Tax evasion and demand for progressivity

This project is built on a large-scale survey experiment to be implemented in Spain. We implement a conventional pre-post information treatment, which exposes randomly selected participants to short videos providing information about the magnitude of tax evasion in the personal income tax (IRPF).
We measure the knowledge participants have before the experimental intervention, and ask similar questions after the treatment, to measure to which extent treated participants updated their beliefs.
We leverage participants’ responses to questions on preferences for tax progressivity to measure how information about the unequal distribution of tax evasion across income groups affects the desired tax burden on each group.

2026-06-22

2026-06-26

The key outcome variable is the demanded tax progressivity, defined as $\text{PROG}^* = t_r^* - t_p^*$, where $t_r^*$ is the respondent's preferred IRPF rate for individuals with monthly income above 43,000 euros (the top 0.1\%) and $t_p^*$ is the preferred IRPF rate for individuals with monthly income below that threshold (a group with mean monthly income of approximately 2,000 euros). Both rates are elicited post-treatment in all arms on a 0--100 slider. A pre-treatment measure of general attitudes towards progressivity is also elicited and used as a baseline covariate.

We randomize participants into three groups with equal probability:

• Control (C): receives no information treatment.

• Rich video (T1): receives a short video informing about the average tax compliance
of the top 0.1% of the income distribution in Spain (xr ≈ 77%).

• Non-rich video (T2): receives a short video informing about the average tax compliance
of the remaining 99.9% of the income distribution (xp ≈ 97%).

The two videos are matched in structure, length, narration style, and visual treatment. Both cite the Spanish Tax Agency (Agencia Tributaria) as the source of the figures.

Not available

Randomization is done at the individual level by Netquest, the
survey provider, at the point of survey administration. Each
respondent is assigned with equal probability (1/3) to one of the
three arms (Control, T1, T2). Netquest applies ex-ante quotas on
gender, age, education, and region to ensure that the final sample
is representative of the Spanish adult population. Balance across
arms on pre-treatment covariates will be checked ex-post and
reported in the empirical section.

individual

No

7,650

7,650

2,550