Incentivizing quality in dairy value chains - experimental evidence from Uganda

Last registered on October 08, 2023


Trial Information

General Information

Incentivizing quality in dairy value chains - experimental evidence from Uganda
Initial registration date
October 18, 2022

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
October 25, 2022, 10:41 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated
October 08, 2023, 10:18 AM EDT

Last updated is the most recent time when changes to the trial's registration were published.


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

Request Information

Primary Investigator


Other Primary Investigator(s)

PI Affiliation
PI Affiliation
PI Affiliation

Additional Trial Information

In development
Start date
End date
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Quality of products transacted within value chains, and the preservation of quality throughout the chain, is central to value chain development. In Uganda, we find that there is a clear demand from dairy processors for better quality raw milk and substantial scope for quality improvement at the dairy farmer level, yet a market for quality does not develop, holding back further value chain transformation. In this study, we test two potential reasons why a market for quality does not develop through a field experiment with randomized interventions at different levels of the value chain. At the dairy farmer level, we conjecture that farmers are paying attention to the wrong quality attributes and design a video-based information campaign to point out what the quality parameters are that matter for processors. We also provide them with a small incentive to put what they learned into practice. Midstream, at milk collection centers where milk is bulked and chilled, we install technology that enables for quick and cheap testing of the milk that is brought in. We look at impact of both interventions at both farmer and milk collection center level and consider outcomes such as milk quality, prices received and quantities transacted.
External Link(s)

Registration Citation

Ariong, Richard et al. 2023. "Incentivizing quality in dairy value chains - experimental evidence from Uganda." AEA RCT Registry. October 08.
Experimental Details


To make relevant quality parameters visible at the level of the milk collection centers, we focus on a technology bundle. In close collaboration with DDA, we install milk analyzers at a random sample of milk collection centers. These can be used to test milk samples of individual farmers or traders that supply to the milk collection centers to establish quality of incoming milk, as well as to test samples from the milk tankers when milk is picked up by traders or processors. Milk analyzers show butter fat, solid non-fats, added water, temperature of milk, protein content, and corrected lactometer coefficient. Taking a sample is non-destructive and takes about 30 to 50 seconds depending on the temperature of the milk. The milk analyzers will be delivered with clear Standard Operating Procedure advising MCCs and MCC staff will be trained. We collaborate with the DDA to set up a system to monitor the milk analyzers and its use. In particular, DDA technicians will visit treatment MCCs at set periods. We also set up a hotline that MCCs can contact in case of problems with the milk analyzers. We also make sure that, over the course of the project, equipment is adequately cleaned and calibrated. In addition to the milk analyzers, the MCC level treatment also consists of a digitized system to keep track of milk quantity and quality delivered to the MCC. To do so, we developed a custom Android application that MCCs can use to register farmers that deliver milk. For these farmers, MCC managers can then record milk deliveries, including quantities delivered and price agreed, as well as a range of quality parameters that can be read from milk analyzer, such as butter fat and protein content. The application can also provide MCC managers with simple reports, such as the average butter fat (weighted by quantities supplied) over a different period (today, yesterday, last week, last two weeks and custom data range). Reports by farmer are also possible, such that MCC managers can determine the total sum to be paid to a farmer for milk delivered in the last 14 days. The application, which is pre-installed on a Samsung galaxy tab A7 with sim-card for mobile internet, backs up data in the cloud, but is designed following an off-line first principle as some MCCs may not have coverage. Finally, for the MCC intervention, we also developed a poster to be displayed at MCCs informing farmers that the MCC now has a milk analyzer that can determine milk quality for free. The poster was designed by a local artist.

To provide information to dairy farmers on the parameters and characteristics that processors are looking for and how farmers can produce milk that adheres to these standards, we use a short engaging video that demonstrates the inputs and practices that can be used to increase milk quality. The use of video has been found to increase technology adoption in different settings, although the effectiveness also depends on a range of design attributes (Spielman et al., 2021). The ability to depict role models in videos seems important to increase both aspirations of the person targeted, as well as creating an enabling environment for adoption in that it may challenge world views and stereotypical thinking (Riley, 2019; Lecoutere et al., 2020).

To provide information to dairy farmers on the parameters and characteristics that processors are looking for and how farmers can produce milk that adheres to these standards, we use a short engaging video that demonstrates the inputs and practices that can be used to increase milk quality. To design the video based extension intervention, we first identified the top five practices and inputs that are known to raise butter fat and Solid Non Fats in milk. This was done through consultations of experts. We found the top 5 practices and inputs were: selection of breed and genetic potential, selection of grasses for high-quality forage, best practice in silage and hay making, correct mixing and dosage of feed, and feed supplements like Methionine and Lysine. To make the information intervention more actionable, we also provide farmers with some free inputs (1 kg of Chloris Gayana also known as Rhodes grass). The video will be screened a first time during baseline data collection and a second time just before the distribution of the milk analyzers. We also developed an appealing handout that summarizes some of the main points from the video using cartoons drawn by a local artist.
Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
The five primary outcomes at MCC level are:
1. average milk quality level of milk sold. This will be sampled from the milk tanks, and based on an index of different quality parameters (at least butter fat content and SNF).
2. average prices at which milk was bought from farmers (during last 7 days)
3. volumes collected in last 7 days
4. sold to top 5 processors (Pearl, Amos, Lakeside, GBK, Vital tomosi) (in last 7 days)
5. price at which milk was sold (in last 7 days)

Outcomes of interest at farmer level, measured in the last seven days:
1. Milk quality (butter fat content and SNF)
2. Production investment and management (based on index of five recommended practices to improve milk quality)
3. Volumes sold (liters during last week)
4. Sold to milk collection center during last week? (1=yes)
5. Price received for milk sold (inclusive of any quality premium that may have been obtained) (average during last week, UGX per liter)
Primary Outcomes (explanation)
Primary outcomes

We define five primary outcomes at each level. These five primary outcomes will be combined in a covariance weighted index to assess overall impact at that level following (Anderson, 2008). As dairy is a continuous activity, we need to define a time frame for measurement. We will use the last full week before the interview.

