An Introduction to Multilevel Modeling for Research on the Psychology of Art and Creativity

This article introduces some applications of multilevel modeling for research on art and creativity. Researchers often collect nested, hierarchical data— such as multiple assessments per person—but they typically ignore the nested data structure by averaging across a level of data. Multilevel modeling, also known as hierarchical linear modeling and random coefficient modeling, enables researchers to test old hypotheses more powerfully and to ask new research questions. After describing the logic of multilevel analysis, the article illustrates three practical uses of multilevel modeling: (1) estimating within-person relationships, (2) examining between-person differences in within-person processes, and (3) comparing people’s judgments to a criterion. The breadth, flexibility, and power of multilevel modeling make it a useful analytic tool for the multilevel data that researchers have been collecting all along. Statistical methods can open new doors by enabling new kinds of hypotheses to be developed and tested. This article describes the usefulness of multilevel modeling—sometimes known as hierarchical linear modeling or random coefficient modeling—for empirical research on art and creativity (Hox, 2002; Luke, 2004). Although it sounds exotic, multilevel modeling is a straightforward extension of conventional regression analyses. Because it is more general, multilevel modeling enables researchers to test hypotheses that cannot be tested with conventional regression or ANOVA models.