📚 Research Digest

Finding: This registered replication (N=3,323) failed to replicate Gino et al.'s (2009) finding that witnessing ingroup dishonesty increases individual dishonest behavior more than outgroup dishonesty—Bayesian analyses provided moderate-to-strong evidence for null effects across all comparisons.

Why it matters: Directly relevant to behavioral science and ethics teaching—challenges a seminal social norms finding about how observing others' unethical behavior spreads dishonesty, which has implications for understanding social contagion in teams and nudge/norm-based behavior change interventions.

Method: Registered report with conceptual replication using both minimal and natural groups (N=3,323), incentivized coin-flip lying task, and Bayesian analysis to quantify evidence for null effects.

Jargon: *Registered report*: study design pre-registered before data collection to reduce publication bias. *Minimal groups*: arbitrary group assignments with no meaningful history. *BF₀₁*: Bayes Factor favoring the null hypothesis over the alternative—values >3 indicate moderate evidence for null.

Finding: Displaying judgment certainty in self-formed information networks reorganizes who people choose to observe (toward more certain sources) but does not improve collective accuracy—and can actually hurt performance by increasing exposure to highly certain but incorrect judgments.

Why it matters: Directly relevant to collective intelligence, team decision-making, and how UI design choices (what information to display) shape information aggregation and belief revision in collaborative settings—with implications for HCI design of social/collaborative platforms and misinformation spread.

Method: 8-session online experiment (n=273) with costly directed observation networks, manipulating visibility of certainty while hiding accuracy and belief direction; tested six predictive models including DeGroot belief updating.

Jargon: *DeGroot model*: a social learning model where agents update beliefs by weighted averaging of neighbors' opinions. *Costly directed networks*: participants pay to observe others' judgments, so link formation reflects genuine information-seeking choices. *Net scores*: payoff structure penalizing miscalibrated confidence.

Finding: The Bridge-or-Trap model proposes that AI companions either facilitate human connection (bridge) or substitute for it (trap) depending on whether AI-mediated emotional relief reduces anticipated interpersonal costs and supports human-directed action versus reinforcing avoidance.

Why it matters: Directly relevant to AI's social impact and behavior change design—specifically how AI affordances can function as either supportive scaffolds or dependency-inducing traps, with implications for persuasive technology ethics and dark patterns.

Method: Theoretical model synthesis drawing on attachment theory, social anxiety research, social exchange theory, and resilience theory; proposes three empirical tests (target-specific transfer, viable-alternative boundaries, asymmetric recovery).

Jargon: *Path-dependent substitution*: once a user settles into AI-only emotional relief, exit becomes progressively harder due to reinforced avoidance; *corrective human experience*: real-world social interactions that update negative interpersonal expectations; *ecological protection and affordance*: the real-world social environment's role in either enabling or blocking human reconnection.

Finding: A Bayesian nonparametric framework (Chinese Restaurant Process) explains how people form and update moral impressions of groups, revealing a "sparsity bias" where rare behaviors disproportionately shape group moral judgments, opposing the common "Common Is Moral" heuristic.

Why it matters: The finding that social biases can be mitigated by strategically altering information distributions has peripheral relevance to behavioral science, decision-making, and nudge/choice architecture work—particularly around how group reputation and collective moral judgment form.

Method: Combines behavioral experiments, computational modeling, and Representational Similarity Analysis (RSA) to map mental representations of groups.

Jargon: *Chinese Restaurant Process (CRP)*: A Bayesian nonparametric model that clusters data into latent groups, where new observations can either join existing clusters or form new ones. *Sparsity bias*: Tendency for rare behaviors to have outsized influence on judgments. *RSA*: A method comparing the structure of neural/behavioral similarity matrices to model predictions.

Finding: Central form properties of concepts trigger stable, fixed beliefs that persist regardless of change, potentially explaining phenomena like functional fixedness, stereotyping, and fixed mindsets through a universal cognitive mechanism.

Why it matters: The concepts of functional fixedness and fixed mindsets are directly relevant to creativity research, design thinking, and learning sciences — this mechanism could explain why people resist novel uses of objects or fail to update mental models.

Method: Two experiments testing belief stability for form vs. non-form properties across familiar and newly learned categories.

Jargon: *Central form* (Form Theory): a compressed, unitary core structure within a concept that exerts default influence on cognition; *functional fixedness*: cognitive bias where objects are only seen as serving their typical function.

Finding: Integrating Positive Psychology principles with AI-supported language instruction produced greater gains in language proficiency and socio-emotional outcomes than AI-only or traditional instruction, with enjoyment partially mediating achievement gains.

Why it matters: Touches on AI's impact on learning and behavior change (enjoyment as mediator, emotionally supportive design), but is narrowly focused on EFL instruction in Middle Eastern higher education rather than HCI, creativity, teams, or behavioral science more broadly.

Method: Cluster-randomized mixed-methods design across 300 learners in 12 universities; mediation analysis linking instructional condition → enjoyment → achievement.

Jargon: *PERMA* = Positive Psychology well-being model (Positive emotion, Engagement, Relationships, Meaning, Achievement); *Academic buoyancy* = ability to manage everyday academic setbacks; *FLE* = Foreign Language Enjoyment scale.

Finding: Montessori's work is reframed as empirical cognitive science, structured around individual variation, designed material environments, and socio-cultural context — dimensions that anticipate recent cognitive science frameworks.

Why it matters: Has peripheral relevance to learning sciences and the role of designed environments in cognitive development, but the focus is primarily historical/philosophical reconstruction of Montessori's scientific legacy, not directly actionable for HCI, creativity, teams, or behavioral science teaching.

Method: Historical analysis drawing on Italian-language and untranslated primary sources to reconstruct Montessori's intellectual lineage.

Finding: Cognitive hints reduce knowledge gaps for lower-ability 5th-grade math students but have negligible effects on high performers; webcam-based eye-tracking can reliably capture learning analytics at scale despite lower data quality.

Why it matters: Marginally relevant to learning sciences and skill acquisition, but focuses narrowly on math education and eye-tracking methodology rather than HCI, behavioral science, or the professor's core teaching areas.

Method: Webcam-based eye-tracking with Linear Mixed Models to analyze gaze patterns across within-subject rotated task sets (n=134).

Finding: An AR-based instructional framework integrating brain-based learning principles and local cultural values improved prosocial, critical thinking, and communicative behavior in preschool children.

Why it matters: Marginally relevant to learning sciences and AR/technology-enhanced learning, but focused on early childhood character education in a specific cultural context (Indonesian/Balinese values) rather than the professor's core interests.

Method: ADDIE instructional design model with quasi-experimental evaluation and expert validity review.

Finding: PBL outperforms Discovery Learning for developing analytical thinking and science process skills in high school biology, with an interaction effect showing PBL is better for low-skill students while Discovery Learning suits high-skill students.

Why it matters: The interaction between prior skill level and instructional method has some relevance to learning sciences and adaptive instruction design, but the domain is narrowly focused on high school biology with limited connection to HCI, creativity, teams, or behavioral science.

Method: Quasi-experimental posttest-only design with Two-Way ANOVA on 52 students, comparing PBL vs. Discovery Learning conditions.

Finding: A Scratch-based interactive learning medium (MEINSIPA) for civic education showed moderate effectiveness (N-Gain 0.33 vs 0.27 control) in improving cognitive, affective, and psychomotor outcomes for 6th graders.

Why it matters: Tangentially relevant to games/gamification and learning sciences, but the context (Indonesian civic education, Pancasila values) and modest effect sizes make it unlikely to be useful for this professor's teaching or research.

Method: Borg and Gall R&D model with quasi-experimental design comparing experimental and control groups using N-Gain scores.

Finding: A 4-year longitudinal study comparing traditional vs. experiential learning (Kolb's model) approaches to architectural site analysis, finding significantly better design outcomes with the experiential approach (Cohen's d = 0.67).

Why it matters: Marginally relevant to learning sciences and skill acquisition, but focused narrowly on architecture/urban design pedagogy rather than HCI, creativity tools, teams, behavioral science, or AI.

Method: 4-year longitudinal quasi-experiment with student focus groups and effect size validation (Cohen's d).