Do Amazon Distribution Facilities Boost Local Economies?

Pathania and Netessine (2026) ask whether opening Amazon distribution facilities increases local economic well-being. They examine county-level data from 2011–2019, including employment-to-population ratios, poverty rates, and median household income, and they use midsized counties as treatment cases. They combine matching with Callaway–Sant’Anna difference-in-differences to address selection and staggered treatment. They find that after Amazon enters a county, the employment-to-population ratio rises by about 0.0087 (about +1.46%), the poverty rate falls by roughly 0.36 percentage points (−2.69%), and median household income increases by about $1,413 (+2.33%). They argue these effects are likely causal, not just correlations driven by faster-growing counties.

Why This Article Was Selected for The Policy Scientist

This article addresses an important and timely policy question: whether large e-commerce facilities materially change local economic conditions. The stakes are broad because many regions compete for distribution centers through sizable tax incentives, yet clear evidence on economic returns has remained limited. This study contributes meaningfully to that gap by providing credible estimates at a moment when warehouse growth and labor market restructuring are accelerating. The authors are established in this research area and employ a county-level panel with credible public data and proprietary headcount information; the data are high quality. Their matched Callaway–Sant’Anna design enhances causal interpretation and supports cautious generalizability to other midsized U.S. counties.

Full Citation and Link to Article

Pathania, V., & Netessine, S. (2026). The impact of Amazon facilities on local economies. Journal of Policy Analysis and Management, 45(1). https://doi.org/10.1002/pam.70065

Central Research Question

The authors ask whether the opening of Amazon distribution facilities produces measurable changes in host counties’ economic outcomes. More specifically, they investigate whether employment-to-population ratios increase, poverty rates decline, and median household incomes rise after Amazon enters a county. Embedded in this question is whether any observed changes can plausibly be interpreted as causal rather than artifacts of Amazon’s tendency to locate in already growing areas. The study therefore treats Amazon facility openings as a quasi-experimental shock to local labor markets and assesses the magnitude, timing, and durability of the resulting economic effects over the first three years following entry.

Previous Literature

The broader research program examines place-based development policies and the economic consequences of large facility openings. Several adjacent strands exist. One examines manufacturing plants and other tradable-sector establishments, with evidence generally suggesting positive multipliers via direct and indirect job creation. Another evaluates large retail entrants such as Walmart, where findings are more mixed: modest net job gains accompanied by price reductions and displacement among smaller competitors. More recent work analyzes e-commerce fulfillment networks as part of the restructuring of retail supply chains and employment. Amazon occupies an intermediate position in this literature because it is both a retailer and a logistics operator, creating tradable-sector characteristics (shipping goods nationwide) without the classical structure of a manufacturing plant. Prior empirical studies on Amazon have been sparse and sometimes contradictory, with think-tank reports yielding divergent conclusions about net employment effects. Methodologically, past research has faced challenges of endogeneity, staggered treatment, and limited visibility into facility-level employment. This study situates itself within that space, extending earlier work by addressing site-selection bias and by providing explicit treatment timing in a multi-period context.

Data

The empirical analysis combines public and proprietary county-level data from 2011–2019. Economic outcomes include employment-to-population ratios from the Bureau of Labor Statistics, poverty rates and median household incomes from Census Small Area Income and Poverty Estimates, and population as well as demographic composition from the American Community Survey. Facility headcount data are derived from Amazon internal records, which provide monthly employment counts for each distribution facility. These headcount data allow the authors to infer start dates and measure the scale and ramp-up trajectory of facility staffing. The sample is restricted to midsized counties (2013 populations between 50,000 and 1,000,000) because small counties rarely host Amazon facilities and large counties are nearly always treated by the mid-2010s, leaving no credible untreated comparisons. Counties adjacent to treated counties are excluded to reduce contamination via spillovers. The authors also employ data on county-level rural–urban classification from the USDA Economic Research Service to model heterogeneous time trends. The resulting dataset is a balanced panel that supports pre/post comparisons while preserving temporal variation in treatment onset, with a minimum of three post-entry years for each treated county due to truncation at 2019 to avoid pandemic-era distortions.

