Why do people bother writing online ratings at all?

A foundational question in recommendation: why do people write online ratings at all? There's no immediate reward, the audience is anonymous, and writing takes time. Lafky's experimental approach isolates motivations by manipulating cost-to-rate.

The findings invert several assumptions. First, raters care about both buyers and sellers — they are not purely altruistic toward fellow shoppers, nor purely punitive toward bad merchants. The distribution of ratings reflects mixed motivations rather than a single coherent goal.

Second, when rating becomes more attractive (no cost), people rate broadly across the satisfaction spectrum. When rating has a small cost imposed, the distribution of ratings becomes U-shaped — only people with very strong opinions, positive or negative, find it worth the effort. The middle of the distribution (mildly satisfied or mildly dissatisfied users) drops out. This biases the average rating away from true quality, since true quality is in the middle.

Third, providing small discounts to consumers who rate is a possible solution: it compensates for the cost and recovers participation across the satisfaction range. The general lesson for recommender systems consuming ratings: the rating distribution is not a sample from satisfaction; it is a sample from satisfaction-among-people-who-found-it-worth-rating, with an added cost-of-rating filter that makes the distribution non-representative. Small policy choices about how easy rating is and what compensation it carries dramatically affect what ratings the system sees — and thus what recommendations it produces.