Experimental Design Details
The trial is an individually randomised trial which involves different variations of letters being sent from the Guatemalan tax authority (SAT) to taxpayers that were due to declare their annual income tax for the year 2013 but failed to do so. The trial involves taxpayers that have self-selected onto the tax regime ‘Regime Over Profits from Lucrative Activities’. For the 2013 tax year this involves a tax rate of 31% (28% in 2014) on taxable income (profit), chosen by taxpayers in the sample instead of the simplified regime of a 6% direct tax on gross income. Consequently the trial is likely to involve the sub-sample of taxpayers with the highest revenues. The tax was due to be declared, and paid, by the end of March and end of April 2014, respectively.
According to SAT, this tax has a higher rate of compliance than other Guatemalan taxes. However, compliance remains relatively low, and the incentives to comply are weak for participants, due to their relative affluence and the higher rate. These two factors (affluence and tax-rate) also make the prospect of increasing compliance with this tax particularly attractive, and increase the likelihood that any effects identified in this trial will be economically significant.
The trial involved a sample of n = 43,389 taxpayers, all taxpayers that were due to declare this tax according to SAT minus those that had already declared by May 16th or had already been contacted by SAT about failing to declare this tax.
Each individual is randomly allocated to one of six arms:
T0) Control group – no letter (n = 12397)
T1) Original SAT letter (n=6198)
T2) Simplification + Deterrent (n = 6199)
T3) Simplification + Deterrent + Social Norm (n = 6198)
T4) Simplification + Deterrent + Omission/Commission (n = 6199)
T5) Simplification + National Pride (n = 6198)
These interventions are explained in detail in the ‘Interventions’ section. We note that more participants are allocated to the control group than to any of the treatment arms. This is because of restrictions placed on the amount of letters that could be sent out by the mail office rather than a methodological reason.
We have four outcome measures of interest, all of which will be measured at three intervals; 3, 7, and 11 weeks after the trial commences:
1) Declaration (binary); defined as the taxpayer completing online declaration. Taxpayers declare their income for the year to the tax authorities through the Declaraguate website (http://declaraguate.sat.gob.gt/)
2) Response (binary); defined as declaration (as above) or a call by the taxpayer to the call centre where the individual gives the alert number provided on the letter.
3) Payment (binary); defined as the taxpayer completing payment for the full amount, or setting up a payment plan.
4) Payment amount (continuous); the total amount paid by the taxpayer at the time of data collection.
The outcome variables are all collected regularly as administrative data by the SAT. This has the advantage of not costing additional money to collect, while simultaneously accurately capturing the dependent variable of interest. The data on declaration is collected automatically from the declaraguate website, the data on phone responses is automatically recorded at the SAT call centres, and the payment data is sent to the SAT by the relevant banks involved in processing tax payments. All of the above data is stored centrally by SAT.
Outcome data will be collected 3, 7 and 11 weeks after the trial commences on 29th May 2014. This is for three key reasons: 1) the exact timeline that the letters will be sent over will remain unclear at the start of the trial due to uncertainty between the SAT and the mail office over the quantity of letters feasible to send out per week, 2) the letters were estimated to be delivered over an 8 week period, the three collection periods were designed to take account of the possibilities that all letters are sent out immediately, or if the process takes much longer.
During trial implementation monitoring data will be collected on the number of letters sent out each week. If it is the case that the letters are sent at a different rate than expected then additional collection dates can be added. Our measurement will therefore adopt an efficient pragmatic design.
The sample was taken from the universe of taxpayers due to pay their annual income tax for the year 2013 according to SAT (n=115,999). From this universe we took all taxpayers that had failed to declare by 16th May 2014 (n=44,952). Finally we removed the taxpayers (n=1563) that had already been contacted by SAT about failing to pay this tax, giving us our sample of n = 43,389 taxpayers.
The 1563 taxpayers already contacted, and hence removed from the sample, were chosen by the tax authority based on their calculations on those most likely to respond to letters. Assuming that the tax authority is accurate in its targeting this creates a selection issue as our sample of taxpayers are systematically harder to reach than the general population. This means that our results are likely to provide a conservative estimate of the impact of the trial letters on the universe of taxpayers that initially failed to declare.
There are no restrictions to eligibility within the groups defined above.
Randomisation was conducted at the individual level. Each participant in the sample was assigned a random number between 0 and 1, using the random function in excel. The taxpayers were then ranked according to this number and split into 6 arms; a control arm of n=12397, and 6 treatment arms of n=6198 or 6199. As mentioned above the control arm is double the size of the treatment arms due to logistical constraints rather than for a methodological reason.
Following Bruhn and McKenzie (2009) we chose to use a pure random draw rather than testing for balance and re-randomising in the event of failure of balance on observables. Following Kahan et al (2014) any significant differences in observables between treatment arms will be controlled for in the analysis.
Data is routinely gathered by the tax authority. To protect participant confidentiality, the tax authority will clean this data of the unique taxpayer identification number and other fields that would allow identification of taxpayers, before transferring data securely to BIT.
Data Storage and Transmission
Data handling and management will be governed by BIT’s data management protocol (available on request). Data will be anonymised and stored on secure hard drives accessible only to BIT and World Bank staff working on the project. Data will not be transmitted to third parties, except where this is appropriate under the conditions of appropriate data sharing agreements.