Duration, Retention, and Engagement in Short Science Videos: A Reanalysis of 48 TikTok Posts

Abstract

Short-form video expands the possibilities for science communication, yet its metrics are often conflated: reaching more people, sustaining viewing, and eliciting actions are not equivalent outcomes. This study characterized associations among duration, retention, discovery sources, reach, and engagement through a reproducible reanalysis of public data from 48 geoscience education videos posted on TikTok between October 2021 and February 2022. A single analytical matrix was reconstructed from the dataset’s primary sheet; rates were recalculated using reach as the denominator, and missingness, duplicates, and denominator consistency were audited. Robust descriptive statistics, Spearman correlations with bootstrap confidence intervals, permutation tests, and Benjamini–Hochberg correction were estimated. Parsimonious linear models with HC3 robust errors and sensitivity analyses excluding the top 5% of reach were also conducted. Median duration was 54.96 s, average percentage watched was 41.73%, completion was 18.46%, reach was 2,345 users, and composite engagement was 8.98%. Average retention (rho = 0.328; adjusted p = 0.033) and completion (rho = 0.390; adjusted p = 0.011) were positively associated with engagement. The share of views from the For You feed showed the strongest association with log reach (rho = 0.785; adjusted p < 0.001), whereas its relationship with engagement did not survive multiplicity correction. Duration was inversely associated with retention (rho = −0.473). Results remained directionally stable after removing extreme-reach videos. Discovery, consumption, and response should therefore be treated as distinct performance dimensions. The observational associations do not establish causality or support generalization beyond the studied account.