Continuing on the same track as the last blog post, I read “Global non-linear effect of temperature on economic production” written by Stanford scholars Marshall Burke, Solomon M. Hsiang, and Edward Miguel called “Global non-linear effect of temperature on economic production.” This article explores the phenomenon that the effect of temp on GDP varies according to the fact that the effect of temperature on GDP is not constant across all temperatures. Instead, the impact changes depending on how hot or cold a country already is.
At cooler average temperatures (below ~13 °C), a small increase in temperature may improve economic productivity. But at warmer temperatures (above ~13 °C), further warming causes increasingly large declines in GDP growth. What makes this study stand out is its nonlinear approach. Instead of assuming every additional degree of warming has the same effect, the authors show that temperature effects vary depending on where you start. That’s why a 1°C rise in Finland might actually help growth, while the same 1°C in Mali could be devastating.
Under a high-emissions future (RCP8.5), global GDP per capita could “decline by ~23% by 2100.” Countries already in hot climates would bear the brunt of that loss. After considering this finding, I began to wonder why 13°C was perceived as an optimum temperature. Different areas around the world logically should have different optimum temperatures depending on factors like geography and agriculture reliance. After doing some research, I realized that this specific number comes directly from the data. While the authors used an impressive dataset–166 countries, spanning from 1960 to 2010, paired with annual average temperature and GDP per capita figures– I noticed some discrepancies. The major one being the exclusion of data from China and the US (page 3). China and the U.S. are unusually large economies with massive populations and diverse regional climates. Including them could skew the results of a global model because their economic output disproportionately influences averages. However, it could also be argued that these countries not only produce a major share of global GDP but also represent very different climate zones and economic structures. Excluding them and other data could also skew the global temperature growth curve, hide large-scale adaptations already underway, and downplay the economic weight of emissions-heavy nations. Their omission also raises broader questions. How generalizable are the results without the full global picture? And can policy implications drawn from this model fully apply if it omits two economic giants?
Ultimately, the study provides a powerful warning about climate inequality, but it also reminds us that even the strongest models have blind spots and the data that gets left out is just as important as what’s included. In lower-income countries metrics like GDP and population are frequently undercounted or unevenly collected. Informal economies (which dominate in many African nations) rarely show up in official GDP numbers. Population figures may rely on infrequent or incomplete censuses. Many regions lack strong weather station infrastructure, meaning climate data itself is interpolated and estimated over large geographic gaps. These limitations are important because if the baseline data is patchy, any statistical model built on it may reproduce existing biases.
This leads us to the key finding that data is never neutral. The numbers we use to model the world are shaped by who collects them, where they’re collected, and what gets left out. When data on GDP, population, or even temperature is inconsistent (especially in lower-income countries) our conclusions must be cautious. If key regions or sectors are underrepresented, models risk reinforcing the very inequalities they aim to expose. As we navigate the relationship between climate and economics, we must not only ask what the data says, but also what it can’t, and how to build better systems for seeing the full picture. Without reliable data, any prediction about the future is only half-formed.
Impact Blueprint 