Secondary Outcomes

Secondary Outcomes (end points)
Secondary outcomes at the milk collection center level include:
1. local sales - previous research found that milk collection centers are also important for local milk supply, often doubling as milk shops. Does the intervention crowd out the local market?
2. reason for selling to buyer (in particular if the buyer pays premium for quality, but also payment modalities)
3. Impact pathway: did MCC measure quality of aggregated milk before selling? In particular butter fat and SNF using a lactoscan? What equipment was used?
4. Who decided on the price? buyer made offer and MCC accepted, MCC made offer and buyer accepted, negotiation — use likert scale slider to get an idea of power balance.
5. Did the buyer pay a quality premium? What was it based on? What is the quality premium?
6. Does the MCC pay a quality premium to suppliers? What was it based on? What is the quality premium?
7. Does market for quality lead to additional investment in quality preservation - milk cans, etc
8. Does the development of a market for quality lead to more formalization (eg written contracts) between farmer and MCC? Between MCC and processor?
9. Changes in mid-stream service provision: Does the MCC provide services related to artificial insemination? Transport? Access to acaracides? Training on milk sanitation? Training on feeding practices?
10. Information on lactoscan use (for ITT-TOT analysis).

Secondary outcomes at the farmer level include:
1. Home consumption of dairy products (liters, in what form, and who consumes diary products) - test if the development of a market for quality milk crowds out animal sourced food intake within the family.
2. Reason for selling to buyer (in particular pays premium for quality, payment modalities,...)
3. Test if intervention leads to quality based market segmentation (with less rejection and more instances of lowering of price when farmer supplies substandard milk)
4. Does the buyer pay for higher quality milk.
5. Buyer checks for quality during last transaction (lactoscan, lactometer, alcohol test).
6. Number of dairy animals (improved/local) - does a market for quality lead to technology adoption for intensification? Is this stronger for the subgroup of farmers that receives the training video, where we explicitly mention that genetics also affect quality parameters?
7. Feed and pasture management - a more detailed analysis than the composite primary outcome 2 at the farmer level. This includes changes in grazing system (paddocking, free range, mixed or zero grazing) and use of different dairy feed types (hay, silage, improved forages, commercial feeds like (brewers) bran, salt and mineral blocks, multivitamin). We will differentiate between practices in the rainy season and the dry season.
8. Price of dairy animals (improved/local) - test if the development of a market for quality has an impact on the price of animals.
9. Gendered decision making outcomes - test if the development of a market for milk impacts who within the household makes the decisions to sell to a particular buyer.
10. Does the development of a market for quality lead to more formalization and less relational contracting?
11. Does the intervention also increases milk sanitation (use of milk cans)?
12. Does the intervention leads to changes in bargaining power? farmer made offer and MCC accepted, MCC made offer and farmer accepted, negotiation — use likert scale slider to get an idea of power balance.
13. Gendered labour outcomes (milking, marketing, feeding and herding or cleaning )
14. Does the intervention affect home processing? Does this have gendered effects?
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
The field experiment consists of two cross-randomized interventions that are implemented at different levels. Outcomes may be measures at different levels. We randomly allocate quality testing equipment to a random subset of milk collection centers (MCCs), while another random subset of milk collection centers functions as the control group for this treatment. In the catchment area of each milk collection center, we then take a sample of dairy farmers, stratifying the sample on whether the farmer is an active supplier to the milk collection center or not. In this sample, we then randomly assign half of the farmers to the information treatment (blocking on whether the farmer is an active supplier to the milk collection center or not).
Experimental Design Details
Not available
Randomization Method
Randomization is done in office by a computer
Randomization Unit
for the MCC level treatment, we randomize at the MCC level (and so dairy farmers are exposed to this treatment in clusters); for the farmer level treatment, randomization is at the individual level.
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
125 Milk Collection Centers
Sample size: planned number of observations
2500 farmers (20 farmers in each catchment area of an MCC)
Sample size (or number of clusters) by treatment arms
60 MCCs control and 65 MCCs treated with lactoscan; 1250 farmers treated with information interventions and 1250 farmers as control
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
We assume that the intervention at the level of the milk collection centers leads to an increase in the price of UGX30 per liter. This seems reasonable in light of the fact that processors told that they either pay a 10 percent premium for quality milk, or UGX100 per liter. However, as we assume a pretty narrow distribution of prices, even though this effect is only a 3 percent increase, this is considered a medium to large effect according to Cohen's D. At the level of the farmers, for the intervention at the MCC level, we expect an effect size of UGX40. While this represents a 4.4 percent increase, the larger variance at this level means that according to Cohen's D, this effect is considered small to medium. Finally, at the level of the farmers, the individual level randomization of the information treatment intervention allows us to estimate small effects. For our power simulation, we assumed and effect size of UGX25, which corresponds to a small effect according to Cohen's D. For the interaction, we assume a large effect (UGX50 per liter).

Institutional Review Boards (IRBs)

IRB Name
IRB Approval Date
IRB Approval Number
Analysis Plan

Analysis Plan Documents

pre-analysis plan



Uploaded At: October 08, 2023