Participation at Dailykos over time is a subject of some interest and speculation for both registered users and politically minded folk everywhere.
With a decade of data in hand it is long past time to take a look.
With this in mind, statistical summaries of participation over 536 consecutive weeks at daily kos were prepared to provide a starting point for discussion.
Jump through the curly queue of orange hue with me to further explore.
How can we, or should we, measure participation?
Our only choice is things we count, or derive from several counts: measures.
I looked at measures in two categories, people and products.
People: number of ...
authors
recommenders
commenters
participants (union of authors, recommenders, and commenters)
Products: number of
posts
recommendations
comments
or
impact (in units of bharns, as defined in High Impact Post lists)
As already noted, these measures were prepared for 536 weeks starting from January 4, 2004, and ending with the week beginning April 6, 2014.
For those interested, a full description of preliminary analysis and the subsequent choice of a few (3) measures to concentrate on is provided as an appendix.
For now, take it for granted that number of posts, number of participants, and impact cover the waterfront in terms of participation.
The graphic below shows weekly impact, number of participants, and number of posts from January 2004 to April 2014.
Behold!
For those who've been around, there are few surprises here.
Q1 2005 was a time of increasing participation, and a sort of joy as many first joined the site. Right?
That is the quarter I used to establish the standard for impact, based on the only daily feature at that time, Cheers and Jeers.
What else do we see?
Steady growth through the Bush years with peaks at election time and around Katrina. Check.
A massive spike around the 2008 election. Check.
A steady drop in participation in 2009 - 2010, with a very feeble bump at mid term. Check.
(Everybody remembers those days, not particularly fondly)
Starting in 2011 the rate of decline is smaller. That's interesting to see.
There is an election bump in 2012, though nothing like 2008.
Any other notable events? Put them in the comments.
So what might we expect the graph to look like with two more years of data, ie after the 2016 election?
Will participation continue to decline, or are we in for another upturn?
What will the 2016 election bump look like: the one in 2004, or 2008, or 2012, or somewhere in between?
I don't know, and neither do you, but that's how the pundits make a living.
Go ahead, Dive in!
Shifting gears somewhat: what else can we use to help us understand this data set?
It is often informative to find other time series tightly correlated with yours.
Who has one of the most interesting time series collections in the world, and has implemented a way to compare our data with theirs?
Google.
Namely,Google Correlate, a feature of Google Trends.
Take a look: Google Correlate: Daily Kos Impact
Here are the top 5 query terms.
- 0.8679 political ticker
- 0.8602 tpm
- 0.8438 cnn ticker
- 0.8416 cnn political ticker
- 0.8375 politico.com
The search terms are all of a political nature, and likely familiar to most.
The time series for tpm closely matches participation at Dailykos over the 2004-2014 interval.
But 2009-2012 is less well matched.
Fortunately, it is possible to select that time range and search again for matching query terms.
The top query term is shown here vs dailykos impact 2009-2012.
Ha ha! Search for conservative talk radio closely matches participation at Dailykos over 2009-2012.
So, over ten years, or over 4 years, it is difficult to argue that anything particular or unique to Dailykos (pie fights, trolls, mass bannings, dramatic exits, site redesign, etc) had major effects on participation.
I think the data shows that Dailykos rides squarely in the middle of national political currents and interests.
What's next? I hope to find a time series derived from external participation at dailykos (by which I mean page views, unique visits, etc).
I hope you find this look over a decades worth of participation at Dailykos interesting and illuminating.
If you do visit Google Correlate and play around, let me know what you find.
-- jotter
APPENDIX - Preliminary Analysis and choice of representative measures
Each measure was normalized by subtracting its mean value and dividing by its standard deviation.
This procedure puts all the measures in the same range, aiding both visualization and correlation.
Exploratory data analysis (plots, correlation, heat maps of correlation), revealed three groups of measures of participation.
In the first group we have number of posts and of authors.
The correlation coefficient (r) between these two measures is 0.98, which is quite close to the maximum possible value (1.0).
This seems reasonable - you can't have posts without authors, and you should expect an average author to produce an average number of posts.
Group one will be represented by number of posts.
In the second group we have number of participants (the union of recommenders, commenters, and authors), as well as, individually, recommenders and commenters.
Recommenders, commenters, and participants have r >= 0.99 among them, while authors is only 0.95 vs any of the others.
Group two is represented by number of participants.
In the third group we have recommendations, comments, and impact, with r=.99 for impact vs recommendations, r=0.91 for impact vs comments, and r=0.84 for recommendations vs comments.
Impact represents group three.
The preliminary analysis allowed us to reduce the number of candidate measures from eight to three.
As you would hope, these three are not as well correlated with each other as with the other measures in their group. For impact vs posts r=0.76, for impact vs participants r=0.84, and for posts vs participants r=0.93.
Continuing on, each measure was smoothed by replacing each point with the average of the adjacent 6 weeks.
This removes some detail (maybe noise, maybe small but interesting changes), but improves graphics.