Bochen Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Bochen Li has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Charles Silber
Department of Mathematics, College of Education, Rutgers University, New Brunswick, NJ 08901, USA
3 shared publications Fernando Munoz Gomez Andrade
Willamette University, Salem, OR, USA
3 shared publications- Justin Gasper Jacques
Threat Assessment and Disaster Behavioral Health Service, Bolante.net, Salem, OR 97304, USA
3 shared publications Alina Hyk
Oregon State University, Corvallis, OR, USA
2 shared publicationsAndrew Do
Oregon State University, Corvallis, OR, USA
2 shared publicationsBriana Hoxha
Oregon State University, Corvallis, OR, USA
2 shared publications
Latest Publications
- Open Access
- 23 September 2026
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 5, e93498
Implicit Bias in Large Language Model Diagnosis of Eating Disorders: Experimental Vignette Study
Background: Large language models (LLMs) are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. Objective: This study aims to examine whether LLMs exhibit implicit demographic biases when generating psychiatric diagnoses. Methods: We developed 1152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race and ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa [AN]) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across 5 model families were tested. Results: Control vignettes produced near-unanimous AN diagnoses (mean 100%, SD 0.1%), while ambiguous vignettes elicited greater variability (mean 23.6%, SD 10.1%). Intermodel agreement was moderate for ambiguous vignettes (Fleiss κ=0.410, 95% CI 0.397‐0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over 6 times more likely to receive a major depressive disorder (MDD) diagnosis than White patients with identical presentations (odds ratio [OR] 6.09, 95% CI 5.13‐7.24), Latine patients were over 9 times more likely (OR 9.57, 95% CI 8.00‐11.45), and Asian patients were nearly 3 times more likely to receive an AN diagnosis (OR 2.88, 95% CI 2.44‐3.42). Female patients were less likely than males to be diagnosed with AN (OR 0.43, 95% CI 0.37‐0.49). Conclusions: These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.
- 23 September 2026
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 5, e93498
Open AccessImplicit Bias in Large Language Model Diagnosis of Eating Disorders: Experimental Vignette Study
Background: Large language models (LLMs) are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. Objective: This study aims to examine whether LLMs exhibit implicit demographic biases when generating psychiatric diagnoses. Methods: We developed 1152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race and ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa [AN]) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across 5 model families were tested. Results: Control vignettes produced near-unanimous AN diagnoses (mean 100%, SD 0.1%), while ambiguous vignettes elicited greater variability (mean 23.6%, SD 10.1%). Intermodel agreement was moderate for ambiguous vignettes (Fleiss κ=0.410, 95% CI 0.397‐0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over 6 times more likely to receive a major depressive disorder (MDD) diagnosis than White patients with identical presentations (odds ratio [OR] 6.09, 95% CI 5.13‐7.24), Latine patients were over 9 times more likely (OR 9.57, 95% CI 8.00‐11.45), and Asian patients were nearly 3 times more likely to receive an AN diagnosis (OR 2.88, 95% CI 2.44‐3.42). Female patients were less likely than males to be diagnosed with AN (OR 0.43, 95% CI 0.37‐0.49). Conclusions: These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.
- 6 September 2026
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
Examining Personality and Decision-Making Style as Predictors of Automated Aid Efficiency
Automated diagnostic aids can improve decision making, yet operators often fail to use them effectively. This study examined whether personality traits and decision-making styles predict aid-use efficiency in a signal detection task. One hundred seventy-five undergraduate participants completed a signal detection task with and without a 93% reliable automated decision aid and completed measures of conscientiousness, neuroticism, satisficing, and decision difficulty. After preregistered exclusions, Bayesian analyses examined whether individual differences predicted aid-use efficiency, quantified as the ratio between observed and optimal aided performance. The aid significantly improved sensitivity, providing evidence of its diagnostic value. However, Bayesian correlations provided evidence against relationships between aid-use efficiency and conscientiousness, neuroticism, satisficing, or decision difficulty. Additional analyses showed that neuroticism was associated with greater decision difficulty, whereas conscientiousness was associated with lower decision difficulty. These findings suggest that individual differences in personality and decision style may not explain inefficiencies in automation use.
- 6 September 2026
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
Examining Personality and Decision-Making Style as Predictors of Automated Aid Efficiency
Automated diagnostic aids can improve decision making, yet operators often fail to use them effectively. This study examined whether personality traits and decision-making styles predict aid-use efficiency in a signal detection task. One hundred seventy-five undergraduate participants completed a signal detection task with and without a 93% reliable automated decision aid and completed measures of conscientiousness, neuroticism, satisficing, and decision difficulty. After preregistered exclusions, Bayesian analyses examined whether individual differences predicted aid-use efficiency, quantified as the ratio between observed and optimal aided performance. The aid significantly improved sensitivity, providing evidence of its diagnostic value. However, Bayesian correlations provided evidence against relationships between aid-use efficiency and conscientiousness, neuroticism, satisficing, or decision difficulty. Additional analyses showed that neuroticism was associated with greater decision difficulty, whereas conscientiousness was associated with lower decision difficulty. These findings suggest that individual differences in personality and decision style may not explain inefficiencies in automation use.
