Self-control and Uninstallment: Evidence on a Large Video-sharing Social Media Platform

How to cite

Hoong, Ruru (Juan Ru); Hickman, Peter; Tang, Yulu. 2022. "Self-control and Uninstallment: Evidence on a Large Video-sharing Social Media Platform." AEA RCT Registry. AEARCTR-0009143. https://doi.org/10.1257/rct.9143-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Ruru (Juan Ru) Hoong* , Peter Hickman , Yulu Tang

* Primary investigator

Registry ID
AEARCTR-0009143
DOI
https://doi.org/10.1257/rct.9143-1.0
First registered
Last updated
Status
In development
Trial dates
2022-07-05 – 2023-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/9143

Abstract

This study will conduct a randomised controlled trial (RCT) with 10 million users on a large video-sharing social media platform in China to answer two sets of questions. First, we document the extent of self-control issues (e.g., overuse of the platform, naivete over time usage and uninstallment, what types of content more "addicted" or "naive" people tend to view or be engaged with) on the platform, providing insight to the state of digital addiction for hundreds of millions of people on one of the worlds largest platforms. Second, we investigate if an intervention opting users into a soft commitment device (alarms that pop up after an exogenously varied time limit) impacts time usage patterns on the platform and propensity to uninstall. Based on the heterogeneous treatment effects we also document whether the firm's incentives over who to target for treatment assignment (to reduce uninstallment) coincides with a social planner's optimal targeted treatment assignment (to reduce self-control issues). Our participants will be randomised into 7 groups: a control group, and 6 treatment groups following a factorial design that (i) varies the time at which alarms pop up on their screen (20/60/120 minutes into the user's day) and (ii) whether or not users are shown a "bedtime alarm" or "usage reminder" pop-up screen that tells them to set an alarm that pops up after X mins of use or set a bedtime reminder that pops up at a certain time of night.

Intervention

Our intervention comes in the form of a pop-up screen that occurs within the app telling participants either that (i) they've used the app for a long time (or (ii) that it's already late in the evening), and they should either (i) set a reminder to take a break from the app every 30 minutes-8 hours (participants can endogenously set this on a slider scale), or (ii) set a bedtime reminder between 9pm to 1am on a slider scale.

The timing of the pop-up depends on the treatment arm the participant is assigned into. The pop ups will occur either 20, 60, or 120 minutes into the subject's daily use.

Experimental design

The study consists of a randomised controlled trial with 7 arms: a control group, and 6 treatment groups following a factorial design that (i) varies the time at which alarms pop up on their screen (20/60/120 minutes into the user's day) and (ii) whether or not users are shown a "bedtime alarm" or "usage reminder" pop-up screen that tells them to set an alarm that pops up after X mins of use or set a bedtime reminder that pops up at a certain time of night. Our target size for the entire study is ±10 million participants, with 2 million in control, and 1 million in each of the treatment groups.

We will sample from two groups of people: 4 million people who have used the platform between 30 mins and 6 hours daily in the last 2 weeks & are between the age of 18-50, and 6 million people who fulfill the above requirements and moreover have uninstalled (and reinstalled) the app in the last 6 months.

To supplement the administrative data we have from the platform, we will conduct two surveys for a subset of 100k participants: a baseline and a endline survey that pops up in the app. In the week following the baseline survey, participants will be randomised into control and treatment, and treatment group participants will face alarms should they exceed their time limit. The endline survey will be scheduled approximately a month after the start of treatment.

Design details

Randomization method
We will pre-randomise using a random number generator which will be implemented by the platform.
Randomization unit
Individual
Clusters
0 clusters.
Observations
10,000,000 users
Arms
2+ million users control, 1+ million users in each treatment.
Primary outcomes
- Average actual daily usage of platform - Average stop time (last seen usage) on platform - Actual - ideal daily usage of platform (proxy for self-control) - Actual - predicted daily usage of platform (proxy for sophistication)
Secondary outcomes
- Uninstallment behaviour - Engagement on the platform (average percentage of each video watched, % of videos skipped, % of videos liked) - Opt-out behaviour / adjustment of time limit - Average "sleep" time (time between last used at night and first used in morning).

Cited by

No citation data available for this registration yet.

Keywords