so a categorical variable got mixed up as a numerical one and produced misleading results.

to the credit of the authors, they released their data sets. -- however, i suspect that proper data exploration and visualisation would have prevented all this. visual inspection would have most likely revealed that there is no visible effect, or even an effect in the opposite direction, and once you see this, all alarm bells should go off if your model predicts otherwise. so i suspect that the authors skipped some basic steps and got carried away by results that promised a nice headline.