Level of Analysis

A social science study using survey data can be set at the micro level when individuals are analyzed, or it can be set at a higher, more macro level when aggregates of individuals such as households, wards, precincts, firms, neighborhoods, communities, counties, provinces, states, or nations become the unit of analysis. This structural level, spanning the range from most micro to the most macro, at which a social scientific investigation is carried out is called level of analysis. A particular study may also cut across several levels of aggregation. For example, a multi-level study of the educational effectiveness of a certain education program may include pupil-specific, classroom-specific, school-specific, and school-district-specific information and analyze the data at each and all of the levels.

The choice of level of analysis should be driven by researchers' theory and, subsequently, their research questions. There are two large, contrasting issues of concern over why the level of an analysis must be carefully chosen and specified. The first is the famous issue, or the infamous problem, of the ecological fallacy, popularized by William S. Robinson in 1950. Simply stated, the ecological fallacy is an incorrect inference about individual or micro-level effects or relationships drawn by analyzing aggregate or macro-level data. Many theories are set at the individual level. However, it is easy to overlook the possible fallacy and study social relations in the aggregate because data are more widely available at that level.

The second issue is that of emergent property, which may appear when a number of simple entities (or individual actors or agents) operate in an environment, social or otherwise, forming more complex behaviors as a collective. Emergent properties are not reducible to the properties of the individual agents. This idea is attributed to Emile Durkheim in The Rules of the Sociological Method, initially published in French in 1895. The idea of emergent property is a potent and power one, and its influence can be found outside of the social sciences today. For example, researchers of artificial intelligence study the so-called emergent functionality. Put another way, a component has a particular functionality, which is not recognizable as a subfunction of the global functionality. For survey researchers, data collected at the individual level should not be aggregate in order to draw inference for a particular behavior at a higher level, which may be emergent.

Both the ecological fallacy and emergent property are important issues for survey researchers because the (primary) sampling unit of a survey sets a limit for the level of analysis a researcher wants to use. A sampling unit is the elementary unit that is sampled or selected for detailed examination, and valid statistical sampling requires that each sampling unit have a deter-minable nonzero chance of selection and that each be selected randomly. Statistical properties aside, sampling unit gives the level at which detailed information is acquired. For example, the General Social Survey (GSS) in the United States samples English-speaking individuals 18 years or older living in noninstitutional arrangements in the United States. Naturally, the GSS is most appropriate for analysis at the individual level.

...