How would you evaluate a family-invitation feature when parents and teenagers share a social network?

Instruction: Identify the causal question, comparison, assignment unit, spillover risk and success measures. Do not assume parent membership itself is randomized.

Context: Fictional research-design case: consenting eligible teenage users and their already verified family links may be invited to try an optional family feature. Participation remains voluntary. The proposed analysis compares teens whose parents joined with teens whose parents did not. Friends can interact across groups. Age, consent and safeguarding requirements have been approved for this research; do not infer relationships from devices.

Updated

Example answer

The proposed comparison does not isolate the effect of a parent joining: families who opt in may differ in prior engagement, relationships and preferences. Randomly sampling the two resulting groups does not randomize that exposure.

I would define the estimand as the effect of offering the feature to eligible participants, then randomize the offer while keeping voluntary uptake and approved privacy controls intact. Analyze by assigned offer (intention to treat), including nonusers. Do not claim this directly estimates the effect of actual parental participation.

Individual assignment may create spillovers when a teen, parent or connected friend changes behavior. Assign verified family units together when that matches the feature, document cross-family interaction and assess whether a larger network-cluster design or a spillover-aware analysis is needed. Cluster assignment reduces the number of independent units, so power and uncertainty must reflect clustering; it does not magically eliminate cross-cluster effects.

Choose a prespecified teen outcome such as retained voluntary use over a fixed window, together with autonomy, privacy complaints and unwanted-contact guardrails. Parent activity is a separate outcome. Check assignment integrity and baseline balance, measure feature exposure and avoid unsupported stereotypes about parents' free time or teenagers' preferences.

Follow-up

Can we interpret a higher opt-in rate as proof of benefit? No. It measures uptake. Establish the effect on prespecified outcomes and investigate adverse effects before expanding.

Related Questions