Search

Recent developments in the causal-inference literature have renewed psychologists’ interest in how to improve causal conclusions based on observational data. A lot of the recent writing has focused on concerns of causal identification (under which conditions is it, in principle, possible to recover causal effects?); in this primer, we turn to causal estimation (how do researchers actually turn the data into an effect estimate?) and modern approaches to it that are commonly used in epidemiology. First, we explain how causal estimands can be defined rigorously with the help of the potential-outcomes framework, and we highlight four crucial assumptions necessary for causal inference to succeed (exchangeability, positivity, consistency, and noninterference). Next, we present three types of approaches to causal estimation and compare their strengths and weaknesses: propensity-score methods (in which the independent variable is modeled as a function of controls), g-computation methods (in which the dependent variable is modeled as a function of both controls and the independent variable), and doubly robust estimators (which combine models for both independent and dependent variables). A companion R Notebook is available at github.com/ArthurChatton/CausalCookbook. We hope that this nontechnical introduction not only helps psychologists and other social scientists expand their causal toolbox but also facilitates communication across disciplinary boundaries when it comes to causal inference, a research goal common to all fields of research.

Knowledge about causal effects is essential for building useful theories and designing effective interventions. The preferred design for learning about causal effects is randomized experiments (i.e., studies in which the researchers randomly assign units to treatment and control conditions). However, randomized experiments are often unethical or unfeasible. On the other hand, observational studies are usually feasible but lack the random assignment that renders randomized experiments causally informative. Natural experiments can sometimes offer unique opportunities for dealing with this dilemma, allowing causal inference on the basis of events that are not controlled by researchers but that nevertheless establish random or as-if random assignment to treatment and control conditions. Yet psychological researchers have rarely exploited natural experiments. To remedy this shortage, we describe three main types of studies exploiting natural experiments (standard natural experiments, instrumental-variable designs, and regression-discontinuity designs) and provide examples from psychology and economics to illustrate how natural experiments can be harnessed. Natural experiments are challenging to find, provide information about only specific causal effects, and involve assumptions that are difficult to validate empirically. Nevertheless, we argue that natural experiments provide valuable causal-inference opportunities that have not yet been sufficiently exploited by psychologists.

Within this study ten different self-emulsifying drug delivery systems (SEDDS) mainly containing Labrasol, Kolliphor EL, Cremophor RH40, Transcutol, dimethyl isosorbide, PEG 400 and various types of Capmul were evaluated for their potential mucosal use. These formulations were evaluated regarding the minimum extent of dilution needed to guarantee self-emulsification, emulsification time, stability in different body fluids including simulated saliva, vaginal and tear fluid, dynamic viscosity and cytotoxicity. Furthermore, log DSEDDS/Waterwas determined for cyclosporine used as model drug. Results showed that SEDDS have to be diluted to a final concentration of at least ≤ 40% to guarantee their self-emulsification. Emulsification time was in dependence of the type of formulation between 30 s and 30 min. The resulting oily droplets maintained their size over the entire observation period of 4 h. An up to 3.8-fold higher dynamic viscosity was observed for concentrated SEDDS (30% (v/v)) in comparison to diluted SEDDS (1% (v/v)). Log D SEDDS/Waterfor cyclosporine was in dependence on the type of formulation between 1.7 and 3.6. According to these results, SEDDS seem to be promising delivery systems for lipophilic drugs even on mucosal membranes where comparatively low quantities of body fluids are available for emulsification of the system.

Correlation does not imply causation; but often, observational data are the only option, even though the research question at hand involves causality. This article discusses causal inference based on observational data, introducing readers to graphical causal models that can provide a powerful tool for thinking more clearly about the interrelations between variables. Topics covered include the rationale behind the statistical control of third variables, common procedures for statistical control, and what can go wrong during their implementation. Certain types of third variables—colliders and mediators—should not be controlled for because that can actually move the estimate of an association away from the value of the causal effect of interest. More subtle variations of such harmful control include using unrepresentative samples, which can undermine the validity of causal conclusions, and statistically controlling for mediators. Drawing valid causal inferences on the basis of observational data is not a mechanistic procedure but rather always depends on assumptions that require domain knowledge and that can be more or less plausible. However, this caveat holds not only for research based on observational data, but for all empirical research endeavors.

In psychological science, researchers often pay particular attention to the distinction between within- and between-persons relationships in longitudinal data analysis. Here, we aim to clarify the relationship between the within- and between-persons distinction and causal inference and show that the distinction is informative but does not play a decisive role in causal inference. Our main points are threefold. First, within-persons data are not necessary for causal inference; for example, between-persons experiments can inform about (average) causal effects. Second, within-persons data are not sufficient for causal inference; for example, time-varying confounders can lead to spurious within-persons associations. Finally, despite not being sufficient, within-persons data can be tremendously helpful for causal inference. We provide pointers to help readers navigate the more technical literature on longitudinal models and conclude with a call for more conceptual clarity: Instead of letting statistical models dictate which substantive questions researchers ask, researchers should start with well-defined theoretical estimands, which in turn determine both study design and data analysis.

Behavioral researchers increasingly recognize the need for more diverse samples that capture the breadth of human experience. Current attempts to establish generalizability across populations focus on threats to validity, constraints on generalization, and the accumulation of large, cross-cultural data sets. But for continued progress, we also require a framework that lets us determine which inferences canbe drawn and how to make informative cross-cultural comparisons. We describe a generative causal-modeling framework and outline simple graphical criteria to derive analytic strategies and implied generalizations. Using both simulated and real data, we demonstrate how to project and compare estimates across populations and further show how to formally represent measurement equivalence or inequivalence across societies. We conclude with a discussion of how a formal framework for generalizability can assist researchers in designing more informative cross-cultural studies and thus provides a more solid foundation for cumulative and generalizable behavioral research.

Path models to test claims about mediation and moderation are a staple of psychology. But applied researchers may sometimes not understand the underlying causal inference problems and thus endorse conclusions that rest on unrealistic assumptions. In this article, we aim to provide a clear explanation for the limited conditions under which standard procedures for mediation and moderation analysis can succeed. We discuss why reversing arrows or comparing model fit indices cannot tell us which model is the right one and how tests of conditional independence can at least tell us where our model goes wrong. Causal modeling practices in psychology are far from optimal but may be kept alive by domain norms that demand every article makes some novel claim about processes and boundary conditions. We end with a vision for a different research culture in which causal inference is pursued in a much slower, more deliberate, and collaborative manner.

Psychological theories often invoke interactions but remain vague regarding the details. As a consequence, researchers may not know how to properly test them and may potentially run analyses that reliably return the wrong answer to their research question. We discuss three major issues regarding the prediction and interpretation of interactions. First, interactions can be removable in the sense that they appear or disappear depending on scaling decisions, with consequences for a variety of situations (e.g., binary or categorical outcomes, bounded scales with floor and ceiling effects). Second, interactions may be conceptualized as changes in slope or changes in correlations, and because these two phenomena do not necessarily coincide, researchers might draw wrong conclusions. Third, interactions may or may not be causally identified, and this determines which interpretations are valid. Researchers who remain unaware of these distinctions might accidentally analyze their data in a manner that returns the technically correct answer to the wrong question. We illustrate all issues with examples from psychology and issue recommendations for how to best address them in a productive manner.