Preventing fraud via mobile phone interactive narrative games in Uganda

This randomized trial will evaluate the impact of a scalable, interactive narrative edutainment intervention on how consumers in Uganda recognize and respond to fraud attempts. Users will have already registered for the Viamo 3-2-1 Interactive Voice Recording (IVR) platform. Users are invited on the platform to participate in the study. Those who provided consent completed a short baseline survey. These users will be randomized into control and treatment groups. The treatment will have the opportunity to play the interactive narrative or Wanji game, with the chance to earn incentives as they complete the game. The content is designed to simulate the experience of receiving a call and decide how to recognize fraud and what to do during and after. To better communicate fraud prevention skills, the heuristics are communicated via a guiding metaphor: Preventing fraud is like crossing the street. Stop, check to verify it is safe, and then act. According to Conceptual Metaphor Theory (CMT), metaphors can enhance understanding by mapping knowledge onto an existing easy-to-understand cognitive frame.

2023-11-10

2024-04-10

a. Fraud prevention capability index
b. Fraud victimization

a. Fraud prevention capability index: Constructed as a z-score average of up to 10 knowledge, attitude, intention and behavioral variables. Knowledge includes knowledge of how to recognize a fraudster, how to respond, and where to report. Attitudes include self-efficacy in fraud prevention domain. Behavioral intention would capture the likelihood of reporting fraud attempts and whether a PIN is shared.

b. Fraud victimization: Self-reported binary measure of whether (a) provided personal information or (b) lost money to fraudsters in a previously specified time period.

a. Complaints registered
b. SIM verification check

a. Complaints registered: Whether the individual registered a complaint with the mobile network operator for (a) fraud or (b) non-fraud issue during the study time period.

b. SIM verification check: Whether the individual verified their SIM card with the regulator's service during the study time period.

To estimate the impact of the interactive narrative or Wanji game, we will conduct an individual-level block randomization. Randomization strata will be created based on language (Luganda, 4R, English), sex (male, female), and age (18-24, 25-34, 35-44, over 44) resulting in 24 blocks. Individuals will then be randomized within each strata into equal-sized control and treatment groups. The Wanji treatment will then be exposed to the treatment until take-up exceeds at least more than 50% or for five months.

Several strategies will be used to ensure intervention take-up: (i) treatment callers have the option of listening to the Wanji game or accessing other platform content in the opening menu, (ii) users will receive an incentive for completing each of the game's two levels, and (iii) weekly SMS encouragements will be sent to individuals. The minimum take-up target is 50%.

Roughly one month after the Wanji treatment ends, IVR reinforcement messages will be sent. To estimate the impact of the IVR reinforcement messages, we will conduct a second-level randomization of treatment group members. The same block randomization will be used as that for the first level to assign individuals into equal-sized control and IVR reinforcement treatment groups.

Not available

Randomization was conducted using the randomize command in STATA 17.

Individual users

No

Clusters were not used.

22,495 platform users

11,247 individual users control
11,248 individual users treatment

Sample and power calculations were estimated for outcomes with binary and continuous variables. Various parameters and sensitivity tests were considered for take-up, attrition, baseline and number of follow up surveys, and multiple hypothesis correction. In the most conservative estimate, to detect a MDE of 0.20 SD, assuming 40% take up, an attrition of 50% between the baseline and first follow up, an attrition of 60% between the first and second follow ups, and adjustment for multiple hypothesis testing for up to three outcomes, the sample size needed at baseline is 19,563. Power increases and/or sample size requirements decrease for a MDE of 0.20 SD with the relaxation of any of these parameters.