Research

Publications:

"Overreaction in Football Wagers" with Gary Smith, Big Data, 2018, 6(4): 262-270.

Abstract: Football scores are an imperfect measure of a team's ability, and consequently exaggerate differences in abilities. Those teams that perform the best and the worst are not really so far from average in their ability; thus their future performances regress to the mean. Betting data indicate that gamblers do not fully account for this regression.

Working Papers:

"No Space Like Home? Selection and Efficiency in Hospital-at-Home" (Job Market Paper).

Abstract: Hospital capacity is costly to expand, and the value of an available inpatient bed varies with congestion. Hospital-at-Home (HaH) provides inpatient-level care in a patient's residence and can release a brick-and-mortar bed without requiring permanent capacity expansion. However, this release is only realized when the hospital offers HaH and the patient accepts. I study how these decisions determine the use and value of flexible hospital capacity. Using clinical, accounting, census, and staffing data from four hospitals, supplemented by a novel patient survey, I jointly estimate patient acceptance and hospital offers, accounting for persistent patient preferences and endogenous offer selection. Patients differ systematically in their willingness to receive home care, and hospitals target likely adopters while balancing financial returns, clinical needs, and capacity relief. At median occupancy levels, the estimated capacity-relief value is equivalent to between $1,150 and $1,570 per discounted bed-day at the two largest hospitals. A counterfactual 10% increase in HaH reimbursement generates approximately 12,000 potential B&M additional bed-days released and raises aggregate expected patient option value by 91%. Although induced transfers raise predicted treatment costs by more than the additional reimbursement, reducing contribution from assessed patients by $4.1 million, filling 50–100% of released bed-days yields a net increase in combined hospital cash contribution of $3.7-$17.0 million. These findings show how consumer adoption shapes flexible capacity deployment and how the value of released capacity can make expansion privately attractive even when induced transfers raise treatment costs.

"Dynamic Treatment Choices and Selection into RCTs: Evidence from Prostate Cancer" with Barton Hamilton, Brian McManus, Juan Pantano, and Justin Trogdon.

Abstract: We study how voluntary participation and treatment protocols affect the interpretation of randomized controlled trials (RCTs) comparing existing medical treatments. When patients can obtain their preferred treatment outside the trial, those closer to indifference between treatment strategies are more likely to participate. When treatment is dynamic, protocols governing when patients may switch from their initially assigned treatment affect both enrollment and potential outcomes. We develop and estimate a dynamic model of trial participation and treatment choices using aggregate data on volunteers and eligible non-volunteers from a prostate cancer study. Selection operates primarily through patients' preferences over treatment side effects rather than differences in treatment effectiveness. Under the observed protocol, estimated volunteer and population treatment effects are similar for 10-year prostate cancer mortality but differ for metastasis and 15-year mortality. Alternative switching protocols change both participation and treatment effects.

"What do Boards Know? The Value of CEO Selection and Monitoring" with Barton Hamilton, Andrés Hincapié, and Noah Lyman, submitted.

Abstract: How much do boards know about their CEOs, and what is that information worth? We estimate a dynamic model of CEO succession in which boards screen internal and external candidates and monitor the incumbent's evolving quality. Boards' information is valuable but limited: screening resolves about sixty percent of the uncertainty about internal candidates but only six percent about external ones. Relative to uninformed hiring and firing, this information raises firm value by 10.6 and 7.7 percent, respectively. Full information would add another 23 percent. Insiders' performance advantage at appointment reflects what boards know, not better candidate pools.

"Patient Outcomes with Hospital-at-Home in Routine Care: Evidence from a Spatial Eligibility Boundary" with Bailey Beauclair, Escher Howard-William, Brian McManus, Carlton Moore, and Patrick O’Shea, submitted.

Abstract: Trial evidence on Hospital-at-Home (HaH) is generally favorable, but causal evaluation in routine care is difficult because treatment reflects clinical screening, operational constraints, geographic eligibility, and patient acceptance. We study patients with HaH-eligible conditions treated at four hospitals in a large southeastern U.S. health system from September 2021 through February 2026. Using a fuzzy spatial regression discontinuity design, we exploit the program's geographic eligibility boundary, which generated a modest but precisely estimated increase in HaH receipt. We estimate that geographic eligibility reduced mortality during the index episode but increased 30-day inpatient readmission and 90-day post-discharge mortality among patients near the eligibility boundary. Emergency department visits without inpatient admission declined modestly, indicating that the increase in downstream utilization was concentrated in inpatient readmissions. We found fewer imaging and consultation orders and suggestive evidence of fewer falls, but no clear effects on 30-day post-discharge mortality, discharge disposition, or length of stay. Estimates of the effect of HaH receipt were not statistically significant at either post-discharge mortality horizon. Complementary propensity-score matching and doubly robust analyses yielded different estimates even within the same local geographic samples. Among patients offered HaH, these methods produced estimates suggesting fewer readmissions, less institutional post-acute care, and fewer clinical orders, but longer stays and more falls. The contrast highlights the importance of treatment selection and of the distinct populations and identifying assumptions represented by observational and quasi-experimental estimands. Overall, HaH’s effects may differ across stages of care, underscoring the need to evaluate performance across both the index episode and subsequent follow-up rather than through the index stay alone.

Works in Progress:

"Estimating Heterogeneous Marginal Environmental Effects on Electrical Grids: Regularized Panel Data Regression Approach" with August Bruno, Valentin Verdier, and Andrew Yates.

Research Topics in Progress:

Hospital-at-home adoption and growth; organizational integration and agency frictions; health care price transparency; employer-sponsored health insurance bargaining.