About

Nathan Chan is a professor at the Department of Resource Economics at the University of Massachusetts Amherst, affiliated with the Computational Social Science Institute. His research investigates key challenges in environmental management and energy policy, with a primary focus on environmental public goods. He studies a wide variety of topics including global climate change mitigation, energy efficiency policy, markets for environmentally-friendly products, nonmarket valuation, and pollution control. He applies methods from theoretical, empirical, and experimental economics in his work, contributing to the understanding and development of policies aimed at addressing environmental and energy issues. His research aims to inform effective strategies for environmental sustainability and energy efficiency, leveraging diverse economic approaches to tackle complex environmental challenges.

Research topics

• Economics
• Microeconomics
• Public economics
• Business
• Computer science

Selected publications

• Legalization and innovation in the cannabis market

  International Journal of Industrial Organization · 2026-03-03 · 1 citations

  articleOpen accessSenior author

  How would legal institutions affect the rate and direction of innovation in restricted or illicit markets? This paper studies the impact of legalization on innovation in the cannabis market. We construct novel data and measures on cannabis-related innovation in clinical trials and patent applications at the state-year level. We use staggered difference-in-differences models and event studies to analyze the impacts of medical and recreational cannabis legalization (MCL and RCL, respectively) on innovation. We find no evidence that cannabis-related clinical trials respond to MCLs or RCLs. However, there is a marked increase in cannabis-related patenting ( ∼ 4.4 patents) in response to an RCL and more modest impacts from an MCL ( ∼ 1.4 patents). Notably, additional RCL-induced patents are focused primarily on downstream products and methods rather than upstream medical innovation; MCL impacts show a similar pattern. Our further analysis of patent content suggests significant growth in areas that may raise public health, safety, and misuse concerns. These results suggest that legalization increases innovation in the cannabis market, but with relatively weak gains in areas pertinent to health and safe use. We corroborate our findings with multiple new empirical methods in the literature.

  PublisherDOI

• Core Properties of Cost Sharing Approaches to Local Public Good Provision

  SSRN Electronic Journal · 2026-01-01

  preprintOpen access1st authorCorresponding
  PublisherDOI

• Accuracy of large language models in generating differential diagnosis from clinical presentation and imaging findings in pediatric cases

  Pediatric Radiology · 2025-07-12 · 1 citations

  articleOpen access

  BACKGROUND: Large language models (LLM) have shown promise in assisting medical decision-making. However, there is limited literature exploring the diagnostic accuracy of LLMs in generating differential diagnoses from text-based image descriptions and clinical presentations in pediatric radiology. OBJECTIVE: To examine the performance of multiple proprietary LLMs in producing accurate differential diagnoses for text-based pediatric radiological cases without imaging. MATERIALS AND METHODS: One hundred sixty-four cases were retrospectively selected from a pediatric radiology textbook and converted into two formats: (1) image description only, and (2) image description with clinical presentation. The ChatGPT-4 V, Claude 3.5 Sonnet, and Gemini 1.5 Pro algorithms were given these inputs and tasked with providing a top 1 diagnosis and a top 3 differential diagnoses. Accuracy of responses was assessed by comparison with the original literature. Top 1 accuracy was defined as whether the top 1 diagnosis matched the textbook, and top 3 differential accuracy was defined as the number of diagnoses in the model-generated top 3 differential that matched any of the top 3 diagnoses in the textbook. McNemar's test, Cochran's Q test, Friedman test, and Wilcoxon signed-rank test were used to compare algorithms and assess the impact of added clinical information, respectively. RESULTS: There was no significant difference in top 1 accuracy between ChatGPT-4 V, Claude 3.5 Sonnet, and Gemini 1.5 Pro when only image descriptions were provided (56.1% [95% CI 48.4-63.5], 64.6% [95% CI 57.1-71.5], 61.6% [95% CI 54.0-68.7]; P = 0.11). Adding clinical presentation to image description significantly improved top 1 accuracy for ChatGPT-4 V (64.0% [95% CI 56.4-71.0], P = 0.02) and Claude 3.5 Sonnet (80.5% [95% CI 73.8-85.8], P < 0.001). For image description and clinical presentation cases, Claude 3.5 Sonnet significantly outperformed both ChatGPT-4 V and Gemini 1.5 Pro (P < 0.001). For top 3 differential accuracy, no significant differences were observed between ChatGPT-4 V, Claude 3.5 Sonnet, and Gemini 1.5 Pro, regardless of whether the cases included only image descriptions (1.29 [95% CI 1.16-1.41], 1.35 [95% CI 1.23-1.48], 1.37 [95% CI 1.25-1.49]; P = 0.60) or both image descriptions and clinical presentations (1.33 [95% CI 1.20-1.45], 1.52 [95% CI 1.41-1.64], 1.48 [95% 1.36-1.59]; P = 0.72). Only Claude 3.5 Sonnet performed significantly better when clinical presentation was added (P < 0.001). CONCLUSION: Commercial LLMs performed similarly on pediatric radiology cases in providing top 1 accuracy and top 3 differential accuracy when only a text-based image description was used. Adding clinical presentation significantly improved top 1 accuracy for ChatGPT-4 V and Claude 3.5 Sonnet, with Claude showing the largest improvement. Claude 3.5 Sonnet outperformed both ChatGPT-4 V and Gemini 1.5 Pro in top 1 accuracy when both image and clinical data were provided. No significant differences were found in top 3 differential accuracy across models in any condition.

