. 2012. “Omissions and Byproducts across Moral Domains.” PLOS ONE 7(10): e46963. Del Carpio, L. 2013. “Are the Neighbours Cheating? Evidence from a Social Norm Experiment on Property Taxes in Peru.” Princeton University Working Paper. Dwenger, N., Kleven, H., Rasul, I. and Rincke, J. 2014. “Extrinsic and Intrinsic Motivations for Tax Compliance: Evidence From A Field Experiment In Germany.” Mimeo. Gerber, A.S., Rogers, T. 2009. Descriptive Social Norms and Motivation to Vote: Everybody’s Voting and so Should You. Journal of Politics 71 (1): 1–14. Haidt, J. 2003. “The moral emotions.” In Handbook of Affective Sciences, edited by R. J. Davidson, K. R. Scherer, and H. H. Goldsmith, 852–70. Oxford, UK: Oxford University Press. Hallsworth, M. 2013. “The Omission Bias.” Unpublished paper. Hallsworth, M., List, J. A., Metcalfe, R. D., and Vlaev, I. (2014). “The Behavioralist as Tax Collector: Using Natural Field Experiments to Enhance Tax Compliance” (No. W20007). National Bureau of Economic Research. Hoffman, M. L. 1982a. “Affect and Moral Development. In New Directions for Child Development: Emotional Development edited by D. Ciccetti and P. Hesse, No. 16. San Francisco: Jossey-Bass. Hoffman, M. L. 1982b. “Development of Prosocial Motivation: Empathy and Guilt. In The Development of Prosocial Behavior, edited by N. Eisenberg. New York: Academic Press. Kettle, Stewart; Hernandez, Marco; Ruda, Simon; Sanders, Michael. 2016. “Behavioral interventions in tax comp- liance: evidence from Guatemala.” World Bank Policy Research Working Paper; No. WPS 7690. Washington, D.C. World Bank Group. 28 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION Kleven, H. J., Knudsen, M. B., Kreiner, C. T., Pedersen, S. and Saez, E. 2011. “Unwilling or Unable to Cheat? Evidence from a Tax Audit Experiment in Denmark.” Econometrica, 79, 651–692. Lewis, H. B. 1971. Shame and Guilt in Neurosis. New York: International Universities Press. Lewis, M. 1993. “Self-conscious Emotions: Embarrassment, Pride, Shame, and Guilt. In Handbook of emotions, edited by M. Lewis and J. Haviland, 563–573. New York: Guilford Press. Mazar, N., O. Amir, and D. Ariely. 2008. “The Dishonesty of Honest People: A Theory of Self-Concept Mainte- nance.” Journal of Marketing Research 45(6): 633–44. Miller, D. W., and L. J. Marks. 1997. “The Effects of Imagery-evoking Radio Advertising Strategies on Effective Responses.” Psychology & Marketing 14(4): 35–50. OECD. 2013. “Country Note – Education at a Glance 2013: Poland.” Paris: OECD. OECD. 2015, “Country Fact Sheet – Government at a Glance 2015: Poland.” Paris: OECD. Ortega, D. and Sanguinetti, P. (2013). “Deterrence and Reciprocity Effects on Tax Compliance: Experimental Evidence from Venezuela.” Working Paper. Pomeranz, D. 2013. “No taxation without information: Deterrence and self-enforcement in the value added tax.” (No. w19199). National Bureau of Economic Research. Samuelson, W., and R. Zeckhauser. 1988. “Status Quo Bias in Decision Making.” Journal of Risk and Uncertainty 1(1): 7–59. Shu, Lisa L., Nina Mazar, Francesca Gino, Dan Ariely, and Max 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 Sciences 109 (38): 15197–00. Sunstein, C.R. 2015. “Choosing Not to Choose.” Oxford: Oxford University Press. Torgler, B. (2007). “Tax Compliance and Tax Morale: A Theoretical And Empirical Analysis”, Cheltenham, UK: Edward Elgar Publishing. Tversky, A., and D. Kahneman. 