Peer Messaging as a Deterrent in Black Markets

The baseline (control) group includes people who continue to see the generic information about popular illegal drugs (e.g., the current version of the app). The current version of the app allows users to click on individual drug checking results (e.g., test result #1 for MDMA), but does not provide any additional or contextual information such as aggregated statistics or user feedback.

We then have four treatment groups as follows:
(1) Statistical Treatment Group: In this treatment group, we provide salient statistical information about a particular drug. Specifically, we aggregate data from drug checking results to create clear and simple statistical facts, that we present at the drug-location-time level. For instance, we provide a histogram of dosage strength at the drug-category level (e.g., MDMA), for those pills that have been tested in a given city over the last 30 days. We then highlight the average dosage for a particular pill type (e.g., the green tesla, containing MDMA) over that same period, showing where the green tesla sits on the distribution of all tested MDMA pills.
(2) The second treatment group is a variant of treatment group (1), but in addition to providing statistical information, we provide a salient illustration of the source of the data that informs the statistics (e.g., official drug checking site X).
(3) Peer Treatment Group: In our third treatment group, we provide information about drugs and experiences related to drugs in the form of narrative and and ratings-based experiences from drug user peers.
(4) The fourth treatment group is a variant of treatment group (3), but in addition to providing narrative and rating-based experiences, we provide a salient illustration of the source of the data that informs the experiences (e.g., Peers).

2025-10-09

2026-01-09

The primary outcomes are obtained from user engagement with the app (where they click) and a survey instrument that is pushed to users after a weekend. The survey asks questions about drug consumption, experienced effects, safer use behavior, and comfort while taking the drug

As a secondary outcome, we measure how people like the additional information we provide by using a feedback button.

Working with the app developer and engineering team, we conduct four redesigns of the KnowDrugs app to reflect our four treatment arms described above. The new versions appear on the landing page of the app and can be fully engaged with. We maintain the current app interface for our control group.

Together with the owner of the app, we randomly assign app users to either the control group or one of the four treatment arms. Groups are assigned as follows: 10% of current app users are assigned to the control group, and the remaining 90% of current app users are equally split into the four treatment groups. Each app user has an ID, and we will use a random number generator to assign the app user ID to one of the four treatment groups or the control group. They will then only be able to view and engage with their assigned version of the app. Any individuals that download and use the app after the start of our experiment will be allocated to either the control or one of the four treatment groups upon joining.

The app automatically collects click data on that user behavior, which we will use as our outcome variables. We also provide an in-app survey, which is constantly available on the landing page of the app and users can fill it in as often as they want. The in-app survey collects other outcome variables related to user experiences.

Throughout the experiment, we will send weekly push notifications to encourage users to engage with the app. Those push notifications will be randomized based on night, so that we can reduce self-selection concerns related to endogenous timing of individual information acquisition.

Not available

Randomization done in office by a computer

We randomize treatment by individual user.

No

60,000 - 100,000 individuals

60,000 for app engagement data 1,320 for survey data

10% of the participants will be placed in the control group. The remaining 90% will be equally randomized into the four treatment arms (1, 1a, 2, 2a),
i.e.,
13,500 for app engagement data all treatments + 6,000 control group
300 for survey data all treatments + 120 control group

All minimum detectable effect sizes based on a 10% significance level and 80% power level. All 5-point Likert scale survey items (expected standard deviation = 1.1): Minimum detectable effect size is: 0.23 points on the 5-point scale All other variables will be coded as dummies. For the power calculation, we assume the maximum standard deviation of 0.5, when the mean of the dummy variable is 0.5. Minimum detectable effect size is: 1.18 percentage points