🔬 Researcher & Data Scientist — Health, Economics, and AI
🏛️ Predoctoral Researcher, Ellison Institute of Technology Oxford
📊 BSc Statistics, UBC

→ Live Dashboard — explore the gap between AI capability and observed AI use, occupation by occupation

💼 LinkedIn • ✉️ Email

I work at the intersection of health data science, empirical economics, and research engineering, designing end-to-end systems for credible analysis. I focus on problems where measurement is difficult, data is imperfect, and decisions depend on careful empirical structure.

My work starts with public, administrative, or unstructured data, moves through data linkage and dataset construction, and culminates in papers, dashboards, reproducible pipelines, or decision-ready outputs. Across projects, I prioritise measurement, transparency, scalability, and results that are easy to inspect.

• Health Data Science, Healthcare Productivity, Access and Inequality
• Ageing, Labour Markets, and Economic Outcomes
• AI Diffusion, Occupational Exposure, and Technology Adoption
• Bayesian Forecasting and Model Uncertainty
• NLP, Entity Resolution, and Document-Level Extraction
• Dashboards, Tableau, and Reproducible Analytical Products

Do More Appointments Lead to Shorter Waits and Better Patient Experience? Evidence from NHS Primary Care in England
Rohan Joseph, Harsh Gupta, Michael Keoghan, Shaun Danielli, and Andrew Scott

Abstract: Primary care productivity in England is evaluated through appointment volume adjusted by waiting times and patient satisfaction, yet the presumption that more output should improve these quality metrics has not been tested directly. Using national data across all 42 Integrated Care Boards (2018–2024), we find that output growth was not associated with shorter waiting times or lower patient dissatisfaction, and shorter waiting times were not associated with lower dissatisfaction, suggesting that either these metrics are driven by factors other than output or fail to capture the dimensions of quality that matter, and that current performance frameworks warrant review.

Are Longer Waits Associated with Lower Satisfaction and Worse Recovery? Evidence from Hip and Knee Replacement in England
Rohan Joseph, Harsh Gupta, Michael Keoghan, Shaun Danielli, and Andrew Scott

Abstract: Waiting times and patient dissatisfaction are widely used to assess healthcare quality, yet whether either reflects patient health, rather than only the experience of care, has rarely been tested directly. Linking respondent-level outcome measures to provider waiting times for NHS hip and knee replacement in England (2017–2023), we find that longer waits were associated with higher post-operative dissatisfaction and smaller functional improvements, and that worse satisfaction was associated with smaller improvements, with no equivalent relationship in general practice, suggesting that satisfaction in this focused pathway reflects patient-reported recovery as well as care experience, and that performance frameworks should account for the breadth of the care episode measured.

Intrinsic Capacity and the Economics of Healthy Ageing
Julian Ashwin, Yannis Kastis, Rohan Joseph and Andrew Scott

Abstract: Economic models of ageing typically measure health through self-assessed status or deficit-based frailty, yet whether retained functional capacity, rather than accumulated deficits, predicts later-life economic outcomes remains untested. Using nine waves of the English Longitudinal Study of Ageing (2002–2019), we find that higher intrinsic capacity is associated with greater employment, lower retirement and benefit receipt, lower mortality, and higher life satisfaction, both across and within individuals, and that these effects persist alongside a deficit-based frailty index, suggesting that intrinsic capacity and frailty are related but distinct measures of physiological ageing that jointly matter for economic outcomes and public finances.

📊 AI Occupation Adoption Gap Dashboard — live app : Interactive Streamlit and DuckDB dashboard for exploring the gap between theoretical AI capability and observed AI use across occupations. KPI cards, filters, occupation-level scatter plots, ranked opportunity roles, SQL validation.

🤖 AI Occupation Adoption Gap Analysis : Empirical analysis linking Anthropic observed-use data, OpenAI occupation capability scores, O*NET, and BLS controls to study where AI capability has not yet translated into workplace adoption. Feeds the dashboard above.

📉 Bayesian Unemployment Forecasting Analysis : Bayesian structural time-series modelling of U.S. unemployment, with rolling validation, latent trend specification, forecast accuracy, and interval calibration.

🛡️ CVE Vulnerability Exposure Pipeline : Integrated dataset combining NVD, CVE, EPSS, and KEV sources for vulnerability risk analysis.

Many projects rely on licensed, restricted, or locally derived datasets. Where raw data cannot be shared, I prioritise:

• Transparent pipeline structure
• Clear schema and transformation logic
• Reproducible workflows with validation checks
• Public-data examples where possible

The goal is to make methods and reasoning inspectable, even when data is not.