The chart from the meta-analysis does actually look very random, and the OP article discusses several confounding factors which might make us doubt the evidence. It’s easy to find patterns in noisy data.