NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
Tax Compliance Among Small Firms in Rwanda
Initial registration date
February 04, 2018
February 05, 2018 12:53 PM EST
J-PAL and Massachusetts Institute of Technology
Other Primary Investigator(s)
Additional Trial Information
How do small firms navigate tax compliance in low enforcement environments? Using the population of administrative tax declarations by Rwandan firms from 2008 to 2016 to evaluate a 2012 reform in Rwanda that introduced an income tax exemption threshold, we find that firms previously below the exemption threshold on average double the amount of taxable income reported to the tax authority and increase the amount of tax paid by 60 percent in the post-reform period. Potential mechanisms behind this response include overestimation of the likelihood of experiencing an audit, the influence of peers on the declaration decision, and changes in the actual likelihood of experiencing enforcement resulting from the choices of local peers. In order to identify which of these mechanisms explain the observed behavior, we will conduct a survey of 1,000 taxpayers to collect detailed information on the tax compliance decisions of small firms and assess their responses to information through an experimental approach. The survey will measure how the characteristics of firms differ across amounts of reported taxable income in order to understand the responses observed in the quasi-experimental analysis. The survey will contain an information experiment, the design of which is detailed in the pre-analysis plan. The experiment addresses the potential mechanisms by assessing how taxpayers respond to information about (1) the likelihood of experiencing an audit or review from the tax authority and (2) the declaration choices of local peers. Outcomes will be assessed through a survey experiment approach and by observing the actual post-treatment tax declarations made by firms in the sample.
(1) Audit information (high vs. low) Information on the likelihood of experiencing an audit will be drawn directly from Rwanda Revenue Authority data on audits and reviews for fiscal year 2016. The information communicated to taxpayers will provide either a high or low number of audits. For the high audit probability group, taxpayers having declared turnover previously within the exemption zone (below 2 million RWF) will be told that the RRA conducted 1,243 audits of taxpayers like them and that this represents a high number of audits and a high audit likelihood. For the low audit probability group, taxpayers having declared turnover previously just to the right of the exemption zone (at 2 million RWF) will be told that the RRA conducted only 37 audits of taxpayers like them — computed by observing audits in fiscal year 2016 for the income bin from which the sample will be drawn, 0 to 4 million RWF — and told that this represents an extremely low audit likelihood for taxpayers like them.
(2) Peer information (above vs. below exemption zone) Information about the behavior of peers will be crossed with the audit treatments only. The high audit treatment will be crossed with information provided about the proportion of taxpayers (30%) like the respondent choosing to locate at precisely 2 million RWF (just to the right of the exemption zone). The low audit treatment will be crossed with information about the proportion of taxpayers like the respondent who locate within the exemption zone (37%). “Taxpayers like the respondent” is defined as havingg declared between 0 and 4 million RWF previously, and the proportions are drawn from RRA data on the population of taxpayers from fiscal year 2016.
(3) Control Taxpayers assigned to the control group will receive a message about the proportion of taxpayers like them who declared their income tax returns on time before the official deadline for fiscal year 2016 (95%). This information could shift when taxpayers declare relative to the other information treatments but should not affect the amount of income declared as it only communicates information about the timing rather than the content of income tax declaration.
