Economic and Financial Models for Pricing and Setting Sanitation Tariffs for the Benefit of the Urban Poor (Ghana)

In advanced peri-urban communities such as Accra, the market for sanitation services is already relatively well developed, but households continue to face difficulties in finding and hiring sanitation service providers. As a result, households may not complete necessary desludgings as quickly as they would like, creating local health and sanitation concerns as they wait to have the desludgings completed. In addition, households often hire middlemen to find desludging service providers who substantially increase costs for desludging. This project tests the impact of improving linkages between desludging operators and households through creating a call center in which they can interact. We test the impact of access to the call center on prices and service quality.

In many developing markets, a lack of competition can lead to an apparent paradox of high prices and underinvestment by firms. On the one hand, when markets are large and decentralized it is not a surprise that prices are high: Sellers have an advantage whenever they bargain with a customer, because both parties know that if negotiations break down, it will be inconvenient for the customer to find another potential trading partner. At the same time, this also means many inefficient firms will operate since they typically make profits, despite the presence of much better firms: Work is misallocated from the best firms to the worst, which makes it difficult for fit firms to expand and mature. Consequently, the market has a large number of participants, but high prices; many firms operate, but few grow.

The goal of the project in Ghana is to design a low-cost, low-complexity call center matching system that addresses these problems and allows us to accurately measure the costs of the firms and the values of the households. The moving average of a desludger’s prices is computed neighborhood-by-neighborhood, and this number is treated as a “moving-average (MA) bid”. The idea is that the average bids for a neighborhood by a firm are “smoothed” so that more weight is placed on recent prices, and less on old prices. This ensures that even though costs may vary idiosyncratically within a neighborhood, the system imposes cost discipline without allowing desludgers to negotiate with household over unexpected costs. When a household calls in for a desludging, the desludger with the lowest MA bid is called and asked if he can complete the job by the end of the day. If that desludger does not reply or says no, the desludger with the next-highest MA bid is contacted, and so on. The household is then contacted within the next few days, and the price they actually paid is recorded. This system has a number of appealing features. First, it uses “real” data to discipline the prices the firms quote. Second, it allocates work away from inefficient firms and toward efficient firms, who will be willing to accept lower prices in return for more jobs. Third, combined with random bonuses paid to the firms upon the completion of the job, the bidding allows us to estimate the idiosyncratic costs of the firms using a structural model. Since the sample is representative and we have a comprehensive list of the desludgers operating in Accra, we can then estimate the impact of centralizing the market on prices and the allocation of work and investment in the long run.

2014-09-02

2015-10-15

Price of mechanical desludging, quality of desludging service, stability of client-provider matches, desludging service provider profitability, search costs (time)

Quality of desludging service will be measured by a combination of customer satisfaction metrics, such as: cleanliness and completeness of service, promptness of service scheduling, and rate of price renegotiation. Stability of client-provider matches will be measured through reported repeat client-provider matches and likelihood to recommend or use the same provider again.

We begin with a census of desludging operators and owners in Accra to enroll in the call center. At baseline, we will survey 4,500 households. Of these, 3,600 participants will receive the phone number for the call center, and 900 will serve as a control group. An additional sample of approximately 2,000 households in treatment clusters will be given the call center phone number, but will not be surveyed.

The call center will remain in operation for a period of just over one year. During this time, the call center will collect data related to calls matching clients with desludgers. The clients who call will state their location, and will be connected to the lowest cost available desludger for their neighborhood selected through the moving average auction. The call center will not pre-negotiate or purchase desludgings. It will only connect clients and desludgers, who will then negotiate a price for the service. Two days after being assigned a desludger, the call center will call the client to conduct a follow-up survey about the quality of service provided during the desludging, as well as the price. Price data will then be used to allocate work over time.

One year after the baseline survey, we will return to the same households from the baseline sample to conduct an endline survey.

Randomization was done at the cluster level by computer using statistical software, and balanced by neighborhood block.

The treatment is randomized by clusters of household groups. We have identified up to 60 eligible households in each cluster, but not all of these households will be surveyed. In control clusters 12 households will be surveyed at baseline. In treatment clusters 24 households will be surveyed at baseline, and up to 26 additional households will receive the call center phone number without completing a survey.

Yes

225 (150 treatment and 75 control)

4,500 at baseline: 3,600 treatment and 900 control

Treatment (receive call center number): 150 clusters.
Total households: 3,600 surveyed at baseline, plus up to 2,000 receiving only the call center number but not surveyed
Control: 75 clusters
Total households: 900

The minimum detectable standardized effect size for this study depends on the value of intra-cluster correlation (rho). We estimate that the value of rho is between 0.05 and 0.2, which corresponds to an MDES of 0.17 to 0.27 with power of 0.9 and significance level of 0.05.