1981. “The Framing of Decisions and the Psychology of Choice.” Science 211(4481): 453–58. Yoo, J., and M. Kim. 2014. “The Effects of Online Product Presentation on Consumer Responses: A Mental Imagery Perspective.” Journal of Business Research 67(11): 2464–72. Wenzel, M. 2005. “Misperceptions of Social Norms about Tax Compliance: From Theory to Intervention.” Journal of Economic Psychology, 26, 862–883. EXPERIMENTAL EVIDENCE FROM POLAND 29 ANNEX 1. APPLYING BEHAVIORAL INSIGHTS TO TAX COMMUNICATIONS IN POLAND: EVIDENCE FROM A PERSONAL INCOME TAX LETTER TRIAL1 Experiment Design: The goal of the trial was to in- cluded a social norms message: “According to our crease tax payments among taxpayers in arrears2 records, [percent] % of residents in [the Taxpayer’s of the Personal Income Tax (PIT) in two regions Region] have already paid their income tax for 2014”; in Poland: Lubuskie and Wielkopolskie.3 The sample (3) provided personal information about the amount covered 31,929 taxpayers in arrears that each had due, and had a table that calculated the interest pay- a tax liability of more than PLN 50. Taxpayers in this ment that would be due related to the date when sample were randomly assigned to one of three the payment was made; (4) explained how to make groups: (1) a control group that was not sent a re- a payment; (5) included a deterrent message: “If you minder letter (n=10,627); (2) a group that was sent do not pay your taxes, you could be subject to admi- the original Polish Tax Office reminder letter, stan- nistrative proceedings…. We will be checking to see dard dunning letter (n=10,534); and (3) a group that how you respond to this letter”; and (6) informed was sent a letter adapted using behavioral design taxpayers what to do if they were unable to pay the (n=10,768). full tax liability at that time. Here is a brief summary of each letter: Results: Results are reported for four outcome measu- res: payment rate, payment amount, outstanding tax • Dunning Letter (n=10,534): This is the original let- liability, and payment delay. ter, sent by registered mail, that the Polish Tax Office regularly sends to taxpayers in arrears • Payment rate (Figure A1.1): Both the original dun- to remind them to pay their taxes. The letter, ning letter and the behavioral letter were effective which has a formal tone, states the legal basis for at increasing the number of late taxpayers who this type of communication.4 It sets out informa- made a PIT payment. After 12 weeks, 27.6 percent tion about the tax liability and asks the taxpayer of taxpayers in the control group had done so. to calculate the accrued interests, providing ge- The payment rate for those who were sent the ori- neral guidelines for the calculation but not provi- ginal dunning letter was 29.8 percent, which is ding any example. The letter also has a deterrent 2.2 percentage points (pp) higher than the control message, highlighting that “failure to perform group. The behavioral letter was most effective [pay] by the indicated deadline shall result in re- at promoting tax compliance: the payment rate for ferral of the case to execution proceedings thereby those who received it was 32.2 percent, 4.6 pp higher generating costs of execution proceedings to be than the control group (i.e. a 17 percent increase covered first.” vis a vis the control group), and 2.4 pp higher than those that were sent the original dunning letter • Behavioral Letter (n=10,768): The original letter was (representing an 8 percent increase over the origi- adapted using behavioral design based on two core nal). The effects are statistically significant. principles: simplicity/clarity, and inclusion of beha- viorally-informed messages. Instead of being sent by • Payment amount (Figures A1.2): The behavioral registered mail, the behavioral letter went by regular letter proved to be the most effective in mobilizing mail. In terms of content, the behavioral letter (1) in- higher tax payment amounts relative to both the cluded a call-to-action at the top explaining what control group and the original dunning letter. the taxpayer needed to do within a defined period: Figure A1.2 shows the effect of the letters on the “Please pay your income tax by June 10, 2015”; (2) in- amount paid in PLN; the behavioral letter increased 1 Based on Applying Behavioral Insights to Tax Communications in Poland: Evidence from a Personal Income Tax Letter Trial, World Bank and Behavioural Insights Team, February 2016, draft available upon request from ekorczyc@worldbank.org or marcohernandez@worldbank.org 2 Taxpayers in arrears were defined as those who had filed their PIT declarations on time (by April 30, 2015) but had failed to pay their tax liability by the deadline. 