why I took down my climate science video

A climate attribution study used to justify strong claims about human-caused warming and the January Los Angeles wildfires was pulled into a public dispute over whether its key result was actually statistically significant—and the fight spilled into media coverage, social posts, and a deleted climate-science video.

The flashpoint centered on a World Weather Attribution press release claiming that “human-induced warming from burning fossil fuels made the peak January [fire weather index…] 35% more probable,” alongside “high confidence” that human-caused climate change increased the likelihood of the “devastating LA wildfires.” In a strongly worded response, Sabine Hossenfelder argued the study’s own analysis did not support those conclusions because the reported probability ratio was statistically insignificant—meaning the data were compatible with climate change having no effect on the January event.

Her initial concern wasn’t only about what the press release emphasized. She also criticized a broader pattern: climate scientists allegedly know when attribution work is shaky but remain silent because speaking up would be politically inconvenient. That context fueled her frustration that the media amplified the headline claim without scrutinizing the underlying statistical result.

The video then disappeared after about a day. The immediate reason was procedural: roughly 12 hours after posting, she received claims that she had misread the study’s result table. With no quick clarification available while she was away from internet access, she set the video to private to avoid spreading what she feared could be misinformation on a policy-relevant topic. Wildfire risk affects real decisions in Los Angeles, including how urgently people plan for future events.

When she returned and rechecked the paper, the dispute narrowed to the study’s statistical presentation. She walked through how extreme-event attribution typically works: climate models are run twice—once with current global warming and once without—and the probabilities of an event matching the target (here, the LA wildfires) are compared via a ratio. She highlighted two methodological weaknesses often associated with this approach: many models may not reproduce extreme events well in the first place, and the outcome can depend heavily on how the “extreme event” is defined.

Her reading of the results table focused on the mean probability ratio and the 95% confidence interval. The mean was slightly above 1 (consistent with the “35% more probable” framing), but the confidence interval appeared to include 1, which would make the result not statistically significant under common conventions. She also argued the paper’s figure caption and color coding were confusing—blue versus orange, and multiple shades of orange—raising the possibility of legend or hue mix-ups. Further complications came from the fact that the paper did not clearly state what threshold it used for “statistical significance,” and she questioned whether the confidence interval was being applied in the way she assumed.

After several days, an author replied directly that the changes were “unsurprisingly, not statistically significant,” and that the correct color label was “light orange.” With that confirmation, the video was restored.

Still, the episode left her with a broader critique: the study was described in the press release without mentioning it was a “rapid attribution” study and without clarifying that it was not peer reviewed. She also emphasized that her argument does not deny that climate change can increase wildfire likelihood in some regions; it targets the specific claim about the LA January event. She concluded by urging caution about attribution claims, while acknowledging that extreme-event attribution methods vary and not all work in the field is necessarily unreliable.