How do you detect novelty or primary effects in experiments?

Novelty and primary effects can both distort the results of an experiment, making it difficult to assess whether a new feature truly adds value. Novelty effects occur when users show unusually high engagement initially because the feature is new and attention-grabbing. These metrics often spike in the first few days or weeks and then taper off. In contrast, primary effects reflect user resistance to unfamiliar or disruptive changes, resulting in temporarily suppressed metrics that may later recover.

To detect these effects, begin by segmenting users into cohorts based on whether they are new or returning, or by how long they’ve been on the platform. Track engagement over time separately for each group. If new users consistently show strong engagement while long-time users do not, this may indicate novelty. Similarly, if returning users initially react negatively but their engagement improves over time, a primary effect may be at play.

Use cohort analysis to track these groups longitudinally. Extend the observation window beyond the usual test period—this helps assess whether early metrics persist or fade. If engagement decays after the initial spike, it's likely a novelty effect. If it gradually rises, a primary effect might be fading.

To isolate these effects, extend the observation window and monitor whether the metrics stabilize or decline. Break down metrics by user tenure and engagement frequency. If metrics remain high for new users and decline for returning users, the result may be primarily novelty-driven. If the inverse is true, primary effects may be suppressing short-term gains. Longitudinal cohort analysis and multi-period testing can help distinguish between sustained product value and short-lived effects.