Bold Statements
We asked our presenters to make bold statements about points they think are important in the context of causal inference from longitudinal data, but with which others may not necessarily agree.
These statements will be discussed in our panel discussions at the end of each workshop day.
Ellen Hamaker
"To make progress in causal inference with longitudinal data it is crucial we theorize explicitly about time, timing, and timescales."
"A major shortcoming in causal inference with observational data is the lack of alignment of our substantive theory, longitudinal design, and statistical analysis."
Kosuke Imai
"One major outstanding challenge for causal inference with longitudinal data is the issues related to spillover and carryover effects. "
Maya Petersen
"Making reliable inferences about causal questions from most (if not all) non-experimental data arguably requires us to integrate machine learning methods. But Machine Learning merely accessorized after the fact with causal and statistical thinking will also fall short."
"Reliable causal inference requires a roadmap that formally distinguishes between representation of our knowledge and our scientific question using the formal tools of causal inference, translation of our causal question into a statistical quantity under clear assumptions, and estimation of that quantity using machine learning informed by formal statistical theory."
Konrad Kording
"Most observational causal inference in the real world produces essentially useless results."
"Quasiexperimental techniques are underappreciated outside of a narrow rage of users in the social sciences."
"The analysis of randomized experiments could be much richer with machine learning."
Martin Lindquist
"The assumptions required to make accurate causal claims using neuroimaging data are not well understood, and often not valid in practice."
Adriene Beltz
"Individuals are biologically, psychologically, and socially unique. Methods for forming causal inferences must consider that."
"Understanding human behavior in the everyday contexts in which it occurs will likely necessitate a compromise: Causal inferences may have to yield to directed, statistical prediction."
Christian Gische
"Yes, it's costly but worth the investment. Let's take causal models seriously and test them!"
“All models are wrong, but some are useful. Are linear models useful for causal inference?”
Julia Rohrer
"Even the most "descriptive" research question requires clear causal thinking."
"Most research questions involving longitudinal data are causal in nature, even if it is not always immediately obvious."
Markus Eronen
"Most psychological constructs are too ill-defined to be suited for causal inference."
"We have empirically discovered few if any causal relationships between psychological constructs, and are unlikely to do so in the future."
"The main obstacles to causal inference in psychology are conceptual, not methodological, and cannot be solved with methodological or statistical advances."
"The brain/mind is too complex as a system to allow for useful causal inference."