Wen Wei Loh

Psychologists often rely on observational studies for causal evidence, which raises the question of how to strengthen the design of observational studies for causal inference. One widely used approach in medical research is target trial emulation (TTE). In this tutorial, we introduce TTE to psychological research, with particular emphasis on its cornerstone concept that is routinely ignored in practice: time zero. Following a brief primer on TTE, we foreground time zero and highlight its critical role in improving valid causal inferences from observational data. Using examples drawn from psychological research, we systematically enumerate all possible permutations of the time zero components to show how their misalignment can introduce systematic biases, including selection bias, post-treatment confounding, treatment-dependent (in)eligibility, and immortal time bias, even when all baseline confounders are measured. To address this, we develop a decision framework for integrating time zero into research practices and offer practical recommendations. Through a running example of workplace AI use and employee burnout, we demonstrate how clearly specifying time zero clarifies causal questions, prevents many design-based biases, and strengthens internal validity. We further emphasize that time zero alignment is only one design feature of a strong observational study, and we delineate the boundaries of TTE, cautioning against the misconception that it is a magic fix that automatically delivers causal conclusions. We conclude that explicitly justifying time zero is a necessary step toward improving the rigor and transparency of causal research in psychology.

Longitudinal designs can fortify causal inquiries of a focal predictor (i.e., treatment) on an outcome. But valid causal inferences are complicated by causal feedback between confounders and treatment over time. G-estimation of a structural nested mean model (SNMM) is designed to handle the complexities beset by measured time-varying or treatment-dependent confounding in longitudinal data. But valid inference requires correctly specifying the functional form of the SNMM, such as how the effects stay constant or change over time. In this article, we develop a g-estimation strategy for linear structural nested mean models whose causal parameters adopt the form of time-varying coefficient functions. These time-varying coefficient functions are smooth semiparametric functions of time that permit probing how the treatment effects may change curvilinearly. Further effect modification by time-invariant and time-varying covariates can be readily postulated in the SNMM to test fine-grained effect heterogeneity. We then describe a g-estimation strategy for estimating such an SNMM. We utilize the established time-varying effect model (TVEM) approach from the prevention and psychotherapy research literature for modeling flexible changes in covariate-outcome associations over time. Moreover, we exploit a known benefit of g-estimation over routine regression methods: its double robustness conferring protection against biases induced by certain forms of model misspecification. We encourage psychology researchers seeking correct causal conclusions from longitudinal data to use an SNMM with time-varying coefficient functions to assess curvilinear causal effects over time, and to use g-estimation with TVEM to resolve measured treatment-dependent confounding. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Background: Randomised trials evaluating psychological interventions for perinatal depression (PND) in low- and middle-income countries (LMICs) frequently provide conflicting findings. To optimise global maternal mental health interventions, we applied robust causal inference and causal machine learning approaches to individual participant data (IPD) to identify underlying causal mechanisms and responsive subgroups.Methods: We searched MEDLINE, Embase, PsycINFO, CINAHL, Web of Science, and the Cochrane Library from inception to November 19, 2025, for trials assessing the effectiveness of psychological interventions to treat PND. Using IPD, we assessed treatment effect heterogeneity using causal forests, and evaluated causal mechanisms using interventional effects for individual trials and across relevant pooled datasets.Findings: Seven of 11 eligible studies provided IPD, yielding a pooled sample of 3,154 participants (1,700 intervention, 1,454 control). Heterogeneity analyses of all trials demonstrated top responders presented with higher baseline depression and lower education. A four-study subset incorporating social determinants revealed those with lower perceived social support and education responded best. This four-study subset comprised two two-trial subsets: one delivered by community health workers (CHWs) and one by peers. In the CHW-delivered subset, top responders were similar to the four-study subset; however, within the peer-delivered subset, this effect was negligible. Causal mediation analyses of the four trials revealed social support mediated a 15% symptom reduction. The two-trial CHW subset indicated social support and reduced food insecurity mediated 15% and 3% improvements, respectively. Furthermore, a specific two-trial subset indicated behavioural activation and social support mediated improvements of 32% and 17%.Interpretation: Despite interventions effectively targeting perceived social support, trials capturing food insecurity demonstrated minimal impact. The failure of peer-delivered interventions to reach vulnerable women, unlike CHW-delivered trials, suggests highly disadvantaged populations benefit least from a lower-cadre workforce. Optimising global maternal mental health requires assertively targeting socioeconomic determinants such as food insecurity and deploying CHWs to support the most vulnerable populations.

