Forming Teams with Information Constraints
How to cite
Kapon, Samuel; Diop, Binta Zahra; Weigel, Jonathan; Tourek, Gabriel; Bergeron, Augustin. 2024. "Forming Teams with Information Constraints." AEA RCT Registry. AEARCTR-0014390. https://doi.org/10.1257/rct.14390-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Can a low-capacity government facing information constraints about the quality of its workers raise revenue by optimizing the composition of tax collector teams? In the context of a large property tax collection campaign in DRC, this paper experimentally compares three assignment policies: random matching, positive assortative matching, and choice-based matching. Random matching pairs up tax collectors randomly. Positive assortative matching (PAM) estimates collector types using available data and then pairs high (low) types together. Choice-based matching pairs collectors based on their elicited preferences. We examine the impact of these three assignment policies on tax collection performance and taxpayer attitudes.
Intervention
In partnership with the Provincial Government and tax authority of Kasaï-Central, DGRKAC, we conduct a randomized controlled trial in the city of Kananga, Democratic Republic of Congo, in which we vary the method by which we assign tax collectors into teams of two to collect tax payments. We use three different methods: (1) random assignment, (2) positive assortative matching (PAM) based on estimated types , and (3) assignment based on preferences elicited from tax collectors.
Experimental design
The current project introduces experimental variation in how tax collectors are assigned into teams of two to collect tax payments. The intervention varies the method by which collectors are paired into teams.
In the first treatment arm, collectors are assigned into teams randomly. We assign 30 tax collectors to this arm, and form 15 teams. Tax collector pairs are rotated every 5 to 6 weeks, and collectors will be assigned to 3 different partners over the course of the campaign.
In the second treatment arm (which we call "PAM "), collectors are assigned into teams based on our assessment of their quality. We rank tax collectors (who have not participated in a tax campaign before) according to our estimate of their skill in tax collection using past data to map characteristics into expected tax collection, and some pre-tax collection tasks that the collectors are required to do. With this ranking in hand, we match collectors in the top half of the rankings (which we call "high types") to each other, and collectors in the bottom half of the rankings (which we call "low types") to each other. Previous work by some of the principal investigators of this project has shown that such a matching has the potential to improve over random matching, if the ranking is sufficiently informative about tax collection skill. The number of partners and timing of partner rotation is identical to the random arm.
In the final treatment arm, we assign collectors into teams based on preferences over potential partners that we elicit from each collector. The number of partners and timing of partner rotation is identical to the random arm. The logic of this treatment arm is that governments facing information constraints about worker quality might be able to improve team formation by eliciting information that workers already possess about one another.
Each pair of collectors works in 3-4 neighborhoods. In the random arm, neighborhoods are randomly assigned. In the PAM and choice treatment arms, we assign neighborhoods with high revenue potential to teams of two "high type" collectors, and low revenue potential to teams of two "low type" collectors, to the extent possible. In the Choice arm, we apply the same method of assigning collectors to types as in the PAM arm.
Design details
Cited by
No citation data available for this registration yet.