Greener on the Other Side? Inequity and Property Tax Collection

We will conduct a survey experiment in a sample of 9,000 households located along tax sector boundaries in February and March 2019 (first payments of tax bills are due in mid March).

The survey experiment is designed to study the impact of two treatments. Our main treatment will inform property owners of how the tax system works and how their tax liability compares to their neighbors on the other side of the tax sector boundary. The experiment informs property owners that there is inequity in the system and that they are treated better (for those on the low tax side) or worse (for those on the high tax side) than comparable peers. The second treatment highlights key tools the government has available to enforce the payment of the property tax by households.

We will then measure self-reported and actual tax compliance to test for the existence of peer-comparison effects, and separately for advantageous equity and disadvantageous equity effects, in the taxation context. We will also test for impacts on policy preferences.

2019-02-21

2019-03-23

Beliefs about fairness of the property tax.
Willingness to pay the property tax.
Actual payment of the property tax.

Beliefs about fairness are measured using the answers to the questions "Do you consider the rules for calculating the IPTU liabilities for properties in Manaus to be fair?" and "Do you consider the amount of IPTU charged for your property to be fair compared to what other properties in Manaus are charged?"

Willingness to pay the property tax is measured in two ways.

First, using the answers to the questions "Not paying IPTU this year can be justifiable for families with a property and IPTU bill similar to mine."; "It may be justifiable for households not to pay their IPTU bill this year if their IPTU bill is higher than the IPTU bill of families with a similar property and neighborhood"; and "It may be justifiable for families like mine not to pay their IPTU bill this year if they are struggling to pay other bills".

Second, we perform two list experiments (and randomize their order).
The first list is (** denotes the response of interest)
1-I will move to a property with a lower IPTU liability
2-I will not pay my electricity bill on time
3-IPTU revenues should be used to improve the road surfaces in Manaus
4- ** I will not pay my IPTU bill on time this year
5- I have held an informal job.

The second list is
1- I am in favor of lowering the IPTU tax rates
2- I will not pay my water bill on time this year.
3- I have made a purchase without asking for a VAT invoice
4- ** I will not pay my IPTU bill on time this year
5- I will protest formally to the municipality about the IPTU liabilities in Manaus.

Finally actual payment will be analyzed by merging the survey data with the government's administrative data which contains actual payment details.

Opinions about the severity of penalties for noncompliance with the property tax.
Preferences about criteria to use to set property tax liabilities.
Beliefs about overall level of compliance with the IPTU and its responsiveness to tax rates.

Opinions about the severity of penalties are measured as agreement with the statement "I find the penalties for taxpayers who do not pay the IPTU to be very severe"

Preferences about the criteria to use to set property tax liabilities are measured using the responses to the question "If you could decide the IPTU calculation criteria, which ones would be more important? Indicate "A" in the
item considered most important to you; "B" in the second most important item; "C" in the third
item, "D" in the fourth item and "E" in the fifth item. If the respondent does not know how to
respond, mark "Z" in do not know.
______ Taxpayers who receive the same services from the city hall must pay the same amount
______ Contributors who have the same income must pay the same amount
______ Contributors who are neighbors must pay the same amount
______ Taxpayers with buildings of the same size (square meter) must pay the same amount"

Beliefs about the overall level of compliance with the IPTU are measured using the responses to the questions:
In your opinion, in general, how many families will fully pay the IPTU due this year (2019)?
Note: read options and select the one closest to the person’s case
o Less than 2 families for every 10 families
o Between 2 and 4 families for every 10 families
o Between 4 and 6 families for every 10 families
o Between 6 and 8 families for every 10 families
o More than 8 families for every 10 families

and

What if the value of the IPTU was cut in half? (for example, an account of 500 reais would be
250 reais)
NOTE: Read options (if necessary repeat) to select the one closest to the person's case. This
question is the same as the previous one, only with the value of the IPTU cut in half.
o Less than 2 families for every 10 families
o Between 2 and 4 families for every 10 families
o Between 4 and 6 families for every 10 families
o Between 6 and 8 families for every 10 families
o More than 8 families for every 10 families

Sampling
There are approximately 600,000 taxable properties in Manaus. Our government partners shared their cadaster with us, anonymizing each property such that for each property, the finest level at which it could be identified is the lot the property is on. There are 456,000 lots in the data. We identified all lots that are on blocks that are either adjacent to the boundary of a tax sector or adjacent to a block that is adjacent to a tax boundary to form a buffer two blocks deep around the boundaries of the tax sectors.
Within these blocks, we drop non-residential properties, apartment buildings (entering apartment buildings requires the superintendent's written permission, complicating surveying significantly), and blocks where the average tax change for residential properties were they to be located on the other side of their closest tax boundary is either less than 4% or less than 15 Reais. This results in a sample of 39,306 properties on 32,908 lots.

Randomization.
We randomized the lots into 6 treatments.
1- Control survey; Short list first (in list randomization)
2- Control survey; Long list first (in list randomization)
3- Enforcement treatment survey; Short list first (in list randomization)
4- Enforcement treatment survey; Long list first (in list randomization)
5- Fairness treatment survey; Short list first (in list randomization)
6- Fairness treatment survey; Long list first (in list randomization)

We stratified the randomization into 40 strata, combining
1- 2 groups for above and below median delinquency in 2017.
2- 4 groups for quartiles of the size of the tax liability.
3- 5 groups for the size of the tax jump. For the negative jumps (people on the high tax side of the boundary) we split them into the 40% largest negative jumps (group 1), the next 40% (group 2) and the 20% smallest negative tax jumps. Similarly for the positive tax jumps (people on the low tax side of the boundary) we split them into the 40% largest positive jumps (group 3), the next 40% (group 4) and the 20% smallest positive tax jumps. The 20% smallest positive and negative tax jumps combined to create group 5.

We then randomized the lots into the 6 treatments, with probabilities that varied with the strata depending on the tax jump stratum the lot was in. In the strata with small tax jumps (group 5) we randomized 50% of lots into the enforcement treatment and 50% into the control survey. In the strata with the large (positive or negative) tax jumps we randomized 20% of lots into the enforcement treatment, 40% of lots into the fairness treatment, and 40% into the control survey. These proportions were picked to maximize overall power subject to the constraint that no lots in the small tax jump group would receive the fairness treatment.

Randomization done by a computer.

Lot

Yes

The randomization was performed on the full sample of 32,908 lots and this is what will be given to the survey teams. However, due to the difficulty with performing unincentivized household surveys in an urban environment, we expect response rates to be modest, at around 25-30%. and so we will run the survey until we have around 9,000 responses. Given the average number of households on a lot is 1.19 this corresponds to around 7,500 lots.

9,000 household survey responses.

1,575 lots in group 1 (Control survey; Short list first)
1,575 lots in group 2 (Control survey; Long list first)
975 lots in group 3 (Enforcement treatment survey; Short list first)
975 lots in group 4 (Enforcement treatment survey; Long list first)
1,200 lots in group 5 (Fairness treatment survey; Short list first)
1,200 lots in group 6 (Fairness treatment survey; Long list first)




1,167 households in treatment group 1
1,150 households in treatment group 2
1,276 households in treatment group 3
1,284 households in treatment group 4
1,759 households in treatment group 5
1,806 households in treatment group 6

We use the standard expressions for the MDE (see attached document) and assume that with the baseline administrative data we will able to achieve an R^2 of 0.4 in predicting tax compliance (previous compliance is an extremely strong predictor). Doing this, we estimate we have an MDE of under 0.1 standard deviations for both continuous and binary outcomes.