[ATM] List Traffic

[Anthony Stillman <atmer@flash.net>]

Not too long ago in response to a dearth of list traffic I analyzed the
archive for patterns of activity (1996-2003).   Here's some of what I found.

Smoothed, the monthly number of messages increased from the inception of
the list until mid 2002.  After that it fell about 35 percent and has been
holding steady since, maybe even climbing a bit.
<http://www.atmlist.net/contrib/atmer-at-flash-dot-net/RAW.JPG>
Smoothed, the number of original threads has, more or less,  stayed constant.

It seems like the number of new subject lines and the number of subjects
discussed would be related to one another.  However, threads evolve, new
questions are asked, tangents are gone off on.  Only occasionally are new
threads born of this.   Based on my assumptions then, I expect that the
true  "new thread" count is larger than what is represented and has
actually grown with the number of replies.

An auto correlation of the month to month reply counts reveals a weak
twelve month cycle.  I chopped the data up by year and averaged each
month's counts across the years.
<http://www.atmlist.net/contrib/atmer-at-flash-dot-net/TOTAL.JPG>
<http://www.atmlist.net/contrib/atmer-at-flash-dot-net/NEW.JPG>
It almost appears as if there is in fact a  relationship between new
subject lines and replies.  A cross correlation weakly confirms this.  Even
more interesting are the graphs features.  The valley of vacation in June
and July makes sense, but the apparent peak of activity in March, that I
don't understand.  As for the September artifact, it was there before 2001.

Don't look too close at the error bars.


Anthony

ex uno disce omnes
>From one, learn of all.



The graphs

1)  New Threads and Total Number of Messages

Note that the new thread count never exceeds 30.


2)  Total Message Count
     Scaled by year end totals   Normalized to system average
     1996-2003

     New Thread Count
     Scaled by year end totals   Normalized to system average
     1996-2003


What I did was, there were fewer messages in the beginning and hence those
years were unfairly under represented, I scaled each month's value by that
years total and then averaged those scaled values.  That is, the Jan '96
value was scaled by the '96 total message count as was the Jan'97 value
scaled by the '97 total message count, and so on.  Then all of the scaled
Jan values were averaged.  This was repeated for each month.  Finally,
those values were scaled by the data set's single average monthly value.


_______________________________________________
ATM mailing list http://www.atmlist.net/