- Open Access
- 19 April 2026
- Published by Center for Open Science in PsyArXiv
AI-Driven Knowledge Extraction of Therapeutic Rupture Patterns from Counseling Corpora Using LIWC and Machine Learning Classification
Therapeutic alliance ruptures predict treatment outcomes, yet clinicians often struggle to detect them in real time because traditional identification relies on labor-intensive manual coding. The linguistic signatures that distinguish confrontation, withdrawal, and mixed ruptures from intact therapeutic alliances remain insufficiently characterized. The present study analyzed 21,056 words drawn from validated rupture demonstration videos (nine sessions across three rupture types) and master therapist reference sessions (three sessions) using Linguistic Inquiry and Word Count (LIWC-2015) and Random Forest machine learning with leave-one-out cross-validation. Thirteen LIWC variables significantly discriminated rupture types at Bonferroni-corrected thresholds. Third-person pronouns (she/he) showed the strongest effect (G² = 31.85), appearing approximately thirteen times more frequently in withdrawal than confrontation ruptures. Confrontation ruptures exhibited elevated certainty language (2.32% vs. 0.80%), whereas mixed ruptures showed the highest levels of negative emotion language (3.25% vs. 1.61%). Random Forest classifiers achieved perfect accuracy for multi-class rupture classification (100%, F1 = 1.00) and binary rupture detection (100%, AUC-ROC = 1.00), with conversational bigrams such as “it s” (importance = 0.0387) contributing most strongly to classification. Unexpectedly, rupture sessions contained more positive emotion language than reference sessions (4.67% vs. 2.54%), suggesting attempts at relational repair during alliance strain. Given the small sample size and structured nature of the corpus, these classification results should be interpreted as an exploratory demonstration rather than evidence of generalizable predictive performance. These findings identify distinctive linguistic patterns associated with therapeutic alliance ruptures and provide preliminary evidence supporting the feasibility of automated rupture detection, although validation using naturalistic psychotherapy data remains necessary.
- 19 April 2026
- Published by Center for Open Science in PsyArXiv
Open AccessAI-Driven Knowledge Extraction of Therapeutic Rupture Patterns from Counseling Corpora Using LIWC and Machine Learning Classification
Therapeutic alliance ruptures predict treatment outcomes, yet clinicians often struggle to detect them in real time because traditional identification relies on labor-intensive manual coding. The linguistic signatures that distinguish confrontation, withdrawal, and mixed ruptures from intact therapeutic alliances remain insufficiently characterized. The present study analyzed 21,056 words drawn from validated rupture demonstration videos (nine sessions across three rupture types) and master therapist reference sessions (three sessions) using Linguistic Inquiry and Word Count (LIWC-2015) and Random Forest machine learning with leave-one-out cross-validation. Thirteen LIWC variables significantly discriminated rupture types at Bonferroni-corrected thresholds. Third-person pronouns (she/he) showed the strongest effect (G² = 31.85), appearing approximately thirteen times more frequently in withdrawal than confrontation ruptures. Confrontation ruptures exhibited elevated certainty language (2.32% vs. 0.80%), whereas mixed ruptures showed the highest levels of negative emotion language (3.25% vs. 1.61%). Random Forest classifiers achieved perfect accuracy for multi-class rupture classification (100%, F1 = 1.00) and binary rupture detection (100%, AUC-ROC = 1.00), with conversational bigrams such as “it s” (importance = 0.0387) contributing most strongly to classification. Unexpectedly, rupture sessions contained more positive emotion language than reference sessions (4.67% vs. 2.54%), suggesting attempts at relational repair during alliance strain. Given the small sample size and structured nature of the corpus, these classification results should be interpreted as an exploratory demonstration rather than evidence of generalizable predictive performance. These findings identify distinctive linguistic patterns associated with therapeutic alliance ruptures and provide preliminary evidence supporting the feasibility of automated rupture detection, although validation using naturalistic psychotherapy data remains necessary.
