I’m a PhD candidate in economics studying how place shapes economic outcomes — combining spatial
analysis, applied econometrics, and causal inference across regional development, public lands, and
labor markets.
Booked, Built, Mobilized: The Three Geographies of U.S. Defense Industrial Production
Tandem J. Young
June 2026 · Dissertation chapter Working Draft
Reframes defense industrial-base assessment around a distinction the field routinely collapses: a weapons
platform has three geographies — where its contracts are booked, where it is physically produced, and where
latent capacity could be mobilized. Conventional metrics weight counties by prime-contract share, routing a
program’s entire recorded value through its final-assembly site: Tarrant County, TX — home to Lockheed’s F-35
plant — records the largest single share of national defense obligations but under one percent of national
defense employment. Re-weighting exposure by production reverses the verdict. Mapped onto an input-output
county network, the F-35 cluster spans 56 counties across 25 states and behaves as roughly sixteen effective
counties; across 100 community-detection seeds no single county’s removal fragments the parts network (median
cohesion ≈ 85%), and about 95% of its import content has a domestic substitute. Of the ten largest defense
clusters, six are genuinely single-anchor fragile — the production-dispersed F-35 cluster a notable exception.
Four public-data instruments — a Leontief closed-loop vulnerability index, co-location spatial weights, a
cascade simulation, and a Hidalgo–Hausmann surge-capacity index — operationalize the framework. The analysis
is descriptive economic geography, not a causal-effects study.
JEL Codes: H56, R12, L64, F52
Wildlife Management Areas, Deer Quality, and Rural Land Values: A Spatial Hedonic Analysis of Arkansas
Tandem J. Young
May 2026 · Dissertation chapter Working Draft
Builds and validates the data and inference infrastructure for a spatial hedonic study of how
public-hunting-land proximity, deer-harvest quality, and Chronic Wasting Disease capitalize into rural
Arkansas land values; the contribution is methodological — the pipeline is validated, not the amenity
hypothesis, which awaits market transaction data. Lacking publicly released parcel-level sales, it is built
on 308,226 tax-assessed vacant-agricultural parcels across 73 counties and 20 deer-management zones, matched
to fourteen spatial data sources. The “WMA within ¼ mile” coefficient is −0.082 (95% CI [−0.130, −0.034]),
stable across ten identification-robustness checks (leave-one-out county median −0.082; 30% smaller, to
−0.057, under coarsened exact matching), and a placebo test relocating 191 pseudo-WMAs places it in the left tail
(one-sided p = 0.01). The negative gradient — opposite the positive-amenity prediction for sale prices —
reflects WMA placement on marginal terrain and is a diagnostic property of use-value assessment, not evidence
that WMAs depress market land values. The model also introduces an age-normalized Boone & Crockett
deer-quality index from administrative harvest records, with a staggered difference-in-differences design
around Arkansas’s phased CWD-zone expansion (2016, 2018, 2021) as the leading follow-on.
JEL Codes: C21, Q24, Q26, Q51, R14
Fueling Access: Gasoline Prices, Remoteness, and the Travel-Cost Elasticity of National-Park Demand
Tandem J. Young
June 2026 Working Paper
Measures how sensitive demand for public-land recreation is to the cost of getting there. In a monthly panel
of 371 contiguous-state national-park units spanning 1993–2025, regional gasoline prices are interacted with
each park’s remoteness from population. Because gasoline prices are procyclical, the level of the travel-cost
elasticity is not identified — but its remoteness gradient is, recovered from cross-region price variation
under a parallel-cyclical-response assumption that a non-fuel placebo supports. A one-standard-deviation
increase in remoteness makes a park’s gas-price elasticity roughly 0.07 log points more negative
(p ≈ 0.04 excluding the pandemic; p = 0.003 before it) — a point estimate stable across remoteness measures,
price geographies, and sub-periods, and one that reappears in pre-2010 data predating the 2014 and 2022 price
cycles. The adjustment is extensive (fewer trips, unchanged visit duration), concentrated at lower-volume
parks, with no detectable diversion to nearer parks. Applied to the fuel-price changes implied by carbon
pricing, the gradient implies a spatially regressive incidence: under the EPA social cost of carbon, parks at
the 90th percentile of remoteness lose roughly seven percentage points more access than those at the 10th — a
recreation-access co-cost concentrated in the rural West and rarely counted.
JEL Codes: Q26, Q51, Q54, R41, L83
Rising to Competition: Peer Effects and a Tripartite Decomposition in Horse Racing
Tandem J. Young
June 2026 Working Paper
Estimates whether the quality of one’s competitors causally raises or depresses individual performance — and
whether the agents who jointly produce it respond alike — using the complete 2023 universe of North American
thoroughbred racing (36,501 races; 261,496 entries), where post-entry scratches and claiming-price constraints
supply plausibly exogenous variation in field composition. The aggregate peer effect is positive and robust in
sign across designs — pooled OLS (0.148), horse fixed effects (0.193), triple fixed effects (0.184), scratch
and network instruments, and a claiming subsample — with a preferred within-horse magnitude of about 0.19,
roughly 1.4 lengths per standard-deviation shift in field quality, so horses “rise to competition,” reversing
the discouragement effect documented among professional golfers. The paper’s other main contribution is a
tripartite decomposition of this aggregate into the three agents who jointly produce performance: horses respond
positively (about +0.23) while jockeys and trainers respond negatively (about −0.10 and −0.07 jointly), so the
aggregate understates the horse’s own response. The sign pattern survives dynamic-panel and errors-in-variables
corrections, though the horse channel is suggestive rather than pinned. A pre-specified placebo does not pass;
the resulting 0.10–0.48 range is a robustness envelope across designs, not a single identified interval.
JEL Codes: D82, J24, L83, Z20