I show the fallacies from omitting endogenous social interactions (peers' average outcome on own outcome) even if the latter do not bear causal interpretations and simply reflect clustering of outcomes. Dependence in outcomes interferes with the calculation of marginal effects from own and peer characteristics and should, therefore, be modeled. To illustrate potential pitfalls, I disentangle endogenous from exogenous peer effects and own effects in classroom ability net of correlated effects and potential endogenous group formation by using two distinct outcomes in the subject of science for 9th-grade German students. First, I conclude that endogenous group formation is not severe in compulsory education. Second, I show that modeling only effects from own and peer characteristics leads to erroneous conclusions about the existence of peer effects from classmates' average age as well as classmates' prior cognitive ability. Third, when interpreting estimated coefficients as marginal effects, I would erroneously think there exist negative peer effects from classmates' prior cognitive ability. Therefore, researchers omitting endogenous effects should be alarmed as implications on peer effects are quite possibly misleading.