From data and complexity to causes: A three-session workshop series

Part of the Future of Work Institute (FOWI) Academy Workshop Series

Session 1 - From Data to Causes Part 1: Cross-Lagged Panel Modeling in Mplus (9.30am - 11.00am)

In the social and health sciences a variety of methods have been developed for causal inference. One method involves using the past to uniquely predict the future in lagged-effects models, particularly the cross-lagged panel model or CLPM (i.e., vector autoregressive modeling). This general approach typically involves observations from T=10 occasions or fewer. The logic of the CLPM will be discussed along with threats to causal inference in the form of stable factors that can be controlled with latent variables in structural equation models (SEM). An example using Gallup World Poll data will be provided in Mplus to illustrate this method, including ways to estimate short-run and long-run effects over time. Extensions will also be described including the estimation of system stability and latent variable interactions to assess dynamic moderation.

Session 2 - From Data to Causes Part 2: Dynamic SEM for Intensive Longitudinal Data in Mplus (11.45am - 1.15pm)

In the social and health sciences researchers are increasingly using research designs that assess individuals at many occasions over time (T > 10), often referred to as ‘intensive’ longitudinal data. These studies typically use ecological  momentary assessment (i.e., experience sampling methods) that allow measuring individuals in their lived environments,  this offering a level of generalizability for study results that is often lacking in experimental lab and other contexts. New methods for modeling these data in a multilevel lagged-effects framework referred to as dynamic SEM (DSEM) will be described. An example using longitudinal data from romantic couples (dyads) to assess day-on-day affective spillover and crossover will be provided, along with model extensions including random slopes and random variances that provide unique insight into between-person and between-couple differences in affective dynamics. Extensions will also be described including experimental interventions, often called microrandomization, that offer the possibility of strong and generalizable causal inferences at multiple levels of analysis simultaneously.

Session 3 - From Complexity to Causes: Empirical Dynamic Modeling for Nonlinear Systems (2pm - 3.30pm)

Experiments and regression models have been the two dominant methodological workhorses in the social and health sciences for over 60 years. However, these 20th century approaches have limitations that make them unsuitable for modeling complex dynamic systems. For example, as people interact with their environments, their emotions and cognitions become causally linked to these environments, such that changes to the environments impact individuals’ cognitions and emotions, which in turn changes how they interact with their environments (and so forth). Such everyday systems involve interactions and feedback loops that can produce seemingly unpredictable behavior that is hard to study experimentally or by using typical regression models such as CLPMs and DSEM. As an alternative, physical sciences such as ecology have recently offered new methods to study complex dynamic systems. Referred to as empirical dynamic modeling (EDM), these methods allow: 1) characterizing the complexity of a system and the degree of nonlinearity that defines its evolution over time; 2) distinguishing correlation from causation to allow causal inference (i.e., a data-driven method for causal discovery); and 3) estimating causal effects as they change over time to better understand when and why causal associations exist. This talk explores EDM with brief examples to illustrate its value, including an assessment of the effect of daily temperature on crime rates, and an assessment of how the relationship between positive and negative emotions (PA and NA) change over time differently for different people. This latter example sheds light on classic debates in the affective sciences regarding the dimensionality of emotions, showing that at times PA and NA may be orthogonal while at other times positive or negatively related in different causal directions, and all of this may occur differently for different people. Some implications of EDM for the social and health sciences will be described.

All slides, data, and Mplus program input and output will be provided at the conclusion of the workshop(s).

About the Speaker

Michael Zyphur is a former ARC Future Fellow and has held academic positions at the University of Melbourne, University of Washington, and the National University of Singapore. Dr Zyphur has taught statistical modeling to over 2,000 researchers and PhD students across the globe, and his workshops in Mplus, R, Stata, and AMOS can be accessed through the Institute for Statistical and Data Science (Instats) here. His services as a consultant and for delivering custom statistics programs can also be accessed through Instats. The new empirical dynamic modeling (EDM) Stata and R packages are now available here.

Venue

In person (78 Murray Street) and online via WebEx

RSVP

Please RSVP by indicating which workshop you would like to attend and delivery preference (in-person or online) to fowi-admin@curtin.edu.au

Health and Safety

Please refer to Curtin’s information about attending campus: https://www.curtin.edu.au/novel-coronavirus/