User:Halfak (WMF)/Sept. 2013 active editor dip of doom
There were some concerns raised about a dip in the number of active editors observed for Sept. 2013 across Wikis. In order to figure out if this dip is real or a normal part of the trend in participation, I performed a timeseries analysis to forecast the expected participation numbers for September.
I borrowed datasets generated by Erik Zachte containing the number of monthly active editors up to Sept. 2013 for all non-commons wikis and EN/DE/NL wikis individually.
> active_editors = data.table(read.table("monthly_active_editors.tsv", sep="\t", header=T))
>
> editors = active_editors$active_editors[length(active_editors$active_editors):1]
>
> myts = ts(editors[1:length(editors)-1], start=c(2001, 1), end=c(2013, 8), frequency=12)
>
> fit <- HoltWinters(myts)
>
> forecast(fit, 1)
Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
Sep 2013 72035.6 69247.44 74823.75 67771.48 76299.71
>
> editors[length(editors)]
[1] 69097
Here's English, Dutch and German:
> forecast(en.fit, 1)
Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
Sep 2013 25971.88 24473 27470.77 23679.53 28264.23
> en.editors[length(en.editors)]
[1] 24956
>
> forecast(de.fit, 1)
Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
Sep 2013 5470.482 5104.712 5836.253 4911.084 6029.88
> de.editors[length(de.editors)]
[1] 5370
>
> forecast(nl.fit, 1)
Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
Sep 2013 1079.964 995.9386 1163.99 951.4581 1208.47
> nl.editors[length(nl.editors)]
[1] 1046
Summary
editThe datapoint for Sept. 2013 falls well within the expected error for the current trend in active editors and does not represent a significant dip.