3 Lubuskie and Wielkopolskie are located in the northwest of Poland. Together their populations totaled about 4.5 million in 2014, 12 percent of Poland’s total population. In 2014, there were about 1.9 million PIT taxpayers in the two regions, equivalent to 12.4 percent of total PIT taxpayers. 4 The Resolution of the Ministry of Finance of May 20, 2014 specifies procedures for creditors of monetary claims in undertaking to apply enforcement measures (Journal of Laws of 2014, item 656). 30 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION the average amount paid per letter by PLN 169 rela- Figure A1.1. Taxpayers Who Paid their tive to the control group (equivalent to a 15 percent Personal Income Tax after 12 Weeks, increase) and by PLN 225 (a 21 percent increase) Percent compared to the original dunning letter. • Outstanding tax liability (Figure A1.3): The beha- vioral letter was also effective at reducing the out- standing tax liability (the debt) among taxpayers in arrears. On average, the group who received the behavioral letter had a 4 percent lower tax lia- bility than recipients of the original dunning letter – a statistically significant difference. • Timing of payment: Both letters were effective in bringing payments in earlier, which reduces Source: Polish Tax Department, Ministry of Finance, and World Bank estimates. administrative costs. On average, taxpayers who Notes: Error bars indicate 95% confidence intervals. Stars indicate were sent a reminder letter paid their PIT 16 days a significant difference from the control group (at the 0.10, 0.05, earlier than the control group. and 0.01 levels) and daggers (†) indicate a significant difference between the two treatments. Figure A1.2. Average Amount Paid, PLN Figure A1.3. Outstanding (log) Tax Liability, Percent Source: Polish Tax Department, Ministry of Finance, and World Bank estimates. Notes: Error bars indicate 95 percent confidence intervals. Stars indicate a significant difference from the control group (at the 0.10, 0.05, and 0.01 levels) and daggers (†) a significant difference between the two treatments. Conclusions: This trial found that sending a letter sent out. The long-run effects might change if, for to late taxpayers is effective in promoting compliance instance, the reminder effect of the letters fades over with Poland’s PIT laws. In particular, the behavioral time. It is thus recommended that outcomes be closely letter induced higher payment rates and payment monitored in subsequent months and the analysis amounts and decreased tax liability relative to both updated accordingly. (2) The behavioral letter diffe- the control group and recipients of the original dun- red from the original dunning letter in more than one ning letter. The letters also helped to speed up pay- respect: it was simplified, included a social norms ment. The results make a compelling case for rigo- message, and had a milder tone than the original dun- rous testing of small adaptations to communications ning letter, and because it was sent by regular mail to taxpayers and show that current communications it did not require the recipient to sign. Thus, a lesson can be enhanced to promote tax compliance at low learned from this trial is that better understanding cost and without the need for new legislation. None- of which small changes to the behavioral letter con- theless, caution is required in interpreting these tributed the most to increase tax compliance would results for two reasons: (1) The data are limited to require isolating each effect by new randomized con- a period of 12 weeks from the date the letters were trolled trials to