- Open Access
- 13 February 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Implicit Bias in Large Language Model Diagnosis of Eating Disorders: Experimental Vignette Study (Preprint)
BACKGROUND Large language models are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. OBJECTIVE To examine whether large language models (LLMs) exhibit implicit demographic biases when generating psychiatric diagnoses. METHODS We developed 1,152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race/ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across five model families were tested. RESULTS Control vignettes produced near-unanimous anorexia nervosa diagnoses (M = 100.0%), while ambiguous vignettes elicited greater variability (M = 23.6%). Inter-model agreement was moderate for ambiguous vignettes (Fleiss' κ = 0.410, 95% CI: 0.397–0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over six times more likely to receive a major depressive disorder diagnosis than White patients with identical presentations (OR = 6.09, 95% CI: 5.13–7.24), Latine patients were over nine times more likely (OR = 9.57, 95% CI: 8.00–11.45), and Asian patients were nearly three times more likely to receive an anorexia nervosa diagnosis (OR = 2.88, 95% CI: 2.44–3.42). Female patients were less likely than males to be diagnosed with anorexia nervosa (OR = 0.43, 95% CI: 0.37–0.49). CONCLUSIONS These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.
- 13 February 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessImplicit Bias in Large Language Model Diagnosis of Eating Disorders: Experimental Vignette Study (Preprint)
BACKGROUND Large language models are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. OBJECTIVE To examine whether large language models (LLMs) exhibit implicit demographic biases when generating psychiatric diagnoses. METHODS We developed 1,152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race/ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across five model families were tested. RESULTS Control vignettes produced near-unanimous anorexia nervosa diagnoses (M = 100.0%), while ambiguous vignettes elicited greater variability (M = 23.6%). Inter-model agreement was moderate for ambiguous vignettes (Fleiss' κ = 0.410, 95% CI: 0.397–0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over six times more likely to receive a major depressive disorder diagnosis than White patients with identical presentations (OR = 6.09, 95% CI: 5.13–7.24), Latine patients were over nine times more likely (OR = 9.57, 95% CI: 8.00–11.45), and Asian patients were nearly three times more likely to receive an anorexia nervosa diagnosis (OR = 2.88, 95% CI: 2.44–3.42). Female patients were less likely than males to be diagnosed with anorexia nervosa (OR = 0.43, 95% CI: 0.37–0.49). CONCLUSIONS These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.
- 1 September 2025
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
- Vol. 69 (1) , 1913-1918
Sample Size Justifications and Open Science Practices in Human Factors (2020–2024)
Low rates of replicability in behavioral and cognitive sciences have prompted large-scale efforts to increase transparency and openness in reporting. Such practices directly impact researchers’ ability to evaluate studies, replicate procedures, and reproduce empirical findings. This study evaluated the prevalence of research transparency practices (e.g., disclosure of sample size determination, preregistration) and research openness (e.g., share of data, code, and stimuli) across all articles published in Human Factors over the years 2020 to 2024. Most articles made no mention of transparency and openness practices. Future initiatives to improve transparency and openness in Human Factors (e.g., adopting formal policies at the journal level) may be worthwhile.
- 1 September 2025
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
- Vol. 69 (1) , 1913-1918
Sample Size Justifications and Open Science Practices in Human Factors (2020–2024)
Low rates of replicability in behavioral and cognitive sciences have prompted large-scale efforts to increase transparency and openness in reporting. Such practices directly impact researchers’ ability to evaluate studies, replicate procedures, and reproduce empirical findings. This study evaluated the prevalence of research transparency practices (e.g., disclosure of sample size determination, preregistration) and research openness (e.g., share of data, code, and stimuli) across all articles published in Human Factors over the years 2020 to 2024. Most articles made no mention of transparency and openness practices. Future initiatives to improve transparency and openness in Human Factors (e.g., adopting formal policies at the journal level) may be worthwhile.
- 1 September 2025
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
- Vol. 69 (1) , 1858-1861
The Prevalence of Replication Studies in Human Factors (2020-2024)
Replication studies contribute to scientific self-correction and assessing the credibility of a given effect. This study examined the prevalence of replication studies published in Human Factors from 2020 to 2024 ( N = 550). Of the studies coded, 8.59% explicitly identified replication as an aim. Of these, 19.51% were direct replications, 24.39% were conceptual replications, and 56.10% were direct replications with extensions. About three-quarters of replication efforts were conducted by non-independent teams (i.e., at least one author shared with the original publication) while the rest were independent (i.e., no authors in common with the original publication). These findings suggest that replication studies are more common in ergonomics and human factors than in some related fields, but that more independent replication efforts might be valuable.
- 1 September 2025
- Published by SAGE Publications in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
- Vol. 69 (1) , 1858-1861
The Prevalence of Replication Studies in Human Factors (2020-2024)
Replication studies contribute to scientific self-correction and assessing the credibility of a given effect. This study examined the prevalence of replication studies published in Human Factors from 2020 to 2024 ( N = 550). Of the studies coded, 8.59% explicitly identified replication as an aim. Of these, 19.51% were direct replications, 24.39% were conceptual replications, and 56.10% were direct replications with extensions. About three-quarters of replication efforts were conducted by non-independent teams (i.e., at least one author shared with the original publication) while the rest were independent (i.e., no authors in common with the original publication). These findings suggest that replication studies are more common in ergonomics and human factors than in some related fields, but that more independent replication efforts might be valuable.