Avoiding overstatement in climate research

Together with a large group of scientists from different disciplines and regions of the world, we wrote a perspective piece with Environmental Research Letters that just came out: https://lnkd.in/dxxE4NEq It cautions against premature or top-down characterizations of areas as uninhabitable, or portrayals of large-scale climate-induced displacement as inevitable—particularly when the perspectives and preferences of affected populations are excluded. While we recognize the importance of modelling and scenario-building to assess future risks, we argue that such efforts must be grounded in local realities and include diverse forms of knowledge. We propose five guiding recommendations: (1) avoid declaring hard limits to habitability without inclusive, context-specific assessments; (2) treat model-based projections as possible, not predetermined futures; (3) reject simplistic global North/South assumptions in assessing vulnerability and mobility; (4) uphold people’s right to remain, alongside the right to move; and (5) prioritize investment in in-situ adaptation that addresses structural inequalities. These principles aim to inform reflexive and justice-oriented approaches to climate mobility and habitability research.

Overstating research findings is one of the quickest ways to destroy public trust... Here is how 𝗢𝘃𝗲𝗿𝘀𝘁𝗮𝘁𝗶𝗻𝗴 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗞𝗶𝗹𝗹𝘀 𝗣𝘂𝗯𝗹𝗶𝗰 𝗧𝗿𝘂𝘀𝘁 One exaggerated headline, one oversimplified study, and public trust in science suddenly erodes. Overhyping research leads to misinformation, unrealistic expectations, and skepticism when results don’t hold up. 📌 The "Breakthrough" Trap → Science is gradual, but people expect instant solutions. ↳ Calling every study a "game-changer" sets up false expectations. ↳ When research gets debunked, trust plummets (e.g., red wine is as good as exercise?). How we can fix this: Use measured language: "Initial evidence suggests..." instead of "Revolutionary discovery!" 📌 Social Media Fuels Misinformation → Research spreads fast ✒︎but misinformation spreads faster. ↳ Viral posts often take findings out of context ↳ Echo chambers amplify misleading or incomplete claims. ↳ People trust repetition, even if the info is false. How to Fix this: Scientists should actively engage in public discussions and counter misinformation. 📌 Science "Changes" & the Public Feels Betrayed → When studies evolve, people feel misled if initial messaging lacked transparency. ↳ The mask debate during COVID-19 led to confusion because guidance kept shifting. ↳ The public expects certainty—but science is about updating knowledge. ↳ Without clear communication, corrections look like contradictions. Fix It: Normalize uncertainty—“Here’s what we know so far, and here’s what we’re still learning.” 📌 The Balance Between Engagement & Accuracy → Scientists and the media must work together to avoid overhyping research. ↳ Too much data? People tune out. ↳ Too little nuance? People get misled. ↳ Fear-based messaging? Causes panic or apathy. ********** When science gets overhyped, credibility suffers. The more we focus on accuracy over attention, the stronger public trust becomes. 💬 What’s an example of overhyped science you’ve seen in the media? #ScienceCommunication #Misinformation #PublicTrust #ResearchEthics

🔎 Here are common mistakes to avoid: 1️⃣ Using Relative Change to Exaggerate Small Differences Explanation: Relative changes can misleadingly make small differences look big. Present absolute differences. ❌ "Incidence increased by 100%" ✅ "Cases increased from 2 to 4 in a population of one million" 2️⃣ Reporting Raw Counts Instead of Percentages Explanation: Raw counts can exaggerate small effects in large populations, while percentages are standardized & account for population size. ❌ "There were 25 million smokers" ✅ "Smoking prevalence was 2%" 3️⃣ Using Odds Ratios Instead of Prevalence Ratios for Common Outcomes Explanation: When an outcome is common, odds ratios can inflate perceived differences. Prevalence ratios are more conservative. ❌ "Smoking odds were 123.45 times higher among X than Y" (odds ratios) ✅ "Smoking likelihood was 10.1 times higher among X than Y" (prevalence ratios) 4️⃣ Misleading Significance with Terms Like "Almost Significant" Explanation: Such terms can imply importance where statistical thresholds haven’t been met ❌ "Results were almost significant (p=0.06)" ✅ "Results were not statistically significant (p=0.06)" 5️⃣ Reporting Unstable Estimates with Large Standard Errors Explanation: Better to omit imprecise estimates ❌ Results presented despite very large standard errors. ✅ Omit results with relative standard error >30% (RSE=standard error/proportion *100) 6️⃣ Truncating Axes in Graphs to Emphasize Findings Explanation: Truncated axes can make small differences appear larger than they are. ❌ Graph only shows data from 75%-100% ✅ Display the full axis range to ensure a fair comparison, such as 0%-100% 7️⃣ Overstating Implications with Terms Like "Proves" or "Needs" Explanation: May imply certainty and necessity that findings often don’t support. ❌ "Our results prove the need to implement XYZ" ✅ "Our findings suggest that implementing XYZ may be beneficial" 8️⃣ Using Data That Isn’t Fit for Purpose but Adding "Interpret with Caution" Explanation: This undermines the credibility of findings. Only include data fit for use and fit for purpose. ❌ "The results should be interpreted with caution due to data limitations." ✅ If data are not fit for use/purpose, do not use them. 9️⃣ Using Causal Language for Observational Studies Explanation: Terms like "cause", "effect", "attributable", "impact" imply causation that observational studies can’t establish. "Association" is more appropriate. ❌ "Our cross-sectional results showed the effect of X on Y" ✅ "Our cross-sectional results showed an association between X and Y" 🔟 Testing Repeatedly to Find Significance (P-hacking) Explanation: This is a fishing expedition, also known as a type 1 statistical error (false positive results). ❌ Testing various subgroups until significant p-values appear. ✅ Predefine hypotheses and analyses and report exactly what was found. Ethical communication preserves trust. Let's commit to clear, honest reporting. 🌱 #WriteRight

The most dangerous thing in climate risk today is not “too little data”. It’s pretending we know more than we do. I just re-read Lars Peter Hansen ’s paper on climate change and central banking. It is brilliant and it should be mandatory reading for anyone building climate stress tests, scenarios, or “AI-powered climate models.” One line really stuck with me: “Cunningly chosen parsimonious models often do provide remarkably useful approximations.” Professor Hansen won the Nobel Prize in Economic Sciences in 2013. When someone with his stature writes, we should all pay close attention. According to Hansen, climate risk lives in a world of: ✔️ deep uncertainty ✔️ model misspecification ✔️ limited historical experience And yet, we keep producing ever more complex dashboards, scenario trees, and long-horizon stress tests. This is often without being honest about what is assumed, what is guessed, and what is unknowable. A parsimonious model is complex enough to be useful, and simple enough to be credible. They: ➡️ Maximize insight with the minimum amount of data required; ➡️ Avoid "unnecessary things," such as redundant variables or overly complex structure; ➡️ Does not sacrifice necessary complexity; it includes only what is essential to describe the phenomenon robustly According to Hansen, quantitative models should be understood as a form of quantitative storytelling. Their purpose is not to fully replicate reality or deliver precise forecasts. The purpose is to discipline thinking under deep uncertainty. In the context of climate risk, Hansen warns that overstating model precision creates a dangerous pretense of knowledge; credibility comes instead from transparency, humility, and models that are simple enough to be interrogated and adapted as understanding evolves. Read the paper here: https://lnkd.in/eNK-tkrh Do you want to learn more how we put this thinking into practice? Reach out to me at gerhard@climateriskservices.com www.adaptationfinance.com