Intervention Start Date
Intervention End Date
Primary Outcomes (end points)
(1) Firm-Level Outcomes from Survey
From the survey data, we will collect three main self-reported outcomes post-treatment. Each of these measures will also be collected pre-treatment, elicited through exactly the same framing. The pre-treatment measures will be asked early on in the survey and the post-treatment measures will be asked at the very end, separated from the information experiment by an unrelated survey section. The post-treatment collection will be preceded by a script telling respondents: “Now I am going to ask you some questions again that I have already asked you. Some people like to keep their answer the same and some people like to change their mind”. The change in these outcome measures pre- and post-treatment will be compared within taxpayers in a survey experimental approach: 1. AuditLikelihood_Survey: Self-reported measures of likelihood of the respondent experiencing an audit or review for their income tax declarations for fiscal year 2017; elicited both on a 4 point scale (Not at all likely to Absolutely certain) and as a percentage probability 2. AuditLikelihoodPeer_Survey: Same 4 point scale measure as above but asked about the likelihood of businesses like the respondent's receiving and audit or review
3. TaxableIncome_Survey: Self-reported amount of annual taxable income the respondent intends to declare for fiscal year 2017
(2) Firm-Level Outcomes from Administrative Data
From the administrative data on tax declarations for fiscal year 2017, we will observe the content of the declaration and the tax payment. These outcomes are considered separately as some taxpayers declare but do not pay the tax owed: 1. Declare_Admin: Whether the firm submits an annual income tax declaration to the RRA for fiscal year 2017
2. DateDeclaration_Admin: The date the annual income tax declaration is submitted to the RRA for fiscal year 2017
3. TaxableIncome_Admin: Amount of declared annual taxable income for fiscal year 2017
4. TaxPay_Admin: Whether the firm makes income tax payments for fiscal year 2017
5. DateTaxPay_Admin: The date the annual income tax payment is submitted to the RRA for fiscal year 2017
6. TaxAmount_Admin: Amount of income tax payments made for fiscal year 2017
7. QuarterlyDeclarations_Admin: After the close of fiscal year 2017, quarterly income tax declarations for fiscal year 2018 due at the end of each fiscal quarter will be measured to observe whether any effects on declarations persist over time
8. QuarterlyPrepayments_Admin: After the close of fiscal year 2017, quarterly prepayments for fiscal year 2018 due at the end of each fiscal quarter will be measured to observe whether any effects on declarations persist over time
9. OtherTaxes_Admin: For both fiscal year 2017 and 2018, declaration and payment of other taxes (VAT, PAYE, fees)
Primary Outcomes (explanation)
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
The unit of randomization for the information treatments is the taxpayer. Each of the 1,000 taxpayers will be randomly assigned to a treatment group (depending on the sample group to which they belong) with equal probability. Randomization will occur within the survey instrument: a random number will be generated by the tablet used to conduct the survey and this number will be used to assign the respondent to a treatment arm during survey data collection. We are not able to randomize ex ante due to concerns about differential response rates — within-survey randomization ensures that all taxpayers who reach the treatment section of the survey instrument have equal probability of being assigned to possible arms within their sample group.
We are reasonably confident that spillovers in our context are not a concern. The individual level treatments (titling, information, rate) will only be provided to individual businesses, the locations of which we can observe in the RRA administrative data. It is possible that taxpayers could share information with others, but given that the sampling frame is the entire population of taxpaying firms declaring between greater than 0 but less than or equal to 2 million RWF (approximately 15,000 firms), we expect the probability of interactions between taxpayers within our sample to be extremely low. We can test ex post whether geographic locations with a greater (lesser) diversity of treatment assignment have attenuated (strengthened) responses to information; however, assessing spillovers will not be a focus of this study.
Experimental Design Details
Randomization done within tablet.
Was the treatment clustered?
Sample size: planned number of clusters
Sample size: planned number of observations
Sample size (or number of clusters) by treatment arms
400 firms control, 150 firms audit high only, 150 firms audit high and peers above, 150 firms audit low only, 150 firms audit low and peers below
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Given the concern noted in Section 3.1 regarding the ability to survey 1,000 respondents before the tax declaration deadline, we provide conservative estimates regarding power, estimated using the sampsi command in Stata. With a cell size of 250 observations (pooling both peer and audit treatments to assess whether taxpayers move in any direction), we are powered at the 5 percent level to detect a 0.25 standard deviation effect with a power level of 80 percent. With cells of at least 400 observations, we are powered to detect a 0.2 standard deviation effect. Comparing the minimum sized cell of 150 observations to the control cell of 400 observations, we are able to detect a 0.27 standard deviation effect. While these minimum detectable effect sizes are not small, the change in behavior (increasing declarations above the exemption zone) that we hope to detect embodies an approximate increase of at least 67 percent (on average) for taxpayers below 2 million RWF (if they move to 2 million RWF) and for taxpayers above, if they move below, an effect of similar magnitude if these taxpayers locate where their peers currently locate in the exemption zone. Therefore, we feel confident that this sample size will permit detection of effects on declared taxable income. Likewise, the information provided about audit communicates appreciably high and low levels of audit likelihood, meaning that if the information is credible it would entail updating perceptions to a substantial degree. Therefore, we also feel confident that we will be able to detect changes in the perceived likelihood of audit with this sample size through the survey experiment. For the peer information experiments, we believe these also provide strong information about peers that will be new information to most taxpayers. Their movement to the locations of the peers described in each message would entail the same sized responses as those discussed above, which are within the range of effects we are able to detect.
INSTITUTIONAL REVIEW BOARDS (IRBs)
Harvard University Committee on Human Subjects
IRB Approval Date
IRB Approval Number
Analysis Plan Documents
February 04, 2018
Post Trial Information
Is the intervention completed?
Is data collection complete?