Can small incentives like discounts recover representative rating participation?
This explores whether lowering the cost of leaving a rating — via small perks like discounts — could pull in the moderate, silent majority and fix the well-known skew where only the delighted and the furious bother to rate.
This explores whether small incentives like discounts can recover *representative* rating participation — not just more ratings, but ratings from the people who currently stay silent. The corpus doesn't test discounts directly, but it maps the mechanism they'd have to fix, and that mechanism is sharper than 'people are lazy.' Lafky's experiments show that even tiny participation costs produce a U-shaped distribution of raters: only people with strong opinions — strongly satisfied or strongly angry — find it worth the effort, so the average rating drifts away from true quality Why do people bother writing online ratings at all?. The logic cuts both ways and is encouraging for your question: if a *cost* hollows out the middle, a small *incentive* that offsets that cost should, in principle, coax the indifferent middle back in and refill the distribution toward its true shape.
But the same corpus warns that participation is not the only thing distorting ratings, so flattening the U-shape may not be enough. Moe and Trusov decomposed ratings into baseline quality, social-dynamics influence, and error — and found that prior ratings meaningfully shape later ones, with effects that compound over time through future ratings Do online ratings actually reflect independent customer opinions?. That means a rating is partly an echo of the ratings that came before it. An incentive can change *who* rates, but it can't undo the herding already baked into what new raters see and anchor on. Worse, the people a discount attracts aren't a random sample either: who shows up depends on the context that surfaced the product, and different recommender pathways pull in audiences with different prior expectations, which changes how they rate Do different recommender types shape opinion convergence differently?.
There's also a subtler trap: a discount is itself a treatment, not a neutral nudge. The deeper lesson the corpus keeps returning to is that you should model selection bias explicitly rather than assume more data dilutes it. YouTube's ranking work shows that without an explicit mechanism to strip selection effects from training data, systems converge on degenerate equilibria that amplify their own past decisions Why do ranking systems need to model selection bias explicitly?. A discount-driven rating is a selected observation — conditioned on the offer, the timing, maybe a sense of obligation — so it risks swapping one bias (opinion-strength selection) for another (incentive-induced gratitude or reciprocity). Representativeness isn't recovered just because the sample got bigger.
So the honest synthesis: small incentives plausibly attack the *right* failure — the cost barrier that empties out moderate voices — but 'recover representative participation' overstates what they can do alone. They can reshape the participation distribution; they can't strip out the social compounding, the recommender-driven audience sorting, or the new selection effect they themselves introduce. The corpus's consistent move is to *model* these biases (a position tower, a social-dynamics term, an explicit selection correction) rather than hope a behavioral nudge washes them out. The most interesting takeaway is that 'getting more people to rate' and 'getting a representative rating' are genuinely different problems — and the research here treats the second as a modeling problem, not a participation problem.
Sources 4 notes
Lafky's experiments show raters care about both buyers and sellers rather than purely one or the other. Small participation costs create U-shaped distributions where only strong-opinion raters engage, biasing average ratings away from true quality.
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.
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.
YouTube's multi-objective ranker uses MMoE for conflicting objectives and a shallow position tower to remove selection bias from training data. Without both mechanisms, models converge on degenerate equilibria that amplify their own past decisions.