Reciprocity in Peer Assessments

Peer assessment's reliability can be undermined when participants behave strategically. Using a formal model we show how reciprocity can lead to reviewers inflating their rating of each others' work, which is exacerbated when review takes place sequentially. We conduct a pre-registered online experiment and we find that reviewers engaged in mutual-review relationships inflate their reports more, compared to when reviews are one-sided. For sequential reviews, a non-trivial fraction of first movers maximally over-report. In accordance to our theoretical model, we also find that second movers are responsive to the review they received, but only when reviews are mutual. This reveals the potential for a quid-pro-quo element in mutual reviews. Our results highlight the importance of appropriately structuring peer assessment to take strategic reciprocity motives into account and ensure the system's reliability.