AEA RCT Registry

Fields Changed

Registration

Field	Before	After
Trial Status	on_going	in_development
Trial Start Date	February 16, 2015	March 14, 2016
Trial End Date	June 05, 2015	December 31, 2016
Last Published	April 03, 2015 09:21 PM	March 10, 2016 11:31 PM
Intervention Start Date	February 16, 2015	March 14, 2016
Intervention End Date	June 05, 2015	July 01, 2016
Experimental Design (Public)	Pass-through -- that is, how a reduction in traders' costs is passed-through to a reduction in the price at which they sell to consumers -- can reveal much about the competitiveness of a market. If markets are perfectly competitive, then any reduction in costs faced by all traders should be passed on in full to consumers, as competing traders underbid each other until the price has dropped by the exact reduction in costs. However, if markets are not perfectly competitive, then the reduction in costs will not be perfectly passed through to the price faced by consumers, as traders account for the fact that any reduction in price will affect quantity sold as well. Moreover, the amount by which the pass-through deviates from 100% gives us a quantifiable measure of the level of competition. The first stage of this study executes exactly this experiment -- offering an exogenous shock to costs in the form of a per-bag subsidy offered to traders -- in order to measure pass-through. If possible, this experiment may simultaneously also allow us to estimate the underlying demand function; if we are underpowered to do this, we will also run a separate experiment to estimate demand directly. From this, we will be able to back out parameters summarizing competition and strategic interaction among intermediaries in the market. To do this, we will take the following specific steps outlined in the design registry document. The second stage of this experiment attempts to measure the impact of entry on competition. In this stage, we encourage (randomly selected) traders to enter (randomly selected) markets in which they do not currently work, by providing a subsidy that is conditional on entry. Reduced form impacts on prevailing price and quantity sold will be measured by comparison to control market-days. Structural estimation utilizing results from Stage 1 will enable us to back out the impact of entry on underlying competitiveness. Further details on these steps are available in the design registry document.	Pass-through -- that is, how a reduction in traders' costs is passed-through to a reduction in the price at which they sell to consumers -- can reveal much about the competitiveness of a market. If markets are perfectly competitive, then any reduction in costs faced by all traders should be passed on in full to consumers, as competing traders underbid each other until the price has dropped by the exact reduction in costs. However, if markets are not perfectly competitive, then the reduction in costs will not be perfectly passed through to the price faced by consumers, as traders account for the fact that any reduction in price will affect quantity sold as well. Moreover, the amount by which the pass-through deviates from 100% gives us a quantifiable measure of the level of competition. The first stage of this study executes exactly this experiment -- offering an exogenous shock to costs in the form of a per-bag subsidy offered to traders -- in order to measure pass-through. If possible, this experiment may simultaneously also allow us to estimate the underlying demand function; to insure against being underpowered to do this (and to measure welfare impacts at counterfactual prices), we will also run a separate experiment to estimate demand directly. From this, we will be able to back out parameters summarizing competition and strategic interaction among intermediaries in the market. To do this, we will take the following specific steps outlined in the design registry document. The second stage of this experiment attempts to measure the impact of entry on competition. In this stage, we encourage (randomly selected) traders to enter (randomly selected) markets in which they do not currently work, by providing a subsidy that is conditional on entry. Reduced form impacts on prevailing price and quantity sold will be measured by comparison to control market-days. Structural estimation utilizing results from Stage 1 will enable us to back out the impact of entry on underlying competitiveness. Further details on these steps are available in the design registry document.
Randomization Unit	The unit of randomization in the market-day, with 48 markets, each observed on 8 days (2 market-days in Stage 1 treatment, 2 market-days in Stage 2 treatment, and 4 market-days as controls). There is also a trader-level randomization to select the traders who will be offered the Stage 2 subsidies for entry.	The unit of randomization in the market-block, with a block consisting of four weeks in a row of market days. The experiments will be conducted with a pool of 60 markets. These markets each have weekly market-days, all of which will be observed for the duration of the 12-week study. The study is broken into three “blocks,” during which each market will cycle through S1 treatment, S2 treatment, and control in a randomized order.
Planned Number of Clusters	48 markets	180 market-blocks
Planned Number of Observations	384 market-days	All transactions occuring in 720 market-days
Sample size (or number of clusters) by treatment arms	48 markets, each observed on 8 days (2 market-days in Stage 1 treatment, 2 market-days in Stage 2 treatment, and 4 market-days as controls)	60 markets, each observed on 12 days (4 market-days in Stage 1 treatment, 4 market-days in Stage 2 treatment, and 4 market-days as controls)
Power calculation: Minimum Detectable Effect Size for Main Outcomes		Power calculations using price variation from the pilot suggest that the sample size of 60 markets should be sufficient to detect a change in price equivalent to a pass-through rate of 20% for the 200Ksh subsidy and 10% for the 400Ksh subsidy (note that, for a larger subsidy, a given pass-through rate results in a larger change in absolute price, which is therefore easier to detect statistically).