Toward an understanding of collaborative tax evasion: A natural field experiment with businesses
Last registered on October 11, 2018


Trial Information
General Information
Toward an understanding of collaborative tax evasion: A natural field experiment with businesses
Initial registration date
October 10, 2018
Last updated
October 11, 2018 7:17 PM EDT
Primary Investigator
Other Primary Investigator(s)
PI Affiliation
Additional Trial Information
Start date
End date
Secondary IDs
Although it has been established that sales from businesses to consumers provide fertile ground for collaborative tax evasion, little is known about the genesis of this phenomenon. We conduct a natural field experiment with 2,900 businesses operating on two German online markets. We advertise small-scale jobs and randomly manipulate the contract conditions. We find that 56% of businesses approach consumers with the intention to evade. The fraction is zero in the regulated market and 72% in the unregulated market. It increases when consumers signal their willingness to collude. Consumers' can save 25% of the legal price on average if they agree to evade.
External Link(s)
Registration Citation
Doerr, Annabelle and Sarah Necker. 2018. "Toward an understanding of collaborative tax evasion: A natural field experiment with businesses." AEA RCT Registry. October 11.
Sponsors & Partners

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
Experimental Details
We run a large-scale natural fi eld experiment in which we take the role of consumers and negotiate with businesses about the terms of condition of the service. The experiment relies on job advertisements that were posted in 42 German cities. We searched for painting and flooring contractors to conduct renovation services in private households. Interested businesses randomly received one of seven treatments that specify the contract conditions. The primary outcome of interest is whether the contractors who were initially interested still want to execute the job once they have learned about our conditions.

The aim of our paper is twofold. First, we quantify the fraction of businesses that approach consumers with the intention to evade and we investigate whether the fraction of evaders varies with different market settings and signals sent by consumers. Second, we quantify the evasion rent of consumers who agree to collaborate, i.e., how much they save if they agree to evade instead of requesting an invoice.

The main experiment was implemented in a 2 x 7 design in which we varied the job advertisement and contract conditions. To understand if our results apply to other volumes of the service, we posted two additional advertisements in which we varied the number of rooms (2 x 2).
The experiment is implemented as a between-subject design.

We implement two treatments to quantify the fraction of businesses that approach consumers with the intention to evade taxes and to quantify the evasion rent. We assume that contractors propose prices with the intention to evade or to declare. In the baseline treatment, we confirm the proposed price and specify the time frame in which the job should be executed. Contractors are informed that we received several offers and will decide in the next few days. They are asked to send us an email (and propose a possible day) if they agree with our conditions. We do not specify whether we need an invoice. Since the email contains the same information as the advertisement, all initially interested contractors should accept the conditions. Reasons, why some may not reply, are, e.g., that they got other jobs in the meantime or forget to answer.

In the invoice treatment, we add the sentence ``I need an invoice, I would like to deduct the costs from taxes.'' In Germany, the government introduced in 2006 a tax subsidy for private households demanding home repair, maintenance, or remodeling services as part of an initiative against tax evasion. The subsidy gives consumers an incentive to declare the transaction to tax authorities. They can deduct 20% of the labor costs from their income tax liability if they prove the transaction through an invoice (up to the amount of 1,200 Euros (about 1,400 US Dollars), §35a Abs. 3 EStG). The existence of the tax subsidy allows us to signal that public authorities will learn about the transaction if the contractor agrees to execute the service. There are two types of contractors that will not accept the invoice treatment. First, informal businesses that are not able to issue an invoice will decline. Second, formal businesses that can issue an invoice but have proposed a price with the intention to evade will reject.

We compare the acceptance of the baseline and invoice treatment to identify the fraction of evaders. We calculate the difference in the proposed prices of businesses accepting the two conditions to identify the evasion rent.

