The 2016 U.S. presidential primary election, characterized by unexpected results, provides an interesting context to study how citizens are influenced in deciding whether to vote and whom to support. Our aim is to determine which of those changes in voting behavior have the largest impact on the election outcome. We address this question by developing a class of models driven either by the effect of mass media or by social interaction among voters and non-voting members of two parties. The dynamics are modeled using four compartments with a transition matrix describing the evolution of a discrete-time Markov chain. Each model is studied and fit to poll data from the 2012 and 2016 U.S. presidential elections using numerical methods. A comparison across elections indicates that the social influence of each group changes from one election to another, but response to media is similar in both cases.