  PublisherOA PDFDOI

• CvdN Equilibrium in Public Good Economies:An Investigation of Existence and the Nature of Equilibrium

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access1st authorCorresponding
  PublisherDOI

• Income targeting in consumer energy efficiency programs

  Energy Economics · 2025-02-03

  article1st author
  PublisherDOI

• CvdN Equilibrium and Share Equilibrium in Local Public Good Economies

  Journal of Public Economic Theory · 2025-04-01 · 1 citations

  article1st authorCorresponding

  ABSTRACT CvdN equilibrium and share equilibrium are both extensions of the cost‐shares‐based ratio equilibrium for global public good economies to local public good economies. CvdN equilibrium and share equilibrium differ in terms of the stability requirements for equilibrium jurisdictions. While share equilibrium keeps agents' relative cost shares fixed across jurisdictions and allows each agent to consider unilateral moves to alternative jurisdictions, CvdN equilibrium employs the use of share functions that allow for equilibrium adjustments of relative shares in each jurisdiction and requires the agreement of all agents in alternative jurisdictions. Despite these differences, we demonstrate that CvdN equilibrium extends share equilibrium: every arrangement of an economy that is supported in share equilibrium is also a CvdN equilibrium. However, the reverse is not true and CvdN equilibrium may exist when share equilibrium does not. Thus, CvdN equilibrium provides predictions in more economies and does not contradict share equilibrium when it exists.

  PublisherDOI

• Abstract 4136636: Analysis of Racial Disparities in the Training Path to Cardiology: Trends from 2012 to 2022

  Circulation · 2024-11-12

  article

  Introduction: Substantial evidence indicates that racial and ethnic diversity among the physician workforce bridges cultural gaps and improves patient care. Trends in racial diversity across the full training pipeline from undergraduate to practicing cardiologist are not yet well-characterized. Aims: We analyze racial disparities at each stage of training towards becoming a practicing cardiologist, and identify changes in these disparities between 2012 and 2022. Methods: We conducted a retrospective analysis of data specifying the racial composition of medical trainees acquired from the Association of American Medical Colleges, Accreditation Council for Graduate Medical Education, and Electronic Residency Application Service. Population data was acquired from the US Census Bureau. We computed the representation quotient (RQ) for each racial group at each stage to compare representation among the trainees or physicians to age-matched segments of the US population. Results: RQ decreases from undergraduate to active cardiologist for African Americans (0.719 to 0.314), Hispanics (0.660 to 0.359), American Indians (0.551 to 0.221), and Native Hawaiians (1.121 to 0.512). The lowest RQs of 0.133 and 0.137 were observed for American Indian active internal medicine residents and cardiology fellows, respectively. Racial disparities in medical education declined between 2012 and 2022, with RQs increasing from 0.541 to 0.735 for African American medical school matriculants and from 0.444 to 0.578 for Hispanic medical school matriculants. Similar increases were seen in the active cardiologist population for African Americans (2013 RQ: 0.227, 2022 RQ: 0.342) and Hispanics (2013 RQ: 0.283, 2022 RQ: 0.375). Conclusion: While racial diversity in the cardiology training pipeline has increased over the last decade, significant disparities persist for under-represented minorities. Addressing these barriers will help achieve a cardiology workforce reflective of the diverse population it serves.

