Does a progressive tax reform affect tax compliance?

The local government implemented a progressive tax reform in the local property tax, with reductions in the tax rate for properties with the lowest values and increases for those with the highest values. We partnered with the municipality to target the universe of around 100.000 taxpayers with letters outlining the reform and whether the recipient's tax rate was reduced, increased or stayed the same because of her property's valuation. The key aspect of our mailing campaign was the salience of the progressive tax reform: while the baseline letter only stated whether the receipient's tax rate had changed, the treatment letter provided details about the progressive nature of the reform.
UPDATE 2024: A second round of the mailing experiment and a new survey were conducted at the end of 2023/start of 2024 for the tax bills for the year 2024. Please see attached document for details.

2023-01-11

2023-01-31

We will analyze the impact of the reform's details on the probability of paying on time in the months following the intervention.

We will include a link to a survey that will only be completed by a subset of our subjects. The treatment letter and the survey will also be sent by email to a subsample of about 15.000 taxpayers with registered email addresses after the mailing campaign. This survey will inquire about respondent's views on taxation, public goods and progressivity.

HETEROGENEITIES

1) We will analyze the impact of the reform's details for the three groups separately (i.e., for those with low property valuations and tax reductions, for those with mid-value valuations and no changes, and for those with high property valuations and tax increases).

There are two reasons to explore this heterogeneity. First, the most interesting group is the "mid-value" one, because it is the only group that does not have "skin in the game". That is, if there is a treatment effect in this group, it will not be explained by any change in the tax they have to pay. On the other hand, the tax reform directly affected the rate that the other two groups have to pay: low property valuations will see a reduction while high-property valuations will see increases.

Another reason to analyze the effect separately is that the sign of the effects could differ by group. In principle, the sign of the effect is ambiguous for every group. That is because, while some individuals may prefer a more progressive tax scheme (and thus be more willing to pay taxes when they are informed about the reform), the opposite could be true for individuals that prefer less progressive tax schemes. This ambiguity is true for the tree groups (low, mid, high property valuations).

However, in addition to this, there are two factors that are relevant specifically for the low and high property valuation groups. Individuals in the high property valuation group (for whom tax rate increased) will have an additional incentive to reduce their willingness to pay, while the opposite is true for individuals in the low property valuation group (for whom tax rate was reduced).

In any case, we will also present the pooled results to maximize power (average effects for (low, mid, high).

2) We will explore the effects by the level of "past compliance". We expect the effect to be maximized among those that were not "never payers" (that is, we expect a smaller effect among individuals that never paid taxes during 2022). This heterogeneity is based on the result of a previous experiment by Cruces et al (2022, available at https://ifs.org.uk/publications/design-two-stage-experiments-application-spillovers-tax-compliance) which found that result (which was also pre-registered) in a very similar setting.

3) The treatment letter and the survey will also be sent by email to a subsample of about 15.000 taxpayers with registered email addresses after the mailing campaign. The impact of the treatment will be gauged on survey outcomes (respondent's views on taxation, public goods and progressivity) but also on payment behavior for the second month after the main mailing treatment.
----------------------------
EXPERIMENTAL DESIGN UPDATE-MID FEBRUARY2023
Given the low take up for the survey from the link included in the paper letter, we will only use survey responses from the about 15,000 taxpayers with registered email addresses (see secondary outcomes above in the original resgitration). We sent a reinforcement of the treatment including the survey link by email to the subgroup of accounts with emails registered for communications with the Municipality regarding tax matters. We sent the message corresponding to the random group to which each account was originally assigned. The treatment message and the survey link were highly salient in the message. We sent a total of 14646 emails on February 23, and repeated the same message on March 10. This timing implies that these messages will only impact on the second payment in the year - the first payment will only be affected by the original paper letter treatment. Since a read receipt will be available for the email sample, we will be able to compute treatment on the treated estimates with instrumental variable for this sample.

Not available

We randomly assigned individuals to baseline or treatment letters within each of the three groups (i.e., those with low property valuations and tax reductions, those with mid-value valuations and no changes, and those with high property valuations and tax increases).

The unit is the individual property tax account.

No

We divided our experiment in three main experimental samples: low, middle and high value properties, according to the tax reform's criteria. Of the total 102,988 total number of residential properties, these were divided into three groups:
Low 37,946 (36.85%)
Middle 43,761 (42.49%)
High 21,281 (20.66%)

The total number of observations 102,988. The total number of observations in the email subsample is 14,646.

About 50% of each of the three groups will be assigned to treatment or control groups:
| Treatment
Group | 0 1 | Total
---------+----------------------+----------
Low | 19,038 18,908 | 37,946
Mid | 22,034 21,727 | 43,761
High | 10,593 10,688 | 21,281
-----------+----------------------+----------
Total | 51,665 51,323 | 102,988
----------------------------------
Email plus letter sample (February, second payment):
The emails were distributed as follows in the experimental groups:
Tax bill falls (low valuation): Control: 2256 Treatment: 2269
Tax bill unchanged (middle valuation): Control: 3080 Treatment: 3045
Tax bill increases (higher valuation): Control: 1704 Treatment: 1791
Commercial properties: 501

We divided our experiment in three main experimental samples: low, middle and high value properties, according to the tax reform's criteria. The outcome of interest is a dummy variable indicating whether the account paid or not in each month, so effects are measured in percentage points. We calculate minimum detectable effects for the full sample and for the sample excluding "never-payers" (accounts that did not pay taxes in 2022), since we found on a pre-registered analysis from a previous experiment in the same municipality that the subsample of never payers is not responsive to this type of interventions. The minimum detectable effects are: Low in full sample (37,946 obs): 0.0144 percentage points Middle in full sample (43,761 obs): 0.0134 percentage points High in full sample (21,281): 0.0156 percentage points Low excluding never-payers (21,348 obs): 0.0169 percentage points Middle excluding never-payers (25,961 obs): 0.0151 percentage points High excluding never-payers (12,884 obs): 0.0212 percentage points In addition, the analysis will consider payments in multiple months, annual payments and measures of support for progressivity from survey data.