Friday, February 14, 2014
In the literature you often find researchers measuring individuals and aggregating their scores to be analyzed at the group level. In organizational research, aggregating individual scores to a group or departmental level opens up countless opportunities of making better sense of the work place compared to traditional single-level research efforts. Practitioners could use this method to provided better analysis to their customers. This practice, in most cases, can be justified. However, improper aggregation can lead to a model being misspecified.
Prior to aggregating data from a lower level to a higher level one needs to determine what type of emergence is this construct or variable emulating. Emergence can best be thought of as a transformational process. The question to ask is, when aggregating a lower-level construct to a higher-level construct, does the characteristics or meaning of the data change? Emergence can be characterized by two qualitative types: composition and compilation (Kozlowski & Klein, 2000). Kozlowski and Klein (2000) describe composition as being isomporhic in which the lower-level phenomenon is essentially unchanged as it is aggregated to a higher-level phenomenon. Alternatively, compilation describes a phenomenon comprised of a “common domain but are distinctively different as they energy across levels” (Kozlowski & Klein, 2000, p. 16).
A simple example that can be utilized to help distinguish the difference between composition and compilation is the simple concept of classroom learning. For example, if you have a classroom of 10 grade-school children and you teach each individual student simple addition for the numbers 0 to 10, you would expect each student would learn how to add numbers from 0 to 10. By accessing the classroom’s average grade on a test of addition (0 - 10), without allowing any students to interact, you would expect to have a general sense of how much each student learned based on the classroom’s average grade. In this case, the individual learning reflects the classroom’s learning. This example reflects the emergence concept of composition, since the individual learning best represents the classroom’s average grade.
Alternatively, if you take the same 10 grade-school children and you only teach one student addition for the numbers of 0, then you teach the second student addition for the numbers of 1, and so forth. Then, allow the students to interact and share their experiences with what they have learned, then test the classroom, would you get a similar grade? The average grade would reflect the individual student’s learning plus the learning from others through interaction with their classmates. Individual learning, in this case, does not reflect the classroom’s learning. The mediating factor in this case, or the catalyst, is the student’s interactions in which they were allowed to share their learning and experiences with one another. This example reflects compilation, where the individual-level (individual student learning) is similar but distinctively different from the higher level (classroom learning). This example is not to compare the effectiveness of the first example to the second example, it is only to compare the differences between composition and compilation.
Composition and compilation needs to be considered during the initial design of a research project, prior to collecting data. If your level of measurement is at a lower-level (e.g., individual level) and your level of analysis is at a higher-level (e.g., team level) than you need to utilize measures that meet a compositional emergence criteria. If you are utilizing compilation constructs then your level of measurement needs to be at the same level of analysis (e.g., team level and team level). Compilation constructs change meaning when they are aggregated which leads to a model that is misspecified.
Aggregation can be useful for both single-model and multi-model research. Careful planning of each construct and the level of measurement as well as the level of analysis needs to be considered. Klein and Kozlowski (2000) described the importance of a-priori planning: “Rigorous multilevel research rests… on the careful definition, justification, and explication of the level of each focal construct in the model” (p. 214). I would add that this applies for single-level research as well, especially when aggregated constructs are being used.
Klein, K. J., & Kozlowski, S. W. J. (2000). From micro to mess: Critical steps in conceptualizing and conducting multilevel research. Organizational Research Methods, 3, 211-236. doi:10.1177/109442810033001
Kozlowski, S. W. J., & Klein, K. J. (2000). A multilevel approach to theory and research in organizations: Contextual, temporal, and emergent processes. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations: Foundations, extensions, and new directions (pp. 3-90). San Francisco, CA: Jossey-Bass.