Since problems associated with the statistical methodology of educational research are becoming increasingly important, this paper examines a subset of problems associated with the analysis and interpretation of aggregated data. Two major questions arise: (1) if a researcher knows the level (e.g., individual, teacher/classroom, school, school district) at which inferences are desired, what complications arise from analyzing data at different levels? and (2) are there general guidelines for determining the appropriate units of analysis in a given research context? Five research contexts in which group observations can be used to estimate relationships among measurement of individuals are examined including contexts with missing observations, fallibly measured variables, the economy of analysis, anonymously collected information, and ecological inference. In choosing units of analysis, appropriateness is a function of the questions asked and of the sampling and/or experimental unit. The former is reflected in the conceptualization of the research objective while the latter can indicate the presence of statistical constraints on the level of inference. Examples of issues and problems that arise with each concern are provided. (Author/DE)