All times in CEST (Berlin-Time)

04:00 pm CEST = 10:00 am EDT (New York) = 07:00 am PDT (Los Angeles) = 01:00 am + 1 day AEDT (Sydney)

Day 1 - March 29th

04.00 - 04.15 pm (CEST)


Session 1:

04.15 - 04.55 pm (CEST)

Ellen Hamaker: The Within-Between Dispute in Cross-Lagged Panel Research

Recently, several papers have appeared that apply causal inference techniques to cross-lagged panel research, explicating the assumptions under which cross-lagged parameters can be given a causal interpretation. An important prerequisite for such formal analyses is that there is a presumed causal structure; this implies researchers have already had to settle on whether there may be unobserved stable between-person differences that confound the causal effect of interest. However, the latter still forms a major point of dispute in panel research. In this talk, I will examine the within-between distinction in panel research from three different perspectives: design, data, and research question. This will show the importance of considering the timescale at which a process operates, and how this relates to the design that is used and the stability patterns that characterize the data. Moreover, it highlights the need to carefully (re)consider our research question, as well as possible alternative designs.

Session 2:

04.55 - 05.35 pm (CEST)

Kosuke Imai: Matching Methods for Causal Inference with Time-Series Cross-Sectional Data

Matching methods improve the validity of causal inference by reducing model dependence and offering intuitive diagnostics. While they have become a part of the standard tool kit across disciplines, matching methods are rarely used when analyzing time-series cross-sectional data. We fill this methodological gap. In the proposed approach, we first match each treated observation with control observations from other units in the same time period that have an identical treatment history up to the pre-specified number of lags. We use standard matching and weighting methods to further refine this matched set so that the treated and matched control observations have similar covariate values. Assessing the quality of matches is done by examining covariate balance. Finally, we estimate both short-term and long-term average treatment effects using the difference-in-differences estimator, accounting for a time trend. We illustrate the proposed methodology through simulation and empirical studies. An open-source software package is available for implementing the proposed matching methods.


Session 3:

5:45 - 6:25 pm (CEST)

Maya Petersen: SMART Designs to Evaluate Dynamic Regimes: The ADAPT-R Trial in Kenya

Maya Petersen will talk about the analysis of Sequential Multiple Assignment Randomized Trials (SMART) to evaluate dynamic regimes. Dynamic regimes are personalized intervention strategies, in which the intervention that an individual receives can depend on characteristics of the individual, including their response to earlier interventions. Dr. Petersen will include results from the recently completed ADAPT-R trial that evaluated sequenced behavioral interventions to optimize retention in HIV care in Kenya. She will continue to discuss briefly how these methods can be applied to observational data, and extensions of this work to adaptively randomized designs and machine learning approaches for detecting and responding to treatment effect heterogeneity.

6:25 - 7:00 pm (CEST)

Panel Discussion

Day 2 - March 30th


04.00 - 04.05 pm (CEST)

Session 4:

04.05 - 04.45 pm (CEST)

Konrad Kording: Observational Causality from Longitudinal Data in Neuroscience

Neuroscience extensively uses longitudinal data to infer causality. Somewhat regrettably, these estimates of causal effects are almost never tested in subsequent RCTs. I will review the methods used in neuroscience. I will then highlight the problems in the approaches including massive confounding and issues of power. I will briefly discuss why in some very rare cases quasiexperimental techniques may be possible while reminding the viewers that most of the time that won't be possible. Time permitting, I will talk a bit about how thresholds can be optimized in the context of regression discontinuities.

Session 5:

04.45 - 05.25 pm (CEST)

Martin Lindquist: Estimating Causal Effects in Neuroimaging Studies

Neuroscientists are increasingly interested in using neuroimaging data to make causal statements about brain function. Often methods for modeling associations among variables are used to make causal inferences, with associations taken to evidence causation. Here we discuss several approaches towards estimating causal effects in studies of human brain function. This includes a multi-subject whole brain causal model that allows for the estimation of task-related effects, and an approach towards high-dimensional mediation analysis that allows for the determination of the brain regions that mediate the relationship between an exposure and an outcome.


Session 6:

05.35 - 06.15 pm (CEST)

Adriene Beltz: Intensive Longitudinal Data Analysis with Group Iterative Multiple Model Estimation (GIMME)

Each person is unique, but this is not reflected in cross-sectional or traditional longitudinal study designs, or in analyses that treat individuals as representative members of a group (e.g., by gender). Intensive longitudinal data (e.g., from daily diary or ecological momentary assessment studies) are necessary to accurately reflect person-specificity, and network approaches hold significant promise for their analysis. A network approach considers behaviors as nodes and relations among them as edges, and it instigates questions about causality via edge temporality and directionality. This presentation will consider those questions with respect to group iterative multiple model estimation (GIMME), a person-specific network analysis approach that generates directed, time-indexed, and even sample-level edges among behavioral nodes uniquely for each person in a sample. GIMME will be conceptually and mathematically described, its implementation briefly discussed, and its potential illustrated through an application to 75-day intensive longitudinal data on gender.

06:15 - 7:00 pm (CEST)

Panel Discussion

Please note that we don't hold the workshop on March 31st.

Day 3 - April 1st


04.00 - 04.05 pm (CEST)

Session 7:

04.05 - 04.45 pm (CEST)

Christian Gische: Graph-Based Causal Inference Using Parametric Structural Equation Models

When conducting research based on panel designs, researchers often aim at answering causal questions. In the behavioral sciences and psychology, applied researchers often use parametric panel data models. In this presentation, we provide a detailed discussion of the underlying assumptions which justify the use of linear additive panel data models for causal inference. We use a graph-based framework to formally define causal estimands based on the do-operator. On the one hand, linear additive models allow to control for unobserved time-invariant confounders (i.e., causal effects might be identified that would not be identified based on the less restrictive assumptions of a general nonparametric model). On the other hand, linear additive models assume the absence of effect heterogeneity and are likely to be misspecified in many research situations. To alleviate this limitation, we discuss the causal assumptions underlying more flexible random coefficient panel data models. These models (i) capture specific types of effect heterogeneity, and (ii) allow to control for time-invariant unobserved confounding.

Session 8:

04.45 - 05.25 pm (CEST)

Julia Rohrer: "Yes, but what's your estimand?" Improving Science Through Increased Research Goal Transparency

In recent years, researchers have pushed for increased transparency about the research process as a means to improve science. In this talk, Julia Rohrer is going to argue that one area of transparency has been neglected: transparency about the goals of statistical analyses. Such transparency is all the more important the more complex and "impressive" statistical models get. Demanding that every quantitative empirical journal article explicitly states the theoretical estimand of interest, as well as the identifying assumptions linking the theoretical estimand to the statistical analysis, could make it much easier for readers to critically evaluate studies, but also push researchers to do better research in the first place.


Session 9:

05.35 - 06.15 pm (CEST)

Markus Eronen: Causal Discovery in Psychology: Problems, Challenges and Ways Forward

Finding causes is a central goal in psychological research. In this talk, I discuss the

problems and challenges in finding psychological causes from the perspective of philosophy

of science, related to (1) the ill-defined nature of most psychological constructs, (2) the

difficulties in manipulating psychological constructs and measuring them in a robust way,

and (3) failures of causal sufficiency (also known as the problem of unmeasured common

causes). I will then expand this discussion to the context of longitudinal data analysis, and

consider how these problems manifest there and what additional problems arise. Although

the conclusions will be rather pessimistic and critical I will end by discussing several

alternative approaches and ways forward.

06:15 - 7:00 pm (CEST)

Panel Discussion