Establishing Causal Order in Longitudinal Studies Combining Binary and Continuous Dependent Variables

• Published in: Organizational research methods Vol. 20; no. 4; pp. 770 - 799
• Main Authors: Kling, Gerhard, Harvey, Charles, Maclean, Mairi
• Format: Journal Article
• Language: English
• Published: Los Angeles, CA SAGE Publications 01.10.2017; SAGE PUBLICATIONS, INC
• Subjects: Bayesian analysis; Causality; Continuity (mathematics); Correlation analysis; Dependent variables; Economic models; Empirical analysis; Longitudinal studies; Qualitative analysis; Regression analysis; Research management; Researchers; Variables; France; vector autoregression; longitudinal research; binary dependent variables; Bayesian statistics; causality
• ISSN: 1094-4281; 1552-7425
• DOI: 10.1177/1094428115618760

Longitudinal studies with a mix of binary outcomes and continuous variables are common in organizational research. Selecting the dependent variable is often difficult due to conflicting theories and contradictory empirical studies. In addition, organizational researchers are confronted with methodological challenges posed by latent variables relating to observed binary outcomes and within-subject correlation. We draw on Dueker’s qualitative vector autoregression (QVAR) and Lunn, Osorio, and Whittaker’s multivariate probit model to develop a solution to these problems in the form of a qualitative short panel vector autoregression (QSP-VAR). The QSP-VAR combines binary and continuous variables into a single vector of dependent variables, making every variable endogenous a priori. The QSP-VAR identifies causal order, reveals within-subject correlation, and accounts for latent variables. Using a Bayesian approach, the QSP-VAR provides reliable inference for short time dimension longitudinal research. This is demonstrated through analysis of the durability of elite corporate agents, social networks, and firm performance in France. We provide our OpenBUGS code to enable implementation of the QSP-VAR by other researchers.