Background: Preeclampsia remains a leading cause of preterm delivery (PTD). Conventional average causal effect estimands may not always be perfectly suited for assessing preeclampsia’s contributions because they rely on uniform counterfactual scenarios, such as everyone suffering from preeclampsia. A complementary approach is the incremental propensity score intervention (IPSI), which can address a more action-oriented causal inquiry: How would the population average risk of PTD change from current observed levels if each individual’s odds of preeclampsia were reduced by a given amount? Methods: We applied IPSI to two US population cohorts of singleton pregnancies, separated by two decades (1998–2002, 16.7 million; 2020–2023, 14.2 million), to evaluate the causal effects of gestational hypertension/preeclampsia (GHTN/PE) on PTD within each cohort. We developed and implemented a sensitivity analysis procedure using bias formulas to gauge the impact of potential unmeasured confounding. Results: Between 1998 and 2002, halving the individual odds of GHTN/PE would reduce the population average risk of PTD from observed [risk ratio (RR) = 0.969, 95% confidence interval (CI) 0.969–0.970]. Between 2020 and 2023, when GHTN/PE was more prevalent, halving the odds of preeclampsia would lead to an even greater reduction in average PTD risk (RR = 0.948, 95% CI 0.947–0.948). Sensitivity analyses demonstrated that these conclusions held under assumptions encoding relatively strong unmeasured confounding. Conclusions: The IPSI approach offers more nuanced understanding and clinically meaningful interpretations of causal effects, such as by how much partially reducing the odds of GHTN/PE, while allowing it to remain non-zero, can lower the average risk of PTD.

Generalizability is a perennial concern in randomized studies. While randomized studies are the gold standard for establishing causality, study samples are rarely representative of broader populations due to factors such as convenience sampling, participant self-selection, and researchers' inclusion or exclusion criteria. To strengthen the generalizability of randomized experiments, the framework of causal effect generalizability provides a solution. However, existing methods require accessing representative individual-level data from the target population, which is often unavailable due to limited resources, data access restrictions, or privacy concerns. In this paper, we develop a novel method to generalize causal effects using only summary statistics on covariates from the target population. We illustrate the estimator using a real-world study by generalizing the impact of a climate change behavioral intervention from the study sample to a broader population. By avoiding the need for individual-level data from the target population, our method offers a practical tool for generalizing causal findings from randomized studies. We hope that the proposed method helps build more accurate theories and enhance the policy relevance of behavioral and psychological research.

Does ostracism disproportionally affect certain groups? Which groups are at greater risk, and in which social domains (e.g., family, workplace)? To address these questions, we adopted a descriptive approach, leveraging a probability-based sample of Dutch adults, with 5,290 participants and 22,075 ratings of ostracism experiences in various social domains. The distribution of ostracism ratings was zero-inflated and extremely skewed: the majority of values clustered at the left end of the spectrum, indicating ostracism as a rare experience; a smaller number of values formed a tail of statistical outliers indicating ostracism as a frequent experience. Low-income (vs. high) earners had over three times higher odds of facing ostracism frequently by family members and people at work. Immigrants had two to three times higher odds of experiencing ostracism frequently than non-immigrants across all domains. These results highlight group-based disparities in ostracism experiences and identify priority populations in need of support.

In behavioral sciences, researchers frequently employ mediation analysis with longitudinal data. A common scenario involves the 1 → 1 → 1 mediation model, where the Predictor X, Mediator(s) M, and Outcome Y are all measured at different occasions (Level 1) within individuals (Level 2). The standard 1 → 1 → 1 mediation model approach fits two multilevel models, one for the mediator and one for the outcome, with two random intercepts and three random slopes in total. However researchers often exclude random slopes from multilevel models and only include random intercepts to account for non-independence across observations of the same individual. We demonstrate that ignoring random slopes in the 1 → 1 → 1 mediation model can result in biased average indirect effect estimators, as well as underestimated standard errors. We provide code from open source and free statistical software that can be used by practitioners to fit the 1 → 1 → 1 mediation model.

The field of ostracism research is witnessing a growing interest in understanding ostracism - being excluded and ignored - as a lived experience outside of the laboratory context. How do researchers draw valid causal conclusions about naturally occurring experiences of ostracism without relying on experimental designs? In this article, we draw on insights from the well-established causal inference framework to emphasize a critical step for strengthening causal rigor: stating the causal estimand. Using an intuitive example, we illustrate what a causal estimand is, how to define it, and why it matters. With this article, we encourage readers to think clearly about causal estimands before conducting any data analysis. This conceptual step holds the potential for enhancing the rigor and precision of research studying ostracism as a naturally occurring phenomenon.