The anomaly in 2012 appears due to a large PTS user(s) deploying a cluster of thousands of systems with PTS and having made their statistics available at that time online, which distorted the overall figures that year
In my view you should republish the article, but then only use robust statistics like percentiles; the median is the 50 % percentile, P50 and possibly the best here.
Popular choices include P01, P10, P25, P50, P75, P90, P99 together with the Min and Max if tabulated.
Comment