Fig. 3

Relevant distributions of our dataset. Please note the following points: First, we performed kernel density estimations (KDEs (kernel density estimations (KDEs; bandwidth selection according to Scott’s rule) to approximate the underlying probability density functions. Second, the dashed line and the colored area reflect the median and the interquartile range (IQR), respectively. Third, the full dataset, i.e., before story exclusion (see Additional file Table view, Fig. 2), was used for panels a and b. a KDE of the number of replies per story; the x-axis was \(\text{log}_{10}(x + 1)\)-transformed. b Time series (KDE-approximated) of all replies (\(N = 2.07 \cdot 10^{5}\)) c KDE of the number of fact-checked tweets per story; the x-axis is \(\text{log}_{10}(x + 1)\)-transformed. d, e KDEs of \(\text{Recall}\) d and \(\mathrm {Overall\ Precision}\) e. Note that these measures are estimates. The density evaluation was restricted to \([0, 1]\). f KDE of the number of matching tweets. These data were accessed via the Tweet count endpoint, and that the number of retrieved tweets for a given story might be lower due to downsampling. g Time series (KDE-approximated) of all retrieved tweets (\(N = 2.69 \cdot 10^{7}\)). h KDE of the estimated story duration. The density evaluation was restricted, as we limited the observation period to 1 year