Levels of data: the whys and hows of selection and analysis

There are many challenges in selecting the level of a variable for data collection and analysis. Meeting these challenges is labor intensive and requires considerable skill; however, if a performance improvement project is to produce sound results, shortcuts cannot be taken. Studies that purport to show the impact of important variables on care influence the allocation of resources. The choice of a working unit level of analysis may be highly useful in meeting project goals and statistical requirements, but familiarity with multilevel techniques remains essential. If all or some of the variables must be aggregated, information about individual response rate, the extent and treatment of missing data, and tests of the collinearity of the aggregate data need to be made before proceeding with the analysis.