  PublisherDOI

• Behavioral science initiatives to reduce individuals' climate footprint

  AEA Randomized Controlled Trials · 2024-05-13

  dataset1st authorCorresponding
  PublisherDOI

• PrP turnover in vivo and the time to effect of prion disease therapeutics

  bioRxiv (Cold Spring Harbor Laboratory) · 2024-11-14 · 4 citations

  preprintOpen access

  Abstract PrP lowering is effective against prion disease in animal models and is being tested clinically. Therapies in the current pipeline lower PrP production, leaving pre-existing PrP to be cleared according to its own half-life. We hypothesized that PrP’s half-life may be a rate-limiting factor for the time to effect of PrP-lowering drugs, and one reason why late treatment of prion-infected mice is not as effective as early treatment. Using isotopically labeled diet with targeted mass spectrometry, as well as antisense oligonucleotide treatment followed by timed PrP measurement, we estimate a half-life of 5-6 days for PrP in the brain. PrP turnover is not affected by over-or under-expression. Mouse PrP and human PrP have similar turnover rates measured in wild-type or humanized knock-in mice. CSF PrP appears to mirror brain PrP in real time in rats. PrP in the colon is readily quantifiable and has a half-life just slightly shorter than in brain. An under-expressed pathogenic mutant PrP, corresponding to D178N in humans, exhibits an accelerated turnover rate. Our data may inform the design of both preclinical and clinical studies of PrP-lowering drugs. Author Summary Prion disease is a fatal brain disease caused by misfolding of the prion protein (PrP). Emerging therapies for prion disease seek to reduce the amount of PrP produced in the brain in order to delay onset of disease or slow progression. Mouse studies have shown that if these therapies are initiated too late, their benefit is limited or they may not help at all. Here we measure the half-life of PrP in the mouse brain, and find that it is about 5 days. When drugs are used to lower PrP by cutting up the RNA that encodes PrP, the RNA drops rapidly while the protein lags behind, and does not reach its minimum level until 4 weeks after the drug is dosed. This half-life is about the same regardless of the species of PrP (mouse or human) and whether or not the brain is infected with prions. Cerebrospinal fluid appears to reflect the real-time levels of brain PrP with no appreciable lag. PrP can be measured in colon, which may be useful in animal studies of systemic drugs to lower PrP. PrP turns over more quickly in the presence of a pathogenic genetic variant, the equivalent of the human D178N variant. These findings suggest that clinical trials can monitor PrP in cerebrospinal fluid to look at drug activity, but should plan timepoints far enough post-dose to account for PrP’s rate of turnover, and should focus on patients who will survive long enough to benefit from the drug.

  PublisherOA PDFDOI

• Measuring strength of altruistic motives

  Journal of the Economic Science Association · 2024-04-26 · 2 citations

  articleOpen access1st author

  Abstract We introduce a novel way to elicit individuals’ strength of altruistic motivation in the context of charitable donations, ranging from pure warm glow to pure altruism. Using the giving-type elicitation task of Gangadharan et al. (2018) and assuming that individuals maximise a Cobb–Douglas impure altruism utility function, as is used in Ottoni-Wilhelm et al. (2017), we can uniquely identify the strength of altruistic motivation for impure altruists, which is typically found to be the largest category of donors. We compare the introduced measure to an alternative survey-based elicitation from Carpenter (2021).

  PublisherOA PDFDOI

Frequent coauthors

• Dieter Stiers

  KU Leuven

  13 shared

• Gunnar Otte

  10 shared

• Amélie Godefroidt

  KU Leuven

  10 shared

• Jan Paul Heisig

  WZB Berlin Social Science Center

  10 shared

• Andrea Förster

  Bradford Royal Infirmary

  10 shared

• Nils D. Steiner

  Johannes Gutenberg University Mainz

  10 shared

• Joan E. Madia

  University of Oxford

  10 shared

• Katharina Burgdorf

  University of Bremen

  10 shared

Education

• Ph.D.

  Yale University

  2014

• M.P.A.

  Columbia University

  2009

• B.S.

  California Institute of Technology

  2008