iiinova — where your digital self keeps evolving

Andrew Wilson

@ajsw.info

730 followers 694 following 12 posts

environmental economist at the University of Virginia previously: Stanford Doerr/FSE, Columbia SDEV, Council of Economic Advisers, UC Berkeley, AmeriCorps ajsw.info 🏳️‍🌈

Posts Replies Media

Andrew Wilson @ajsw.info · 29/01/2026

But it's not just that. Due to the fixed, network nature of railways, efforts to manage weather risk actually result in a congestion externality—propagating delays—with most delay coming from decisions at nearby stations. Here, that's "spatial lag" > 0:

A two-panel chart plotting rail delays against daily average temperature (°C). The top panel shows "Cumulative minutes of delay" categorized by spatial lag (0=purple, 10=teal, 20=yellow). Delays spike significantly in extreme cold (<-20°C) for higher cumulative lags, with the 20-lag series reaching ~150 minutes and the 10-lag series ~100 minutes. Between 0°C and 30°C, delays are near zero. Extreme heat (>30°C) shows a minor increase (under 30 minutes). The bottom histogram displays temperature frequency, showing that most station-days occur between 5°C and 25°C, making extreme temperatures rare.

120

Andrew Wilson @ajsw.info · 29/01/2026

But there's good news: we also find that railroads are adapted to their local climate, might be learning from prior accidents, and change train speed and throughput (cancellations!) in response to weather. Sorry for the delay, it's because they care.

A coefficient plot with error bars showing the "Change in speed of Amtrak trains" (y-axis) across different bins of "Daily average temperature (°C)" (x-axis).

Extreme Cold: There is a significant reduction in train speed during extreme cold. At temperatures below -20°C, the speed change drops to approximately -2.2 miles per hour.

Moderate Temperatures: As temperatures rise from -20°C, the speed deficit gradually shrinks, reaching a baseline of 0 mph in the 15–20°C range.

Heat: In the 20–30°C range, there is again a decrease in speed (around -0.5 to -0.8 mph). However, for temperatures above 30°C, the data points show a positive change, indicating an increase in speed relative to the baseline.

Uncertainty: Error bars indicate the confidence intervals for each temperature bin, which are widest at the extremes (<-20°C and 30+°C).

110

Andrew Wilson @ajsw.info · 29/01/2026

Unsurprisingly, we find that exposure to extreme cold, heat, and rain increase accidents. But the magnitude surprised us: while rail is very safe overall, weather is responsible for 8–10% of accidents and related casualties—and railroads vastly underreport it as a cause.

A two-panel vertical chart sharing a horizontal axis labeled "Daily average temperature (°C)," divided into bins ranging from <-20°C to 30+°C.

Top Panel: A plot showing the "Percent change" in rail accidents compared to a baseline (10–15°C). Two data series are plotted with error bars:

Red (All rail accidents): Shows a sharp increase as temperatures drop, peaking at a 75% increase when temperatures are <-20°C. In hot weather (>30°C), the increase is smaller (below 10%).

Blue (Rail accidents with 1+ casualties): Rising to 65% at <-20°C. However, it shows a larger 25% increase at temperatures above 30°C.

Bottom Panel: A gray histogram showing the "Percent of county-days" for the corresponding temperature bins. The distribution indicates that extreme cold (<-10°C) and extreme heat (>30°C) are rare. Most county-days fall between 5°C and 25°C, a peak at 20–25°C.

100

Andrew Wilson @ajsw.info · 29/01/2026

Was your train delayed or cancelled this weekend? Mine was. Do you crave research that speaks to this particular experience? Look no further. In a new paper, @xinmingdu.bsky.social and I quantify how much rail safety and operations are affected by the weather.

A screenshot of the first page of the linked published article from the Journal of Public Economics.

282

Andrew Wilson @ajsw.info · 16/12/2025

For those who want a figure! "... before 2003, sequencing a hot day after another hot day can nearly double the mortality effect of that hot day ... the compounding model yields an [Aug. 2003] death toll that is closer to the total ... than the standard model ..."

071