test each message specifically. EXPERIMENTAL EVIDENCE FROM POLAND 31 ANNEX 2. MAIN REGRESSION MODEL RESULTS: SIGNIFICANCE IN TERMS OF THE REGULAR DUNNING LETTER Outstanding Outstanding Payment Payment Payment Payment Payment Payment Payment Liability Liability Rate Rate (PLN) (PLN) (PLN) (log) (log) (log) (log) Controls Controls PPLN w/Controls Controls Controls Dunning 0.005 0.006 36.357 38.281 35.781 0.050 0.057 -0.012 -0.022 registered (0.009) (0.009) (88.101) (81.563) (80.478) (0.062) (0.062) (0.072) (0.066) Standard 0.061*** 0.062*** 248.699*** 260.068*** 224.646*** 0.433*** 0.434*** -0.347*** -0.346*** behavioral (0.008) (0.007) (73.030) (67.611) (55.760) (0.052) (0.051) (0.060) (0.055) 0.035*** 0.036*** 177.070** 182.741*** 164.420*** 0.263*** 0.265*** -0.225*** -0.234*** Social norms (0.008) (0.007) (72.867) (67.459) (58.597) (0.051) (0.051) (0.060) (0.055) Positive public 0.046*** 0.045*** 121.721* 137.417** 115.564* 0.301*** 0.299*** -0.269*** -0.256*** good (0.008) (0.007) (72.900) (67.489) (60.685) (0.052) (0.051) (0.060) (0.055) Negative public 0.027*** 0.028*** 167.235** 155.223** 155.893*** 0.209*** 0.211*** -0.150** -0.166*** good (0.008) (0.007) (72.950) (67.537) (60.402) (0.052) (0.051) (0.060) (0.055) 0.045*** 0.046*** 238.938*** 218.164*** 216.627*** 0.331*** 0.326*** -0.286*** -0.291*** Deterrence (0.008) (0.007) (72.888) (67.481) (58.476) (0.052) (0.051) (0.060) (0.055) Deterrence with 0.069*** 0.069*** 219.358*** 238.546*** 200.365*** 0.475*** 0.474*** -0.432*** -0.417*** execution form (0.008) (0.007) (72.989) (67.573) (56.465) (0.052) (0.051) (0.060) (0.055) Omission 0.071*** 0.072*** 195.273*** 204.078*** 180.034*** 0.474*** 0.477*** -0.405*** -0.411*** commission (admin) (0.008) (0.007) (73.018) (67.600) (57.191) (0.052) (0.051) (0.060) (0.055) Omission 0.084*** 0.084*** 346.603*** 370.048*** 302.064*** 0.588*** 0.592*** -0.467*** -0.462*** commission (admin) + deterrence (0.008) (0.007) (73.026) (67.606) (51.819) (0.052) (0.051) (0.060) (0.055) Omission 0.063*** 0.064*** 228.013*** 233.245*** 207.583*** 0.445*** 0.448*** -0.381*** -0.389*** commission (taxpayer) (0.008) (0.007) (73.010) (67.592) (56.389) (0.052) (0.051) (0.060) (0.055) 0.402*** 0.683*** 1122.670*** -1453.107*** 2.535*** 2.555*** 4.905*** 0.559** Constant (0.006) (0.034) (61.724) (310.812) (0.044) (0.234) (0.051) (0.252) N 149925 149925 149925 149925 149925 149925 149925 149925 149925 Notes: Stars indicate a significant difference from the control group at the 0.10, 0.05, and 0.01 levels. 32 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION ANNEX 3. MAIN REGRESSION MODEL RESULTS: SIGNIFICANCE IN TERMS OF THE STANDARD BEHAVIORAL LETTER Outstanding Outstanding Payment Payment Payment Payment Payment Payment Payment Liability Liability Rate Rate (PLN) (PLN) (PLN) (log) (log) (log) (log) PPLN Controls Controls w/Controls Controls Controls Dunning -0.056*** -0.056*** -212.342*** -221.787*** -0.383*** -0.377*** 0.335*** 0.325*** -0.584*** registered (0.008) (0.008) (73.996) (68.504) (0.052) (0.052) (0.061) (0.055) (0.114) -0.061*** -0.062*** -248.699*** -260.068*** -0.433*** -0.434*** 0.347*** 0.346*** -0.336*** Dunning regular (0.008) (0.007) (73.030) (67.611) (0.052) (0.051) (0.060) (0.055) (0.113) -0.026*** -0.026*** -71.629 -77.326 -0.170*** -0.169*** 0.122*** 0.112*** 0.009 Social norms (0.006) (0.006) (54.983) (50.903) (0.039) (0.038) (0.045) (0.041) (0.082) Positive public -0.015*** -0.017*** -126.978** -122.651** -0.132*** -0.135*** 0.078* 0.090** 0.033 good (0.006) (0.006) (55.028) (50.944) (0.039) (0.038) (0.045) (0.041) (0.082) Negative public -0.034*** -0.034*** -81.464 -104.844** -0.224*** -0.223*** 0.196*** 0.180*** -0.037 good (0.006) (0.006) (55.093) (51.005) (0.039) (0.038) (0.045) (0.041) (0.083) -0.016*** -0.016*** -9.761 -41.904 -0.102*** -0.108*** 0.060 0.056 -0.117 Deterrence (0.006) (0.006) (55.012) (50.930) (0.039) (0.038) (0.045) (0.041) (0.082) Deterrence