The investigators used adolescent data from the Boston Adolescent Neuroimaging of Depression and Anxiety (BANDA) study to examine how analytic choices alter observed associations between childhood stress and medial temporal volumes. They did not rely on a single pipeline. Instead, the team systematically crossed three region-of-interest operationalizations (left, right, total) with four intracranial volume scaling methods and varied sociodemographic covariates to generate a large set of plausible specifications. This strategy allowed the researchers to map which choices most strongly influence outcomes and which associations are robust across choices.
The analysis focused on lifetime stress severity as the primary exposure. The paper reported results across the many specification models rather than presenting a single “best” model. The source did not report specific sample sizes, exact scaling algorithms, or all covariates in full detail within the abstract; those methodological specifics were not provided in the previewed material.
Across the primary multiverse, greater lifetime stress severity associated with smaller amygdala volume in 72% of specifications. That effect emerged in the majority of analytic permutations. Statistical significance, however, was sensitive to how the region of interest was defined. When investigators used left, right, or total operationalizations, the pattern of significance shifted—indicating that hemisphere choice materially affected inference.
For the hippocampus, the relationship was less consistent. Greater lifetime stress severity associated with smaller hippocampal volume in 52% of specification models. The study team observed that hippocampal findings varied more across analytic choices than amygdala findings, suggesting a weaker or more model-dependent association in this cohort. The authors reported these proportional outcomes directly; no additional effect sizes or p-values are reported in the source abstract.
The source explicitly attributed a substantial portion of prior literature inconsistency to reasonable but arbitrary analytic decisions—especially the operationalization of hemisphere. By quantifying the proportion of specifications yielding an association, the analysis parsed robustness from specification-dependent results. Investigators reported that hemisphere operationalization was among the analytic decisions exerting the largest influence on statistical outcomes.
The team performed a secondary multiverse analysis that tested six operationalizations of stress exposure: combinations of total, acute, and chronic exposure crossed with count versus severity measures. This secondary sweep examined whether how stress is defined would change the laterality pattern observed in the primary analysis.
Data indicated that the laterality pattern—differences tied to left versus right versus total region definitions—was not strongly altered by which stress operationalization investigators selected. In other words, varying stress definitions did not eliminate the sensitivity of results to hemisphere choice. The source did not present detailed comparative statistics for each stress operationalization in the abstract; it only reported that laterality effects persisted across the secondary specifications.
The authors framed this finding as evidence that the stress–medial temporal relationship shows degree of robustness (particularly for the amygdala) while pointing to hemisphere choice as a reproducibility risk. They did not claim universality; rather, they provided a mapped landscape of which analytic decisions changed outcomes.
Investigators noted that the ethics committee/IRB at the University of Pittsburgh approved the work and that participant consent procedures were followed. The source declares no competing interests. Data produced for the study are available through the Human Connectome Project release linked in the preprint, and the funding source listed is the National Institute of Mental Health (R01MH136109) as provided in the source information.
The preprint emphasized analytic scope as both a strength and limitation. While the multiverse approach clarifies which choices drive variability, the source did not report all granular methodological parameters in the abstract (for example, exact covariate lists per specification or precise intracranial volume scaling formulas). Those details are necessary to reproduce every specification and were not included in the previewed text. The authors also did not claim that the multiverse resolves causal inference; rather, they positioned the method as a framework for interpreting inconsistent volumetric findings across studies.
The study closes by identifying specific unanswered operational questions that remain: standardizing hemisphere operationalization and detailing covariate treatment across studies. The preprint positions multiverse analysis as a tool to reduce interpretive ambiguity and recommends that future work explicitly report how hemisphere and scaling choices were made so that replication and meta-analysis can proceed from a clearer methodological baseline.