Best-Practice Recommendations for Estimating Cross-Level Interaction Effects Using Multilevel Modeling

By Herman Aguinis, R. Gottfredson, Steven A. Culpepper. 2013

1,069 citations recorded by the source corpus.

Multilevel modeling allows researchers to understand whether relationships between lower-level variables (e.g., individual job satisfaction and individual performance, firm capabilities and performance) change as a function of higher-order moderator variables (e.g., leadership climate, market-based conditions). We describe how to estimate such cross-level interaction effects and distill the technical literature for a general readership of management researchers, including a description of the multilevel model building process and an illustration of analyses and results with a data set grounded in substantive theory. In addition, we provide 10 specific best-practice recommendations regarding persistent and important challenges that researchers face before and after data collection to improve the accuracy of substantive conclusions involving cross-level interaction effects. Our recommendations provide guidance on how to define the cross-level interaction effect, compute statistical power and make research design decisions, test hypotheses with various types of moderator variables (e.g., continuous, categorical), rescale (i.e., center) predictors, graph the cross-level interaction effect, interpret interactions given the symmetrical nature of such effects, test multiple cross-level interaction hypotheses, test cross-level interactions involving more than two levels of nesting, compute effect-size estimates and interpret the practical importance of a cross-level interaction effect, and report results regarding the multilevel model building process.

Subfields: Organizational Behavior and Human Resource Management, Sociology and Political Science, Strategy and Management

View Paper: https://www.semanticscholar.org/paper/e7c0b9b84b444134f96bf57ce2bd6646b0296e39

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Organizational Behavior and Human Resource Management

In Sociology and Political Science

In Strategy and Management

Citations in the corpus

Enhancing Rigor in Quantitative Entrepreneurship Research (2019, background): “…research, which often focuses on such rare events as venture capital backing or IPOs of young firms (Davidsson, 2016).9 However, not all outliers are equally interesting or influential, indicating that different types of outliers require different types of treatment (Aguinis et al., 2013).”
Necessity or Opportunity? The Effects of State Fragility and Economic Development on Entrepreneurial Efforts (2019, background): “In our case, individuals belong to a determined region (Aguinis et al., 2013; Rabe-Hesketh & Skrondal, 2006).”
From the Editors: Explaining interaction effects within and across levels of analysis (2014, background): “Multilevel modeling can help identify the directionality of the interaction effects in that it is logical that the contextual variable moderates the relationship between lower level variables (Aguinis et al., 2013).”
Surveying the forest: A meta-analysis, moderator investigation, and future-oriented discussion of the antecedents of voluntary employee turnover (2018, background): “Yet, there are problemswith these twomethods, particularly with regard to not appropriately weighing studies by sample size, as well as statistical power concerns (Aguinis, Beaty, Boik, & Pierce, 2005; Aguinis et al., 2013; Steel & Kammeyer-Mueller, 2002).”
CEO Succession Origin and Firm Performance: A Multilevel Study (2017, positive): “Research tradition in multilevel modeling suggests that this approach is appropriate for examining how effects at a given level are shaped by factors nested at different levels (Aguinis et al., 2013; Preacher et al., 2006).”
A multilevel study on servant leadership, job boredom and job crafting (2018, background): “Although group-mean centering individual-level independent variables is often done when examining cross-level interactions, it is not recommended if the primary aim is not to investigate whether deviation from the group mean affects the outcome (Aguinis et al., 2013; Enders and Tofighi, 2007), as wa”