with 0.008 0.007 -29.341 -21.522 0.042 0.040 -0.085* -0.070* -0.061 execution form (0.006) (0.006) (55.145) (51.053) (0.039) (0.039) (0.045) (0.041) (0.081) Omission 0.010* 0.010* -53.426 -55.990 0.041 0.043 -0.058 -0.065 -0.008 commission (0.006) (0.006) (55.184) (51.089) (0.039) (0.039) (0.045) (0.041) (0.081) Omission 0.022*** 0.022*** 97.904* 109.980** 0.155*** 0.158*** -0.120*** -0.116*** -0.127 commission + deterrence (0.006) (0.006) (55.194) (51.098) (0.039) (0.039) (0.045) (0.041) (0.080) Omission 0.002 0.003 -20.686 -26.823 0.012 0.014 -0.034 -0.043 -0.131 commission (taxpayer) (0.006) (0.006) (55.173) (51.078) (0.039) (0.039) (0.045) (0.041) (0.081) 0.463*** 0.745*** 1371.369*** -1193.039*** 2.968*** 2.989*** 4.559*** 0.213 8.822*** Constant (0.004) (0.034) (39.032) (307.542) (0.028) (0.232) (0.032) (0.249) (0.058) N 149925 149925 149925 149925 149925 149925 149925 149925 68002 Notes: Stars indicate a significant difference from the control group at the 0.10, 0.05, and 0.01 levels. EXPERIMENTAL EVIDENCE FROM POLAND 33 ANNEX 4. TREATMENT EFFECTS (VERSUS THE STANDARD DUNNING LETTER) OF DIFFERENT LETTERS BY TAXPAYER AGE 34 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION ANNEX 5. TREATMENT EFFECTS (VERSUS STANDARD DUNNING LETTER) OF DIFFERENT LETTERS BY INITIAL TAX LIABILITY EXPERIMENTAL EVIDENCE FROM POLAND 35 ANNEX 6. DOES THE CUT-OFF DATE MATTER? The results presented in the report are based on the 2. All behavioral letters become less effective, espe- data obtained from the authorities by June 13, 2016. cially in terms of payment amount. Up to that date, the tax administration undertook no other enforcement activities. After that date it is 3. The effect of behavioral letters on days to repay- harder to interpret the findings because there were ment becomes negative (with regards to dunning other interventions, such as enforcement activities regular) – an indication that the effect on the by tax offices. For example, a number of taxpayers composition on the sample of those who pay from all treatment arms who had failed to pay by attenuates and becomes more homogeneous June 13 were sent dunning letter by registered mail, across treatments – and those who were sent the and executive proceedings were initiated for tax- regular letter start repaying – an indication that payers who had been sent registered dunning letters for them slower repayment is slower than for as part of the experiment. These factors create poten- other taxpayers. This confirms that the executive tially confounding effects that might affect the validi- measures partly make up for the fact that non- ty of the estimates. Nonetheless, the results using as -behavioral messages are not as effective. cut-off dates the first (June 13) and the last (August 8) monitoring rounds are compared here (Table A3.1). 4. From the administration point of view, omission commission messages lose most of their effective- The findings: ness edge over the behavioral standard letter; it appears that these messages were effective 1. The registered letter becomes more productive, in showing “seriousness of intent” on part of the especially with regard to the amount paid and the administration but once the intent is carried out number of days for repayment. via executive proceedings, the messages lose bite. Table A3.1. Results at Different Cut-off Dates Payment rate Log Payment Amount Log Liability Payment Delay June 13 August 8 June 13 August 8 June 13 August 8 June 13 August 8 0.005 0.009 0.026 116.204 0.006 0.068* -0.249* -0.715** Dunning regular (0.009) (0.009) (0.031) (110.439) (0.042) (0.035) (0.139) (0.363) 0.061*** 0.038*** 0.193*** 172.743* -0.146*** 0.137*** 0.336*** -1.421*** Baseline behavioral (0.008) (0.007) (0.026) (91.546) (0.035) (0.029) (0.113) (0.300) 0.035*** 0.018** 0.120*** 87.464 -0.088** 0.079*** 0.345*** -0.984*** Social norms (0.008) (0.007) (0.026) (91.341) (0.035) (0.029) (0.114) (0.300) 