Uncovering Sensitivity of Tax Evasion: A Combined Vignette and List Experimental Approach

2024-05-08

2024-05-22

The average number of statements reported by respondents across the four experimental groups.

This study employs a combination of traditional list experiments and vignette experiments. This dual approach is designed to not only uncover the extent of untruthful disclosures but also to assess whether tax morale significantly drives sensitivity bias. Participants are randomly assigned to either a control or a treatment group. The control group receives a list of four morally neutral or insensitive behaviours unrelated to tax evasion. The respondents are not asked to indicate which particular behaviour they have carried out but only the sum of behaviours. The treatment group, in contrast, receives a list that – next to the four behaviours presented to the control group – includes a fifth behaviour pertaining to tax evasion. Comparing the average mean level from both groups enables an estimation of engagement in tax evasion. Later in the survey, the sensitive question about undeclared work is asked separately and directly to all the respondents to estimate the degree of sensitivity bias and the assessment of ignorable treatment assignment, i.e., that the assignment of participants to treatment or control groups does not influence their responses to the direct question about undeclared work.

Further, two additional experimental groups will incorporate a vignette element. Prior to presenting the (either with four or five behaviours), these groups will estimate the percentage of their citizens who find undeclared work completely unacceptable. This aspect of the experiment is crucial, as it examines how sensitivity bias fluctuates when the injunctive norm is held by either a minority (Latvia) or a majority (Italy & Denmark), thus offering insights into the interplay between perceived social norms and individual reporting behaviour.

Recent critiques of list experimental methodology focus on their potential to misrepresent true preferences. Specifically, concerns have been raised that longer lists may inadvertently inflate the average mean reported in the lists, a phenomenon called “artificial deflation”. To address this and accurately gauge the level of sensitivity bias, this study incorporates a placebo test involving a list of five distinct, non-sensitive behaviours. This list is designed to serve as a control, ensuring that any observed effects are due to the sensitive nature of tax evasion questions rather than the list length. Same as for the list that seeks to identify the tax evasion level, the placebo list will have a random assignment, with half of the participants receiving a four-item list and the other half a five-item list. As a result, there are 16 possible experimental paths for each individual.

The hypotheses guiding the study are as follows:
• In scenarios where individuals are not primed with the levels of tax morale in their respective countries, the sensitivity bias will be higher in Latvia and Italy relative to Denmark. This is attributed to the higher prevalence of the shadow economy in Latvia and Italy.

• When individuals are primed with information regarding the levels of tax morale, the sensitivity bias is expected to be greater in Denmark and Italy as compared to Latvia. This hypothesis is based on the premise that tax morale is relatively lower in Latvia.

Randomization is done in office of a pollster company by a computer

Individual

No

6,600 respondents

6,600 respondents

The survey, conducted via Computer Assisted Web Interviews, will encompass a sample of at least 2,200 respondents in each of the three countries (Latvia, Italy, and Denmark), culminating in a total sample size of 6,600 respondents. As a result, each of the 16 experimental paths will be experienced by approximately 410 respondents.