Under standard assumptions, costly voting models predict that the supporters of the underdog –i.e., of the candidate that is expected to lose– are less likely to abstain than the supporters of the expected winner (Palfrey and Rosenthal, 1985; Herrera et al., 2014). While some empirical/experimental studies identify this underdog effect (Levine and Palfrey, 2007), in others bandwagons emerge: the supporters of the expected winner are found to abstain less often than the supporters of the underdog (Agranov et al., 2018). We focus on large elections and follow Kőszegi and Rabin(2006, 2007) by considering that voters experience losses, with respect to their expected equilibrium payoffs, more intensely than gains. When the election is sufficiently close (i.e., when the shares of the supporters of the two alternatives are not too asymmetric), we find that bandwagons emerge in every equilibrium. To our knowledge, this is the first formal study that explains bandwagons in large elections, by incorporating a commonly accepted behavioural model in an otherwise standard context of costly voting.