It appears that even if you don't enter `ci` argument, it's set to `95` by default in [`barplot`][1] method, which gives you those error bars. If you don't want them, set `ci = None` while plotting. The bar is the confidence interval for the variable as the library describes, i.e. it's interval where the `95 %` of your variable lies in, which is estimated by [bootstrapping][2].


  [1]: https://github.com/mwaskom/seaborn/blob/fe172c44777f87159e9e2b00e919bbd055d6f696/seaborn/categorical.py
  [2]: https://www.wikiwand.com/en/Bootstrapping_(statistics)