Centering Decisions in Hierarchical Linear Models: Implications for Research in Organizations
讨论了在分层线性模型中如何对第一层自变量进行中心化（原始度量、总均值中心化、组均值中心化），并针对组织科学中四种多层级研究范式（增量、中介、调节、分离）给出了建模建议。
Organizational researchers are increasingly interested in model ing the multilevel nature of organizational data. Although most organi zational researchers have chosen to investigate these models using traditional Ordinary Least Squares approaches, hierarchical linear models (i.e., random coefficient models) recently have been receiving increased attention. One of the key questions in using hierarchical linear models is how a researcher chooses to scale the Level-1 indepen dent variables (e.g., raw metric, grand mean centering, group mean centering), because it directly influences the interpretation of both the level-1 and level-2 parameters. Several scaling options are reviewed and discussed in light of four paradigms of multilevellcross-level research in organizational science: incremental (i.e., group variables add incremental prediction to individual level outcomes over and above individual level predictors), mediational (i.e., the influence of group level variables on individual outcomes are mediated by individual perceptions), moderational (i.e., the relationship between two individ ual level variables is moderated by a group level variable), and sepa rate (i.e., separate within group and between group models). The paper concludes with modeling recommendations for each of these paradigms and discusses the importance of matching the paradigm under which one is operating to the appropriate modeling strategy.