Passive suicidality in a repressive U.S. political context

Abstract

Bodily autonomy reversal (BAR) policies, exemplified by the U.S. Supreme Court’s Dobbs decision overturning abortion access, concurrent with the staggered adoption of state-level gender-affirming care criminalization laws, may represent under-examined drivers of suicidal ideation (SI). Strain theory recognizes aspiration strain, ruptured life potential; deprivation strain, relative socio-material scarcity; and value strain, internalized ideological conflict, as passive SI components. Critical-suicidological principles implicate structural/systemic determinants and dehumanizing political rhetorics in SI formation. Through the lens of strain theory, grounded in critical suicidology, I leveraged public archival Reddit data ($N$ = 1.2M posts) and fine-tuned pretrained LLMs to examine BAR policymaking as an SI driver among BAR-targeted Redditors. A preregistered protocol was filed 2024-10-08 to the Open Science Framework; open Python and Stata code are available on this repository. Aim I: Co-annotators ($n$ = 2) independently and iteratively labeled random subsets of training data, achieving $\kappa$ = 0.72–0.96 by leveraging a novel GPT-4o-assisted human-in-the-loop inter-annotator agreement technique. Classification performance via 5-fold cross-validation was trialed across base BERT, RoBERTa, DistilBERT, and an 8B-parameter Llama 3.1 (applying memory-safe quantization and LoRA weights); hyperparameter-optimized 125M RoBERTa models outperformed across the board: $acc$ = 0.84–0.99, $F_1$ (macro) = 0.77–0.85. SP-LIME post hoc model inspection indicated face-valid human-interpretable learned representations. Explicit SI was encoded using regex; a custom GPT-4o disambiguation function validated self-referential and sincere SI utterances. Aim II: Among posts by disclosed pregnancy-capable Redditors ($n$ = 21K), in Model 1 interrupted time series estimation demonstrated a 1.42% level increase in explicit SI post–Dobbs release ($\beta$ = 4.78, Newey-West SE = 1.10, $p$ = 0.02); the effect was null against a time-matched control series. Descriptively, value strain and explicit SI spiked fleetingly, concurrent with the Politico leak and Dobbs release. Among posts by disclosed transgender and gender-diverse Redditors ($n$ = 146K), LIWC-22 extracted fatalistic (“all-or-none” absolutist and negatively emotive) language, a recognized SI marker. In Model 2, causal inference via the T meta-learner adjusted for relevant text covariates encoded via term frequency–inverse document frequency weighting. Post-level mentions of gender-affirming care criminalization laws and compound strain decreased the likelihood of fatalistic comment threads ($\tau$ = –0.146 and –0.001, respectively). Post-level explicit SI increased comment-thread fatalism ($\tau$ = 0.119). Aim III: Reflexive thematic analysis identified individual (within-case) resilience and cultural repertoires (across-case) of active resistance among multiply marginalized BAR-targeted Redditors: “Duty to the present,” survival for communities and family; “Chains of trauma,” a drive to surmount inter-general transmission of post-traumatic stress; “Embodied joy,” reveling in gendered selfhood; and “Gatecrashing the gatekeeping,” accessing BAR-restricted healthcare informally, though often despite safety and effectiveness concerns. While strain theory offers compelling a priori orienting constructs to conceptualize politically determined SI, its predictive validity was not demonstrated. Nevertheless, institutional failure and oppressive societal structures, including BAR policymaking, are overwhelmingly shown to drive SI qualitatively. Geocoded social media data will offer valuable future opportunities for atheoretical BAR policy impact analysis.

Pre-analysis plan

A pre-analysis plan is filed on the Open Science Framework at osf.io/uc4kh/.