In many industrialized nations such as the US, political polarization has increased within the last decades and is linked to important behavioral outcomes such as social distancing during the Covid pandemic. Prior research indicates that polarization emerges based on complex social behavior of agents. One stream of literature showed that polarization can emerge in the co-evolution of network structure and decision-making based on homophily and social influence mechanisms, which can erode socially beneficial coordination. Building on this research stream, we develop an agent-based model employing a coordination game in dynamic networks in which agents are reinforced based on the similarity of their choices or the preferences underlying their choices. We replicate the emergence of polarization when agents observe other agents choices and show that the emergence of polarization can be mitigated if agents observe the preferences underlying other agents’ choices but not the choices themselves. Our results are robust to different proportions of learning weights in the reinforcement of preferences and edge weights. Our findings point to the importance of engaging in interactions to observe the motivations and reasons underlying other individuals’ choices to mitigate polarization.