Methods

The empirical identification strategy has two components. First, the authors form a matched panel to mitigate selection bias. They match each treated county to a control county based on 2013 values of population, median household income, share of population aged 19–64, and recent population growth. Matching uses nearest-neighbor selection with inverse-variance weighting of covariates in the distance metric. Second, they implement a difference-in-differences design that accommodates staggered treatment timing. They estimate two models: (1) traditional two-way fixed effects (TWFE) with county and year fixed effects plus county-type linear trends and (2) the Callaway–Sant’Anna (CSDID) estimator, which isolates comparisons between treated cohorts and never-treated units while avoiding the negative weighting and contamination issues identified in the recent staggered-treatment econometrics literature. Event-study specifications assess parallel trends in the pre-period and trace dynamic treatment effects in the post-period. Standard errors are clustered at the county level. The authors justify causal language on the basis of selection correction, credible pretrend diagnostics, and placebo tests across employment sectors where no effects are theoretically expected.

Findings/Size Effects

Results indicate economically and statistically meaningful improvements in treated counties over the first three years after Amazon entry. In the preferred CSDID matched-panel specification, the employment-to-population ratio increases by 0.0087, equivalent to a 1.46 percent gain relative to the mean. Poverty rates fall by approximately 0.36 percentage points, representing a 2.69 percent reduction. Median household income rises by an estimated $1,413, a 2.33 percent increase. Event-study plots show muted or statistically insignificant pretrends for matched comparisons, supporting the identification assumptions. Dynamic estimates suggest that impacts materialize gradually, consistent with facility ramp-up patterns, which show county-level headcount surpassing 3,000 by year three. The employment gains in trade/transport/warehousing (about +5 percent) align with expected direct job creation, and placebo tests reveal no effects in government or manufacturing sectors. Retail employment shows no statistically significant decline, in contrast to some prior studies that found displacement effects around e-commerce warehouses. The authors also translate estimated employment effects into implied multipliers: using mid-sample population characteristics, they approximate ~3,900 net new employed individuals against ~2,000 direct Amazon jobs, suggesting a multiplier near two, consistent with earlier research on tradable-sector employment spillovers. Effects are averaged over multiple facility sizes and county contexts, so treatment heterogeneity likely exists, though the primary estimates focus on aggregated mean effects rather than distributional variation across counties or facility types.

Conclusion

The study provides credible evidence that Amazon distribution facilities produce positive local economic effects in the short to medium term, increasing employment, reducing poverty, and raising median income. The identification strategy is thoughtful and aligned with modern causal inference standards for observational policy research. The matched sample improves internal validity by balancing treated and control counties on pretreatment characteristics, while the Callaway–Sant’Anna estimator avoids the well-documented biases associated with TWFE under staggered adoption. The dataset is robust in both coverage and granularity, combining public economic statistics with proprietary headcount data that strengthen treatment timing and mechanism assessment. The authors are appropriately cautious in drawing causal inferences but support their claims with diagnostic tests, dynamic estimates, and sector-specific placebo checks. Generalizability likely applies to other midsized U.S. counties with similar labor markets and infrastructure, though the analysis does not address international contexts or very small rural counties. The findings matter for ongoing debates about tax incentives, local development strategies, and the future geography of e-commerce logistics. They also refine an emerging literature that sits between research on traditional manufacturing multipliers and research on retail displacement, showing that fulfillment centers occupy a distinctive role in local labor markets. While the study does not evaluate long-run effects beyond three years or distributional wage impacts within counties, it establishes a strong empirical baseline for subsequent research on spatial spillovers, land use, and workforce dynamics. Overall, it represents a substantive contribution to the policy evaluation literature at a time when warehouse expansion and regional incentive competition are accelerating.