There is increasing interest in understanding the role of neighborhood-level factors on the health of individuals. Many large-scale epidemiological studies that accurately measure health status of individuals and individual risk factors exist. Sometimes these studies are linked to area-level databases (e.g. census) to assess the association between crude area-level characteristics and health. However, information from such databases may not measure the neighborhood-level constructs of interest. More recently, large-scale epidemiological studies have begun collecting data to measure specific features of neighborhoods using ancillary surveys. The ancillary surveys are composed of a separate, typically larger, set of individuals. The challenge is then to combine information from these two surveys to assess the role of neighborhood-level factors. We propose a method for combining information from the two data sources using a likelihood-based framework. We compare it with currently used ad hoc approaches via a simulation study. The simulation study shows that the proposed approach yields estimates with better sampling properties (less bias and better coverage probabilities) compared with the other approaches. However, there are cases where some ad hoc approaches may provide adequate estimates. We also compare the methods by applying them to the Multi-Ethnic Study of Atherosclerosis and its Neighborhood Ancillary Survey. [PUBLICATION ABSTRACT]