Perceptions of Corruption and Tax Compliance: A Survey Experiment in Honduras

In this project, we will employ various messaging interventions to assess their relative impact on taxpayer perceptions and willingness to pay or evade taxes within the context of the Honduran tax administration (SAR, as per its Spanish acronym). There are four treatment arms:
• The control group, which received basic information about SAR's role in tax collection, together with an image of the locations of the SAR offices in Honduras.
• The "perception" treatment arm, which provides basic information about SAR's role in tax collection and information about SAR's reputation as one of the less corrupt tax authorities in Latin America, backed by a graph using survey data indicating it as the second most honest tax authority institution in the Latin American region.
• The "purge" treatment, which transmits SAR's recognition of corruption as a significant issue and its recent efforts to combat it. The message highlights the personnel renewal done by the institution for the purpose of reducing corruption and includes a simple figure to illustrate the magnitude of the reform.
• The "combined" treatment arm, which combines the two previous information treatments with the "perception" message followed by the "purge" treatment.

2023-10-02

2023-10-09

Perception of Corruption of the SAR, Citizens’ Perception of their Fellow Citizen’s Corruption, Probability that a SAR Agent will Accept a Bribe, SAR's Capability.

The set of primary outcomes covers those related with the beliefs and the perceptions of the respondents on the tax authority and corruption in general. These outcomes include:
1. Perception of Corruption of the SAR: This metric comprises a self-reported outcome in which respondents’ rate, on a scale from 0 (not likely at all) to 10 (very likely), their perception of the likelihood of SAR officials participating in acts of corruption. We will also study the difference between this variable and the (pre-treatment) perceptions of corruption of the public administration in general (explained in the Experiment Design section).
2. Probability that a SAR Agent will Accept a Bribe: This metric comprises respondents' beliefs regarding the probability that a SAR official might request gifts or money in exchange for resolving or preventing tax-related issues for taxpayers (e.g., allowing tax evasion, avoiding penalties, or expediting processes). Respondents rate this likelihood on a scale from 0 (not likely at all) to 10 (very likely).
3. Citizens’ Perception of Their Fellow Citizen’s Corruption: This metric comprises a self-reported outcome in which respondents’ rate, on a scale from 0 (not likely at all) to 10 (very likely), their perception of that their fellow citizens (in their city of residence) might be involved in acts of corruption. We will also study the difference between this variable and the (pre-treatment) perceptions of corruption of Hondurans in general (explained in the Experiment Design section).
4. SAR's Capability: This metric comprises respondent’s assessment of SAR's capability in collecting the amounts owed by individuals with tax obligations on a scale ranging from "Not capable at all” to "Highly capable". This outcome intends to capture potential mechanisms underlying treatment effects on perceptions.

Accept VAT Evasion, Crosswise Measure of Willingness to Evade Taxes, Lie in Dice Game.

Our secondary outcomes will be the ones related self-reported measures of willingness to evade taxes, and other outcomes regarding tax morale and honest behavior in a dice game, which include:
1. Accept VAT Evasion: This measure presents respondents with a scenario in which a store owner offers a 10% discount in exchange for a sale without the VAT included during a shopping transaction. Respondents indicate whether they would accept or reject this discount (accepting the discount implies being willing to evade the corresponding tax).
2. Crosswise Measure of Willingness to Evade Taxes: Self-reported measures of willingness to evade tax can be biased (for instance, because of a reluctance to declare illegal activity, or because of social desirability bias of respondents). We thus employ a Crosswise measure of this willingness to evade taxes. As in a simple list experiment, we ask respondents to report how many of two statements are true in their case. One of these questions is whether the respondent is willing to evade taxes. The second non-sensitive question asks if the respondent’s mother was born on November or December, highlighting that the research team does not have information on this matter and this that the response privacy is guaranteed. The population distribution of the second question’s probability is known, so that following Jann (2012) we can obtain unbiased estimates of the effect of the treatment on the stated willingness to evade taxes.
3. Lie in Dice Game: This binary outcome originates from an implementation of a dice game (see below in the Experimental Design Section) and indicates whether an individual provided non truthful information in their self-report regarding the outcome of a virtual dice roll.

