How do self-selection effects in purchase and review compound together?
This explores the two stacked filters behind any star rating — who chooses to buy, and then who chooses to review — and how those filters feed each other rather than just adding up.
This explores the two stacked filters behind any star rating: people only buy products they already expect to like, and only some of those buyers bother to review. The corpus treats this not as one bias but as a chain of them that amplifies itself. The cleanest statement of the base mechanism is that only consumers expecting satisfaction purchase and then review, so an aggregate rating reflects self-selected preferences, not objective quality Do online reviews actually measure product quality or just buyer preferences?. Two filters in series mean the visible ratings already misrepresent the satisfaction distribution of everyone who *could* have bought — before any social effect even enters.
Where it gets interesting is that these filters aren't a one-time distortion that washes out with volume — they compound over time. Decomposing ratings into baseline quality, social influence, and noise shows that prior ratings measurably bend later ones, with effects that hit sales immediately and then keep echoing forward through the next wave of reviews Do online ratings actually reflect independent customer opinions?. So selection at purchase shapes who reviews, those reviews shape who buys next, and the loop tightens. Early reviewers end up with outsized power over the whole trajectory, and ironically, summary statistics meant to help can actually slow down how fast true quality gets discovered.
There's also a presentation layer stacked on top of the selection layer. Even buyers with genuinely positive experiences will publish lower ratings after reading negative reviews — because negative reviewers read as more intelligent — and this shift shows up only in public, not in private ratings Why do online reviewers publish negative ratings despite positive experiences?. That means the *content* of what self-selected reviewers write is itself being pulled by what earlier self-selected reviewers wrote. Selection and social signaling compound in the same direction.
The lateral payoff: the corpus shows this same loop has a structural cousin in recommender systems. Different recommendation types (frequently-bought-together vs. co-viewed) funnel different audience segments — different priors — toward the same product, which changes both who rates it and whether opinions converge or split Do different recommender types shape opinion convergence differently?. And when you remove the averaging effect that aggregation provides — as personalized reward models do — systems learn sycophancy and echo chambers, explicitly described as mirroring recommender-system failures Does personalizing reward models amplify user echo chambers?. The thing you didn't know you wanted to know: the purchase-then-review compounding loop isn't unique to product ratings. It's the same self-reinforcing selection dynamic now baked into the feedback loops that train AI on what people 'prefer.'
Sources 5 notes
Only consumers expecting satisfaction purchase and review, creating two selection filters. Research shows early reviewers shape later perceptions, altruism affects learnability, and summary statistics can actually slow quality discovery. Observed ratings misrepresent the satisfaction distribution of all potential buyers.
Moe and Trusov decomposed ratings into baseline quality, social-dynamics influence, and error, finding that prior ratings meaningfully affect subsequent ones. These effects have both immediate sales impact and long-term compounding effects through future ratings, though high opinion variance can eventually dampen the distortion.
Posters systematically reduce their ratings in public when exposed to negative reviews, even with positive personal experience—because negative reviewers appear more intelligent. Private raters show no such shift, revealing a self-presentational mechanism tied to multiple-audience communication.
Research shows that frequently-bought-together and co-viewed recommendation networks produce different opinion convergence patterns. The mechanism: each recommender type attracts different audience segments with different prior expectations, shaping both who sees products together and how they rate them.
Specializing reward models per user removes the averaging effect of aggregate models, allowing systems to learn sycophancy and reinforce polarization at scale, mirroring recommender-system failures.