Re: [Analytics] weekly periodicity mode

Hey James, I'm not sure if this will be helpful but I've analyzed the weekly (and yearly) periodicities in *editor* data for various Wikipedias. At https://github.com/fasiha/wikiatrisk#figure-day-of-week-trends-of-the-numbe… I show the day-of-week breakdown for the number of editors on English and Japanese Wikipedias. I was surprised to see how, on English, Monday–Thursday see the most editing, with editors taking a break Friday–Sunday, but on Japanese, Saturday and Sunday see more edits. Perhaps more useful to you might be how I *quantify* the weekly (or yearly) periodicity. I have two ways of doing this: you can look at the auto-correlation with 7-day lags and you can look at the strength of the periodogram (or any such spectral estimate) at 7-day periods. (These two are tightly related, since the periodogram is basically estimating the Fourier transform of the auto-correlation function.) Both these are discussed at that link, and you should be able to adapt the method/code to pageview data. (And feel free to reach out if you want me to summarize those findings, my writing there is very long because I'm trying to both understand and describe my findings.) I was just using daily data (the finest-grained data for editors). Since the pageviews data is available at hourly-granularity, I'm sure you can run the same kinds of analysis for pageviews and make a choropleth/heatmap. The only thing that I might mention is that the hourly pageview data is only available for the last couple of years (unlike editor data that I show in the link above, which is available since 2000), but this shouldn't be a big problem. On Mon, Apr 23, 2018 at 2:29 PM, James Salsman <jsalsman(a)gmail.com> wrote: