AI, Equity, and the Statistics Classroom

• 867 enrollments (852 students; 1.7% repeat), 17 semesters, Fall 2018 to Fall 2025
• A single instructor at one large, diverse California community college
• Enrollment-level logistic regression with semester-clustered SEs; Spearman rank trend; composition-adjusted robustness checks

• Course success rose 42% → 69% across the reform era (pre-AB705 42% → AB705 49% → post 59% → AB1705 69%)
• Odds of passing +10% per semester (OR 1.10, 95% CI [1.07, 1.13]); Spearman ρ = 0.61, p = .009
• Success rose for every group (White 69→84%, Latinx 44→61%, Black 27→62%, Asian 61→73%)
• White and Latinx gap held at ~24 points; race×time interactions underpowered

Direct placement was associated with rising, not eroding, statistics success, contradicting the premise that removing remediation depresses outcomes. But broadened access alone did not close racial gaps.

A single-instructor design makes the temporal trend correlational: it cannot separate the reforms from the instructor's own growth over the same period. Reported as such throughout.

• Bloom's revised taxonomy; matched pre-AI vs. AI-assisted final exams, coded item by item
• Two courses, same instructor: intro statistics (Winter 2025 vs. 2026; 34 vs. 25 MC items) and college algebra (Spring 2025 vs. 2026)

• Statistics: higher-order (Analyse) items 18% → 48% (+30 pts); procedural (Apply) 47% → 28% (−19 pts)
• Student performance did not decline on the more demanding exam
• Three Universal Design for Learning features appeared only in the AI version
• Algebra: cognitive profile stable (85% Apply); AI showed in structure; mean final 61.8% → 78.7%

AI-assisted design works as an amplification mechanism: it lets instructors act on assessment-design intentions that severe time constraints had made impractical, not a replacement for instructor judgment.

The paired-Bloom's analysis of matched pre- and post-AI exams is replicable across courses, instructors, and institutions, a portable way to audit whether AI raises or lowers the demand of assessments.

• Official gradebook records, N = 106; four sections, two courses, same instructor
• Quantile decomposition (10th to 90th percentiles); Holm family-wise error control across ten comparisons
• Welch t & Mann-Whitney; Hedges g; percentile bootstrap (10,000); Lee-style trimming bound; attendance-adjusted ANCOVA

• Algebra final +16.9 pts (p = .004, Hedges g = 0.75), a floor effect
• 10th percentile 22 → 58 (+36); share below 50% 31% → 5%; 90th percentile essentially unchanged
• Statistics: performance maintained on a harder instrument
• Alignment decides value: AI-aligned quizzes stayed predictive of the final; item-bank quizzes lost it

Means hide the students open-access colleges most need to see. AI-assisted design lifted the lowest performers most, and whether formative work keeps its diagnostic value depends on how well it aligns with the redesigned exam.

Small cells; unproctored scores inflated toward the ceiling. Effects are reported distributionally and per course before pooling, a reporting discipline this literature still lacks.

• State-year panel: 50 states, 350 state-years, 3,783 incidents (Gun Violence Archive, 2018 to 2024)
• Negative-binomial regression with population offset, calendar-year fixed effects, CR2 cluster-robust SEs
• Standardized incidence-rate ratios per SD; LASSO + tree-ensemble triangulation

• Incidence roughly doubled at the 2021 peak (IRR 2.02, CI 1.68 to 2.43), elevated through 2024 (1.55)
• Poverty IRR 1.45 [1.18 to 1.79], p = .006, the most stable predictor; inequality 1.44 (r = 0.50 with poverty)
• Gun-law strength 1.11, ownership 1.07, serious mental illness 1.04, none associated
• LASSO retained only the socioeconomic-disadvantage measures

The post-2020 surge was broad and sustained, and its geography tracks socioeconomic disadvantage, chiefly poverty, rather than gun-law permissiveness or mental-illness prevalence.

Ecological and observational: no individual-level or causal inference, and the conclusions are sensitive to how “mass shooting” is defined (here, the GVA four-or-more definition).