Granger Causality Testing with Intensive Longitudinal Data

• Published in: Prevention science Vol. 20; no. 3; pp. 442 - 451
• Main Author: Molenaar, Peter C. M.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2019; Springer Nature B.V
• Subjects: Biophysics; Causality; Child and School Psychology; Coherence; Cointegration analysis; Criteria; Data; Derivation; Economic models; Health Psychology; Longitudinal Studies; Medicine; Medicine & Public Health; Models, Theoretical; Networks; Numbers; Prevention; Prospects; Psychopathology; Public Health; Regression analysis; Representation; Time; Variables; Partial directed coherence; Standard VAR; Hybrid VAR; Structural VAR; Granger causality
• ISSN: 1389-4986; 1573-6695; 1573-6695
• DOI: 10.1007/s11121-018-0919-0

The availability of intensive longitudinal data obtained by means of ambulatory assessment opens up new prospects for prevention research in that it allows the derivation of subject-specific dynamic networks of interacting variables by means of vector autoregressive (VAR) modeling. The dynamic networks thus obtained can be subjected to Granger causality testing in order to identify causal relations among the observed time-dependent variables. VARs have two equivalent representations: standard and structural. Results obtained with Granger causality testing depend upon which representation is chosen, yet no criteria exist on which this important choice can be based. A new equivalent representation is introduced called hybrid VARs with which the best representation can be chosen in a data-driven way. Partial directed coherence, a frequency-domain statistic for Granger causality testing, is shown to perform optimally when based on hybrid VARs. An application to real data is provided.