1: \begin{abstract}

2: Optimizing recommender systems based on user interaction data is mainly seen as a problem of dealing with selection bias, where most existing work assumes that interactions from different users are independent.

3: However, it has been shown that in reality user feedback is often influenced by earlier interactions of other users, e.g.\  via average ratings, number of views or sales per item, etc.

4: This phenomenon is known as the \emph{bandwagon effect}.

5: 

6: In contrast with previous literature, we argue that the bandwagon effect should not be seen as a problem of statistical bias. In fact, we prove that this effect leaves both individual interactions and their sample mean unbiased.

7: Nevertheless, we show that it can make estimators inconsistent, introducing a distinct set of problems for convergence in relevance estimation.

8: Our theoretical analysis investigates the conditions under which the bandwagon effect poses a consistency problem and explores several approaches for mitigating these issues.

9: This work aims to show that the bandwagon effect poses an underinvestigated open problem that is fundamentally distinct from the well-studied selection bias in recommendation.

10: \end{abstract}

11: