The core of the problem lies in how temperature is actually measured at the locations where people live and work. Physical weather stations are typically spaced miles apart, which means that for most cities, neighborhoods, and farm fields, no station directly records the local temperature. Researchers therefore rely on statistical estimates, often called proxies, that interpolate or model temperatures between stations or draw from gridded climate products. These proxies are the best available guesses, but the new study demonstrates that they carry measurement errors far larger than the assumptions embedded in standard econometric methods. Traditional statistical frameworks generally treat such error as negligible or as classical measurement error that merely attenuates estimated effects. The reality, Yu and Carson show, is more troubling: the errors persist over time, correlate with terrain and monitoring density, and distort the temperature signal in ways that conventional corrections cannot fix.

The magnitude of these errors is striking, and it varies enormously with geography. In mountainous regions such as Boulder, Colorado, where temperature can change dramatically over short distances due to elevation, slope, and aspect, standard weather proxies differed from actual temperatures by an average of nearly 15 degrees Fahrenheit. In flatter, more densely monitored areas such as Chicago, the corresponding errors were as low as 1 degree Fahrenheit. This spatial heterogeneity matters because climate-impact studies frequently pool data across many locations. If the proxy error is systematically larger in some places than others, and if those places differ in the outcomes being studied, the resulting estimates inherit a geographic bias that no amount of additional statistical sophistication within the regression can remove. The error is baked into the input data itself.

To put the scale of the problem in perspective, the study reports that for one commonly used weather proxy, temperature measurement errors averaged 3.3 degrees Fahrenheit, or 1.85 degrees Celsius. That figure is larger than the entire warming limit enshrined in the Paris Climate Agreement, which commits signatories to holding global temperature rise well below 2 degrees Celsius and pursuing efforts to limit it to 1.5 degrees Celsius. In other words, the uncertainty introduced by mismeasured temperature inputs in a single class of research data exceeds the full amount of warming that international policy seeks to prevent. For a field in which effect sizes are often estimated from differences of a degree or two between hot days and moderate days, an error of this magnitude is not a rounding issue. It is a first-order problem that can reshape the conclusions of an entire literature.

The empirical demonstration at the heart of the paper uses one of the most policy-relevant relationships in climate economics: the link between extreme heat and violent crime. The researchers assembled more than 20 million crime records and first estimated the heat-crime relationship using city-level crime and temperature data, treating this as a benchmark against which coarser approaches could be judged. They then aggregated both the crime data and the temperature data to the county level and re-estimated the same relationship. The result was consistent with the theoretical prediction: when both outcomes and temperatures were measured at a coarser geographic scale, the estimated effect of extreme heat on violent crime became smaller. The attenuation was not a property of the underlying physics or human behavior but an artifact of how temperature was matched to the units of observation.

This finding has implications that extend well beyond criminology. The same mismatch between temperature proxies and the true exposure of people, crops, and infrastructure affects studies of heat-related mortality, labor productivity, energy consumption, and agricultural output. Whenever the proxy error varies systematically with geography, the estimated dose-response relationship between temperature and the outcome of interest is biased. In some settings the bias runs downward, making heat appear less damaging than it truly is; in others it can run upward. The authors emphasize that the direction and size of the bias depend on how the error correlates with both the local climate and the outcome being modeled, which is precisely why simple fixes, such as adding more control variables or switching to a different functional form, cannot be relied upon to restore accuracy.

The policy stakes are considerable. Economic estimates of climate damages feed directly into cost-benefit analyses, infrastructure investments, emergency preparedness budgets, and adaptation planning. If those estimates systematically understate how severely extreme heat drives crime rates, crop losses, or health emergencies, then the scale of the policy response will be miscalibrated as well. “If climate-impact studies misestimate how severely extreme heat drives crime rates, crop loss, or health emergencies, policymakers may also misestimate the scale of the responses needed to address those impacts,” Yu said. A 15 percent understatement, compounded across dozens of sectors and repeated in every new study that inherits the same proxy problem, could translate into billions of dollars of underinvestment in cooling centers, grid hardening, heat-health early warning systems, and agricultural adaptation.

Importantly, the study does not merely diagnose the problem; it offers practical guidance for researchers choosing among the growing menu of temperature data products. No single proxy performs best everywhere. In regions with dense weather-station networks, measurements from nearby stations can outperform more complex gridded products, because direct observations at close range capture local conditions that interpolated grids smooth away. In areas where monitoring is sparse, gridded data may perform better, since they incorporate additional information such as satellite retrievals and reanalysis models that partially compensate for the absence of ground stations. The practical lesson is that proxy choice should be an explicit, location-specific decision grounded in validation against known temperatures, rather than a default driven by convenience or habit. Researchers can and should quantify the likely proxy error for their study region before drawing substantive conclusions from it.

The broader prescription, according to the authors, is to reduce measurement error at its source by improving the spatial resolution of temperature monitoring itself. Expanding and maintaining dense weather-station networks would shrink the interpolation distances that generate proxy error in the first place. Equally important is more carefully matching temperature measurements to the locations where exposure actually occurs: where people live, where farms grow crops, where businesses operate, and where ecosystems are under stress. A temperature recorded at an airport runway may poorly represent the heat experienced in a nearby low-income neighborhood with little tree cover, and a county average may poorly represent the conditions on any given field within it. Closing that gap between what is measured and what is experienced is, the study suggests, one of the most cost-effective ways to improve the reliability of climate-impact science.

For a research community increasingly asked to inform high-stakes decisions about adaptation and mitigation, the message is sobering but constructive. The tools exist to do better: denser station networks, careful proxy validation, and matching data to the true units of exposure. What the Yu and Carson study adds is a quantified warning that the status quo carries a measurable cost, one large enough to rival the very climate signal that decades of research have worked to detect. As extreme heat intensifies and its consequences for health, safety, agriculture, and energy systems grow, ensuring that the temperatures feeding into those analyses are accurate is not a technical footnote. It is a prerequisite for knowing, with confidence, how much is truly at stake.

Subject of Research: Measurement error in temperature proxies used in economic climate-impact studies

Article Title: Temperature measurement error may cause climate-impact studies to understate impacts by 15% or more

Article References: Temperature measurement error may cause climate-impact studies to understate impacts by 15% or more. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: climate impacts, temperature measurement error, extreme heat, violent crime, weather stations, gridded climate data, econometrics, Paris Agreement, public health, crop yields, adaptation policy, Wake Forest University