The primary goal is to identify how the different information treatments impact subject’s perceptions of corruption of the tax administration in Honduras, the potentially subsequent willingness to evade taxes and related outcomes. Data will be collected through an online survey. Participants will be selected randomly from the general population of Honduras (with internet access) and will be assigned to different communication treatments (explained in the Intervention section):
1. Control.
2. "Perception" Treatment.
3. "Purge" Treatment.
4. "Purge" + "perception" treatment ("combined" treatment).
The primary outcomes include a set of self-reported outcomes regarding the beliefs and perceptions of the tax authority (presented in the Primary Outcomes Section). The secondary outcomes are related to self-reported measures of willingness to evade taxes, and other outcomes regarding tax morale and honest behavior in a dice game (presented in the Secondary Outcomes Section). Specifically, the dice game involves rolling a virtual die which produces a random number. Participants then report this number, and in return, they receive a certain number of tickets for a specific lottery based on the reported value. The more tickets they accumulate, the better their chances of winning. Participants can lie to get more tickets in the game. The virtual dice will be unbalanced towards 1 to maximize the incentive to untruthfully report a number in this game.
The hypotheses of this experiment are the following:
1. "Perception" vs. control Hypothesis: Changing individuals' beliefs about corruption by emphasizing lower corruption levels ("perception") will positively affect perceptions of the Honduras tax authority compared to a neutral message (control), assuming that individuals believe the information provided in this treatment arm.
2. "Purge" vs. control Hypothesis: Communicating government efforts to combat corruption ("purge") will positively affect perceptions of the Honduras tax authority compared to a neutral message (control). However, as a recent literature highlights, such messages may backfire depending on individuals' initial beliefs about corruption - specifically, the purge treatment may reinforce an initial negative perception of the tax authority (Cheeseman and Peiffer, 2022). We will explore this heterogeneity for this and other treatment arms.
3. "Combined" vs. control Hypothesis: Combining both strategies ("combined") implies changing beliefs about corruption first (with the "perception" treatment) and emphasizing government actions next (with the "purge" treatment), and this sequence could positively affect perceptions of the Honduras tax authority compared to a neutral message (control). Its effect could be higher than the combined effect of treatments separately if it compensates for a potential backfire of the "purge" treatment. The "combined" effect, however, can backfire if the "purge" treatment makes the problem of corruption too salient (see Cheeseman and Peiffer, 2022).
4. "Combined" vs. "purge” Hypothesis: The combined message strategy ("combined") will be more effective than emphasizing government actions alone ("purge") because it could partially offset the potential backfire effects of "purge" alone.
6. "Perception" vs. "purge" Hypothesis: Changing perceptions about corruption ("perception") will be more effective compared to emphasizing government actions ("purge"), assuming that perceptions of corruption of the public administration play a significant role (Cheeseman and Peiffer, 2022) and that individuals believe the information provided in the "perception" treatment.
Our hypothesis implies that there may be heterogeneous effects according to individuals’ initial beliefs about corruption. We will study treatment heterogeneity along this dimension, for instance by studying whether high initial beliefs corruption perceptions trigger a backfire in the "purge" treatment (and, consequently, in the "combined" treatment). The main variable for heterogeneity is the respondents’ pre-treatment perception of government agents’ corruption, which will be used for subgroups analysis and as a control in regressions. At the same time, to observe whether the perceptions of corruption of the city's inhabitants changed, we will include a pre-treatment question that asks about the pre-treatment perceptions of Hondurans in general, which will also to serve as control in this case. As an alternative, we will use an index of trust in institutions from pre-treatment questions on trust in the Church, in Congress, in the Executive Branch and in the police.
We will collect a series of baseline characteristics, such as the age of the respondents, their sex, their ideology, the region in which they live and their education level. The firm in charge of the survey includes checks for attention and identifies outliers in the time spent on the questionnaire.

The participants will first answer a baseline survey and then will be randomly assigned to a treatment arm. We will randomize at the individual level to our four different treatment arms. The survey software will be in charge of randomizing the individuals into the treatment arms.

Randomization is done at the individual level.

No

Our total sample size will be n=2,000.

Our total sample size will be n=2,000.

A quarter of the participants will be assigned to the "perception" treatment (n=500), a quarter of the participants to the "purge" treatment (n=500), a quarter of the participants to the "combined" treatment (n=500), and the last quarter of the participants to the control group (n=500).

We conducted a preliminary pilot study with a sample size of n=147, during which we observed significant effect sizes concerning one of our primary outcomes, which pertains to the Perception of Corruption of the SAR. Specifically, the effect size approached 0.485 when comparing the "perception" group to the control group. This observed effect is substantially larger than our initial expectations, taking into account that we assume a level of statistical test equal to 0.05. Nevertheless, we plan to employ a larger sample size to determine effects across all our primary outcomes comprehensively. Additionally, this larger sample size will provide us with adequate statistical power to detect effects, considering the multitude of tests we intend to conduct.