- The paper demonstrates that heat mitigation policies at origins significantly increase out-migration, while adaptation policies at destinations attract inflows.
- It utilizes a DistilBERT-based NLP pipeline and fixed-effects gravity models to robustly estimate heterogeneous policy effects across 4,713 U.S. counties.
- The findings highlight that socioeconomic factors like income, aging, and diversity mediate migration responses, urging context-sensitive policy design.
Contrasting Effects of Heat Mitigation and Adaptation Policies on U.S. Internal Migration
Introduction
This paper systematically investigates the causal impacts of heat-related climate policies—specifically, adaptation policies (APs) and mitigation policies (MPs)—on internal migration patterns across U.S. counties. Leveraging machine learning to classify 4,713 policy actions reported between 2017 and 2020 and granular IRS migration flows, the study deploys gravity-type fixed-effects models for robust identification of causal effects, while explicitly accounting for key socioeconomic moderators.
Methodological Framework
The analysis aligns heat-related policies reported in the CDP dataset with IRS county-to-county migration data. Policies are classified into technological/infrastructural, institutional, behavioral/cultural, and nature-based categories using a DistilBERT-based NLP pipeline, achieving high classification accuracy (F1 score = 0.9274). Migration flows are modeled using fixed-effects gravity specifications allowing for control of time-invariant county pair characteristics and annual shocks. Socio-demographic moderators—including median income, education, racial diversity, and aging rates—are incorporated using linear and nonlinear interaction terms to assess heterogeneous treatment effects.
Empirical Results
Aggregate Policy Effects
- Heat Mitigation Policies: MPs implemented at origin counties exhibit a statistically significant, positive effect on out-migration rates. Each additional origin-based MP increases the annual out-migration rate by 0.21% (90% CI, P < 0.001). MPs at destinations are associated with a 0.15% reduction in outflow from origins to those destinations.
- Heat Adaptation Policies: APs at origins do not produce statistically significant changes in out-migration. However, APs at destinations yield a 0.20% increase in migration inflow from the origin to the destination county (P = 0.04).
- Temporal Asymmetry: The effects of MPs are essentially contemporaneous, whereas destination-based APs show lagged and reinforcing effects—becoming more influential with one- and two-year lags.
Policy-Type Heterogeneity
Effects vary substantively across policy types and spatial context:
- Institutional and Behavioral/Cultural MPs at Origins: Each additional policy of these types raises out-migration by 0.37% and 0.46%, respectively.
- Destination-Based Policies: Behavioral/cultural APs at destinations increase migration inflows, while technological/infrastructural MPs at destinations initially reduce out-migration but, with lag, induce an increase—reversing sign and effect magnitude over time.
- Nature-Based MPs: At destinations, these exhibit a substantial positive association with in-migration (0.82% per policy, P = 0.04), though at origins the effect is negligible.
Socioeconomic Moderation and Nonlinear Dynamics
Socio-demographic context exerts pronounced, nonlinear moderation:
- Income: Institutional MPs at origins have U-shaped effects, suppressing migration at income extremes but increasing it in mid-range contexts; behavioral/cultural MPs show decreasing effectiveness with rising income.
- Aging Rate: Institutional MPs display an inverted U-shaped moderating effect; origin counties with moderate aging rates experience peak migration response, while both low and high aging rates dampen this effect.
- Racial Diversity: MPs’ marginal effects on migration become positive and significant only at higher levels of diversity, notably for behavioral/cultural and institutional types.
- Educational Attainment: Institutional MPs trigger a U-shaped migration response, while behavioral/cultural MPs are mostly insensitive to education levels.
Discussion and Implications
These findings demonstrate that heat mitigation and adaptation policies exert contrasting and spatially asymmetric effects on U.S. internal migration, directly challenging assumptions prevalent in simulation-based migration forecasting. In particular, adaptation policies at migration destinations can function as attractors, while mitigation policies—especially those with regulatory, behavioral, or cultural components—may trigger contemporaneous out-migration at the origin.
Heterogeneity by policy type is substantial and underpinned by economic and social costs associated with implementation. For example, institutional and behavioral/cultural MPs often have direct economic impacts, aligning with prior literature on regulatory cost pass-through and sectoral structural change, supporting the mechanism through which these policies act as migration "push" factors.
Socio-demographic factors condition the realized effect sizes and directions, indicating that universal, untargeted climate policy interventions may inadvertently magnify regional disparities or trigger unintended migration responses—particularly in socioeconomically polarized or demographically aging contexts.
Theoretical and Practical Implications
Theoretically, the results underscore the necessity of accounting for policy-induced economic adjustment and social heterogeneity when modeling climate-migration feedbacks. Practically, the findings recommend that policymakers incorporate migration effects into the design and communication of both mitigation and adaptation interventions. Moreover, given the lagged effects observed for adaptation policies at destinations, there is potential for anticipatory planning to accommodate probable migration inflows following such interventions.
Empirical strategies that harness administrative migration data, policy databases, and advanced text classification set a methodological precedent for more accurate causal assessment in the climate-migration-policy nexus.
Future Directions
Key limitations include potential selection bias in the policy action dataset and the inability to decompose direct and indirect pathways of impact. Incorporating individual- and firm-level data, as well as richer mediational designs, would enhance mechanistic understanding. Expansion to other climate hazards and international contexts is warranted, as is deeper examination of long-term demographic consequences of sustained policy interventions.
Conclusion
This study provides robust causal evidence that climate-related heat mitigation and adaptation policies exert divergent effects on internal migration in the U.S., with substantial heterogeneity driven by policy type and local socio-demographic context. These findings invite re-evaluation of prevalent assumptions in climate adaptation planning and migration projection, highlighting the imperative for context-sensitive, evidence-backed policy design.
