> We wanted to see why Uzbekistan didn’t jump out, so we reproduced it in our comment (Extended Data Fig 1). It turns out that Uzbekistan wasn’t even the biggest outlier, but that the version they had published had the axes cropped so you couldn’t see the outliers (see red boxes in our version). This seemed indicative of a different issue, which is why we documented it in the comment.
Cropping the chart to hide the outliers is so bad that I can't tell if they're incompetent or malicious. I wouldn't be surprised if this is the kind of thing an LLM would produce in the hands of an operator not paying too much attention, but the paper was published in the time period before LLMs were everywhere in publishing.
Figure 1a (the leftmost subfigure in TFA’s lead image) shows the data for Uzbekistan from the DOSEv1 dataset (green), DOSEv2 dataset (red) and World Bank (black). The authors of the retracted study used the DOSEv2 dataset in order to model climate effects on the economy at a sub-national level, as opposed to the country-level analyses used in prior work. However, it looks like the DOSEv2 data was just bad for all 14 provinces in Uzbekistan (a 90% drop in GDP for all provinces in 2020!).
The typical correlation between weather and the economy is going to be fairly noisy across the dataset, but if you have 14 extra datapoints all saying there’s a catastrophic GDP crash in one year together with some coincidental weather effect, that’s going to bias the model hard. Notably, they also extrapolate losses forward all the way to 2100, so the effects of such a bias will compound.
We need something akin to the international geophysical year, but for data integrity. Make it an interdisciplinary priority to clean house and root out papers that are hanging by a thread of included / excluded outliers, biased samples, and outright fraud. It would be humbling, but we'd be in much better shape afterwards.
The article suggests it's unreasonable numbers in the original Uzbekistan data source and that other datapoints may have been worse, the authors just didn't correctly execute their basic checks.
"It turns out that Uzbekistan wasn’t even the biggest outlier, but that the version they had published had the axes cropped so you couldn’t see the outliers..."
They don’t specify, but based on the period they’re talking about I’d put money on it being related to the cotton scandal, to pripiski - that is, the Soviet tendency to make up production figures. When glasnost happened in ‘88 the fiction collapsed, although not immediately - most cotton producers continued to bullshit about their numbers until the mid 90s, while the industry dwindled due to lack of water for irrigation and desertification.
The cascading effect is what makes this particularly dangerous. One bad data point doesn't just produce one wrong conclusion, it gets cited, incorporated into meta-analyses, and eventually shapes policy. By the time someone traces it back to a cropped chart and a suspicious outlier, the conclusions drawn from it have their own citation momentum. The fix isn't just better peer review, it's making raw datasets reproducible enough that anomalies like a 90% GDP drop across 14 provinces get flagged automatically before publication.
Cropping the chart to hide the outliers is so bad that I can't tell if they're incompetent or malicious. I wouldn't be surprised if this is the kind of thing an LLM would produce in the hands of an operator not paying too much attention, but the paper was published in the time period before LLMs were everywhere in publishing.
reply