This dissertation uses the Ecological Inference (EI) method to produce subgroup estimates of health outcome proportions for 22 US counties. Unlike individual level models based on survey data, the Ecological Inference method begins with fractions among aggregates and estimates proportions for the subgroups of interest. This advance in aggregate data methods allows for lower level inference in the absence of micro-level information. However, the EI method has not been extensively tested outside of research on voting behaviors. This project unpacks the critical pieces of the EI method and tests the assumptions of the model using demographic indicators and fractions of health-related outcomes. Each level of analysis and interactions that take place across levels are considered using quantitative methods. The results from these models are then compared to the subgroup proportions estimated using EI. Finally, EI is used in an attempt to replicate results from a multilevel model for smoking with measurable contextual influence. Data for this project are taken from the 2000 US Census and 2001 Behavioral Risk Factor Surveillance Survey (BRFSS). Data from the BRFSS are aggregated to the county level for use as outcome measures with EI. A consideration of the consequences associated with the aggregation of survey data suggests that the use of BRFSS data at the county level does not ensure adequate representation of all relevant groups. The results of this project suggest that the EI method assumes a compositional difference for subgroup proportions. Further, the accuracy of EI estimated proportions is strongly dependent on the level of contextual influence present in the grouped data, the strength of the relationship between grouped measures, and the level of aggregation. The solution to the ecological inference problem is identified as a solution for aggregation bias present in the explanatory variable only when using the basic model. Data structure, relationships between measures, and the level of aggregation are all identified as having an impact on EI's ability to make accurate estimates of subgroup proportions. Contextual effects influencing variation in the outcome measure is unaccounted for in the estimation procedure and is lost to aggregation bias.