Do Earlier Winter Weather Advisories Help Reduce Vehicle Crashes?

This study asks whether issuing winter weather advisories earlier helps reduce vehicle crashes. Using county-date data from 11 U.S. states between 2008 and 2019, the author links advisory lead times with crash rates, snow forecasts, weather observations, mobile phone location data, and snowplow activity. The analysis shows that an additional hour of lead time reduces daily crashes by 0.75%, with no diminishing returns at longer horizons. Cumulatively, longer lead times prevent about 13 crashes per 100,000 people annually and yield economic savings of $190 million per year. Mechanisms include fewer discretionary trips and more intensive road maintenance in response to earlier advisories.

Full Citation and Link to Article

Vaibhav Anand (forthcoming 2025). Does Getting Forecasts Earlier Matter? Evidence from Winter Advisories and Vehicle Crashes. American Economic Journal: Economic Policy. https://doi.org/10.1257/pol.20230247 

Extended Summary

Central Research Question

This study investigates whether earlier issuance of winter weather advisories improves road safety by reducing vehicle crashes. Specifically, it explores whether longer lead times—defined as the number of hours between the advisory’s issuance and the predicted onset of a weather event—enable individuals and institutions to better prepare for hazardous driving conditions. The author also examines whether the marginal benefits of lead time diminish as forecasts are issued further in advance and seeks to understand the mechanisms behind any observed effects. This question is especially relevant given the trade-off between forecast accuracy and early availability.

Previous Literature

The value of weather forecasts has been widely recognized in the context of risk mitigation, particularly in climate-sensitive sectors like agriculture, health, and infrastructure. Several studies highlight the benefits of more accurate forecasts (e.g., Shrader, Bakkensen, and Lemoine 2023; Molina and Rudik 2022), while others examine the role of forecast access in influencing decisions (e.g., Rosenzweig and Udry 2019; Downey et al. 2023). However, little empirical work has addressed the benefits of simply receiving forecasts earlier, irrespective of accuracy. Some theoretical models (e.g., Millner and Heyen 2021) suggest that the value of longer lead times may diminish when individuals can substitute between short- and long-term adaptations. Moreover, prior studies on public alerts and weather warnings (e.g., Simmons and Sutter 2008) show mixed results regarding whether earlier warnings reduce harm. This study contributes new evidence by using real-world data to quantify the effects of lead time on crash rates and to assess the behavioral responses that drive these effects.

Data

The study combines several large-scale datasets from 2008 to 2019, focusing on 11 U.S. states in the Midwest and Northeast. Key data sources include:

1. Vehicle crash reports from state Departments of Transportation (DoTs), offering daily county-level counts of all reported crashes, including fatal and non-fatal incidents.

2. Winter weather advisories issued by the National Weather Service (NWS), accessed via the Iowa Environmental Mesonet. These advisories include time of issuance, effective period, and predicted weather type.

3. Weather observations from NOAA’s Global Historical Climate Network (GHCN), providing county-level estimates of snowfall, rainfall, and temperature.

4. Forecast data from the National Digital Forecast Database (NDFD), used to construct lead time and forecast error variables.

5. Snowplow activity data from the Iowa Department of Transportation, offering daily measures of road treatment (e.g., salt application, snowplow movement).

6. Mobile phone location data from SafeGraph, used to measure changes in individual travel behavior.

The unit of analysis is the county-date. Multi-day advisories are aggregated into a single observation to prevent biased estimates from varying severity across days.

Methods

The primary empirical strategy is a fixed-effects regression that exploits within-county variation in advisory lead time. The model controls for observed and forecasted weather conditions, the nature of the advisory (e.g., blizzard, ice storm), the time of issuance, day-of-week effects, and county-by-year-by-month fixed effects to isolate the causal impact of lead time on crash rates.

The author bins lead time into hourly intervals and estimates both same-day and cumulative effects over a five-day window surrounding the advisory. A distributed lag model captures possible intertemporal shifts in crashes. The specification also includes forecast error terms to adjust for the inverse relationship between lead time and forecast accuracy.

Two mechanisms are studied: individual responses (reduced trips) and institutional responses (increased road maintenance). SafeGraph mobile data measure changes in discretionary and non-discretionary visits; Iowa DOT data track variations in plow activity by lead time.

Robustness checks include alternate lead time definitions, exclusion of multi-day advisories, placebo tests using fatal crash data, and interaction terms to assess heterogeneity in effect.

Findings/Size Effects

The study finds robust evidence that longer advisory lead times significantly reduce crash rates and generate sizable economic benefits.

1. Crash Reduction

   
   

   • Each additional hour of lead time reduces daily crashes by about 0.75%.

   • The cumulative benefit over a five-day period remains strong, indicating that early advisories lower total crash risk rather than merely shifting crashes across days.

   • The crash reduction is more pronounced at longer horizons. For lead times over 48 hours, each additional hour prevents 0.25 crashes per 100,000 people—more than triple the short-horizon effect.

   
   

2. Non-Diminishing Marginal Returns

   
   

   • The marginal benefit of longer lead time does not diminish with horizon, contradicting theoretical expectations.

   • On holidays and weekends, the benefits of longer lead times are even greater, suggesting higher costs or complexities in adjusting travel plans on such days.

   
   

3. Forecast Accuracy vs. Lead Time

   
   

   • Longer lead times are associated with higher forecast error.

   • Nonetheless, lead time is more predictive of crash reduction than forecast accuracy.

   • A “cry wolf” effect is observed: advisories following recent false alarms (where snow fails to materialize) show diminished effectiveness.

   
   

4. Economic Value

   
   

   • The estimated reduction of 13 crashes per 100,000 people annually translates into $190 million in avoided crash costs across the 11 states.

   • This figure represents roughly 19% of the NWS annual budget and 4% of the federal investment in meteorological services.

   
   

5. Mechanisms

   
   

   • Road treatment: Longer lead times increase snowplow activity, especially pre-treatment (e.g., anti-icing) efforts. These responses scale with lead time, especially after 24 hours.

   • Reduced travel: People make fewer visits when they receive earlier advisories. The drop is larger for discretionary visits (e.g., restaurants, shopping) than for non-discretionary ones (e.g., work, school).

   • The travel reduction is modest but meaningful—around 0.2% per additional hour of lead time.

   • No consistent evidence was found that treated roads increased visits. In fact, places closer to treated roads sometimes showed larger declines in visits, possibly due to their higher sensitivity to weather.

     
     

Conclusion

This study offers the first empirical evidence that earlier issuance of winter weather advisories leads to meaningful reductions in vehicle crashes. The benefits are robust across specifications and persist even when forecast accuracy is lower. Moreover, the marginal value of lead time does not diminish as horizon increases—longer warnings appear to enable both individual and institutional adaptations that short-horizon advisories cannot.

These findings carry significant policy implications. They support further investment in forecasting systems that can issue timely advisories, even at the expense of some accuracy. Agencies like the National Weather Service can use this evidence to justify infrastructure upgrades or expanded public outreach. Additionally, the results highlight the importance of ensuring that people and institutions can receive and respond to advisories—suggesting a potential role for improved communication systems and user training.

The author cautions, however, that the study focuses only on the benefits of longer lead times. It does not estimate the full cost of adaptation behaviors, such as lost productivity or increased salt use. Future work could examine these tradeoffs, explore variations across regions or socioeconomic groups, and assess how advisory systems might be optimized for even greater impact.

Overall, the study underscores that even incremental improvements in the timing of public weather alerts can save lives, reduce injuries, and deliver high returns on public investment.