In everyday life, social information often comes from individuals with limited and complementary knowledge, posing the challenge of integrating fragmented perspectives into coherent judgments. While previous research has largely assumed that informants share a uniform background knowledge, or has not explicitly specified their knowledge structures, this thesis examines how people aggregate judgments from agents who each possess only partial information about a problem. Across three pre-registered experiments, social agents were designed to observe only part of the relevant evidence. Participants, acting as information coordinators, were required to integrating judgments from these social agents under two conditions: one where they had full information about the agents' knowledge, and another where their own knowledge was also partial, requiring additional inference about what others knew. Participants' responses were modeled using two competing accounts: (1) a Normative Bayesian model, when participants engage in mentalizing by inferring the hidden evidence behind agents' responses and use it to inform their decision; (2) heuristic models, which rely only on surface-level cues such as vote counts, diversity of expertises, or diagnosticity, without engaging in second-order reasoning. Results revealed that although a minority (20-30%) of participants engaged in mentalizing, the majority relied on heuristic strategies. Specifically, the Majority Vote heuristic, which ignores differences in knowledge and focuses solely on the distribution of responses, provided the best overall fit. Additionally, participants' individual difference in social reasoning strategies correlated with participants' beliefs about the informativeness of agents with different knowledge: those who valued knowledge diversity were more likely to use mentalizing. These findings suggest that while people are capable of sophisticated mentalizing under the condition of partial knowledge, they predominantly adopt efficient heuristics. This tendency reflects both differences in perceived informativeness of knowledge and the high cognitive cost of mentalizing under partial knowledge contexts.