To study if businesses are more willing to evade when consumers signal their willingness to collude, we run five treatments. Four contract conditions are implemented to study businesses' behavior when consumers ask for a price discount. Since collaborative tax evasion is assumed to be related to a lower price, asking for a price discount may signal consumers' willingness to collude. In the baseline discount treatments, we modify the text from the baseline treatment by asking for a 10% or 20% discount. We expect that a fraction of the businesses who proposed a price with the intention to declare switch to evasion when they receive a baseline discount treatment. No such reaction is possible among the contractors that receive the modified invoice treatment in which we ask for a price discount and an invoice.

We expect to observe a lower decline of acceptance rates in the baseline discount treatments than of the invoice discount treatments. This will result in a higher fraction of evaders as compared to the condition without discount.

In treatment TQ, we assess contractors' willingness to deviate from their initial declaration intention when consumers send a stronger collusion signal. The treatment text consists of two questions. We ask if the price includes an invoice and how much the price would be if we would pay cash. The word ``cash'' is commonly used as a collusion signal since cash payment ensures that a transaction cannot be tracked.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
In all treatments, we are interested in the proposed price(s) and the reaction to our treatment. The primary variable of interest in the baseline and invoice treatments is whether interested contractors are still willing to do the job after they have learned the contract conditions. Concerning treatment TQ, we are interested in the initial declaration intention and the fraction of those that are willing to deviate from that intention. Also, we record the "cash price" and the "invoice price" to obtain additional evidence on the evasion rent.
Primary Outcomes (explanation)
We use contractors’ original price offers to calculate the discount requested in the discount treatments, to check that offers are balanced across treatments, and to calculate the evasion rent. We calculate the fraction of evaders as follows: We divide the absolute difference by the acceptance rate in the baseline treatment. We calculate the evasion rent as follows: The average proposed price of those who accept the baseline treatment is the average of the proposed prices of declarers and evaders weighted with the share of the group. We assume that the average proposed price of declarers equals the average price proposals of contractors accepting the invoice treatment. Assuming equal shares and average prices of declarers across the baseline and invoice treatment, we solve for the average proposed price of evaders.
Secondary Outcomes
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
The experiment consists of a two-step procedure. In the first step, we advertised one-time painting and floor installation jobs. The advertisements did not specify the contract conditions (in particular, whether or not an invoice is requested). Interested contractors applied by sending a price proposal via email. In this way, we obtained information on the number of businesses initially interested in the job and their proposed prices. In the second step, every contractor who sent a price proposal was informed about the contract conditions. Our main variable of interest is whether the contractors are still willing to do the job after they have learned the contract conditions. In addition, we compare the prices of those who accepted our conditions to obtain an estimate of the fraction of the price that consumers can save if they agree to evade.
Experimental Design Details
Randomization Method
We randomly assign seven contract conditions (treatments) across interested contractors on the city and advertisement level. We randomly select the number of the treatment send to the first applicant of each advertisement and send treatments to subsequent applicants in chronological order.
Randomization Unit
We randomly assign seven contract conditions (treatments) across interested contractors on the city and advertisement level.
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
42 German cities for the painting job, 22 German cities for floor installation job
Sample size: planned number of observations
2,900 businesses
Sample size (or number of clusters) by treatment arms
Approximately 365 per treatment
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
IRB Name
Ethics Committee of the University of Freiburg
IRB Approval Date
IRB Approval Number
Post Trial Information
Study Withdrawal
Is the intervention completed?
Intervention Completion Date
May 15, 2017, 12:00 AM +00:00
Is data collection complete?
Data Collection Completion Date
May 15, 2017, 12:00 AM +00:00
Final Sample Size: Number of Clusters (Unit of Randomization)
Was attrition correlated with treatment status?
Final Sample Size: Total Number of Observations
3,226 applied, but only 2,900 could be treated, as about 9% did not comply with the conditions posted in the advertisement
Final Sample Size (or Number of Clusters) by Treatment Arms
Approximately 365 per treatment
Data Publication
Data Publication
Is public data available?
Program Files
Program Files
Reports and Papers
Preliminary Reports
Relevant Papers