0.046*** 0.025*** 0.121*** 16.779 -0.107*** 0.073** 0.369*** -1.113*** Public good positive (0.008) (0.007) (0.026) (91.383) (0.035) (0.029) (0.114) (0.300) 0.027*** 0.012 0.100*** 89.185 -0.056 0.058** 0.298*** -0.784*** Public good Negative (0.008) (0.007) (0.026) (91.445) (0.035) (0.029) (0.114) (0.301) 0.045*** 0.022*** 0.149*** 117.128 -0.114*** 0.085*** 0.219* -1.581*** Deterrence (0.008) (0.007) (0.026) (91.368) (0.035) (0.029) (0.114) (0.300) 0.069*** 0.042*** 0.202*** 121.178 -0.188*** 0.138*** 0.275** -1.842*** Deterrence with executive order (0.008) (0.007) (0.026) (91.495) (0.035) (0.029) (0.113) (0.299) 0.071*** 0.039*** 0.193*** 28.626 -0.161*** 0.105*** 0.328*** -2.109*** Omission (0.008) (0.007) (0.026) (91.531) (0.035) (0.029) (0.113) (0.300) 36 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION Payment rate Log Payment Amount Log Liability Payment Delay June 13 August 8 June 13 August 8 June 13 August 8 June 13 August 8 Omission 0.084*** 0.051*** 0.259*** 240.580*** -0.193*** 0.179*** 0.208* -2.377*** + deterrence (0.008) (0.007) (0.026) (91.541) (0.035) (0.029) (0.113) (0.299) Omission 0.063*** 0.037*** 0.192*** 146.909 -0.147*** 0.138*** 0.205* -1.966*** taxpayer view (0.008) (0.007) (0.026) (91.521) (0.035) (0.029) (0.113) (0.300) 0.402*** 0.552*** 4.888*** 1880.912*** 6.258*** 5.343*** 8.486*** 17.007*** Constant (0.006) (0.006) (0.022) (77.374) (0.029) (0.024) (0.098) (0.256) N 149925 149925 149925 149925 149925 149925 68002 87218 Notes: Stars indicate a significant difference from the control group (at the 0.10, 0.05, and 0.01 levels). EXPERIMENTAL EVIDENCE FROM POLAND 37 ANNEX 7. CONTROL LETTERS AND EXPERIMENT LETTER VARIANTS REGISTERED Dunning Letter – ENGLISH translation 38 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION REGULAR MAIL Dunning Letter – ENGLISH translation EXPERIMENTAL EVIDENCE FROM POLAND 39 BEHAVIORAL BASELINE LETTER – ENGLISH translation 40 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION SOCIAL NORM LETTER – ENGLISH translation EXPERIMENTAL EVIDENCE FROM POLAND 41 PUBLIC GOOD POSITIVE LETTER – ENGLISH translation 42 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION PUBLIC GOOD NEGATIVE LETTER – ENGLISH translation EXPERIMENTAL EVIDENCE FROM POLAND 43 DETERRENCE LETTER – ENGLISH translation 44 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION DETERRENCE LETTER + EXECUTION ORDER – ENGLISH translation EXPERIMENTAL EVIDENCE FROM POLAND 45 OMISSION LETTER – ENGLISH translation 46 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION OMISSION + DETERRENCE LETTER – ENGLISH translation EXPERIMENTAL EVIDENCE FROM POLAND 47 OMISSION TAXPAYER’S VIEW – ENGLISH translation 48 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION REGISTERED MAIL DUNNING LETTER – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 49 REGULAR MAIL DUNNING LETTER – Original Polish Version 50 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION BEHAVIORAL BASELINE LETTER – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 51 SOCIAL NORM LETTER – Original Polish Version 52 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION PUBLIC GOOD POSITIVE LETTER – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 53 PUBLIC GOOD NEGATIVE LETTER – Original Polish Version 54 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION DETERRENCE LETTER – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 55 DETERRENCE LETTER + EXECUTION ORDER – Original Polish Version 56 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION OMISSION LETTER – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 57 OMISSION + DETERRENCE LETTER – Original Polish Version 58 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION OMISSION TAXPAYER’S VIEW – Original Polish Version EXPERIMENTAL EVIDENCE FROM POLAND 59 ANNEX 8. SAMPLE EXECUTION ORDER 60 APPLYING BEHAVIORAL INSIGHTS TO IMPROVE TAX COLLECTION EXPERIMENTAL EVIDENCE FROM POLAND 61 NOTES