November 23, 2010

Hashtag Analytics: Part 2 = Engagement

Today, we're going to explore a few introductory metrics, metrics that help us understand how the #blogchat community behaves.

I created a variable called "Engage".  This variable has a 1/0 value, a "1" if the user issued at least one tweet within the #blogchat community for the week ending November 11, 0 otherwise.

I then summarized activity for the prior four weeks, weeks ending November 4, October 28, October 21, and October 14.  I created two new variables, and transformed the remaining eight variables.
  • New Variable = Recency ... defined as "weeks since last participation."  This variable can have a value of 1, 2, 3, or 4 ... one means the user last participated one week ago, four means that the user last participated four weeks ago.
  • New Variable = Weeks ... a sum of the number of times the user participated in the last four weeks.  This variable can have a value of 1, 2, 3, or 4 ... four means that the user participated in each of the past four weeks, one means that the user only participated in one of the past four weeks.
  • Statement, Re-Tweet, Amplify, Converse, Link, Tweets, RT, and ANSW are all "averaged" for the four weeks.  In other words, if a user participated in each of the four weeks, and had 2 statements, 9 statements, 4 statements, and 17 statements, I calculate an average ... 8 statements per week.  This process is repeated for each variable.
At this point, I have a dataset with eleven variables.
  • Engage:  Did user engage the week of November 11 (1/0)?
  • Recency:  Weeks since last participation (through November 4).
  • Weeks:  Number of times user participated in past four weeks.
  • Tweets:  Average number of tweets per week.
  • Statements:  Average number of statements per week.
  • Re-Tweets:  Average number of times re-tweeting other comments per week.
  • Amplify:  Average number of times amplifying the comments of others, per week.
  • Converse:  Average number of times conversing with others, per week.
  • Links:  Average number of links mentioned, per week.
  • RT:  Average number of times re-tweeted by others, per week.
  • ANSW:  Average number of times answered by others, per week.
Finally, I transformed the last two variables, into 1/0 indicators.
  • RT:  1 if user was ever re-tweeted, 0 otherwise.
  • ANSW:  1 if user was ever answered, 0 otherwise.
This dataset provides us with a rich set of community dynamics, dynamics that can be analyzed in many interesting ways.

For instance, I measured "engagement", the percentage of users who participated the week ending November 11, based on "recency", the number of weeks since the user last participated.  Take a look at the findings:


Recency Users Engage Totals % Totals
1 496 38.7% 192 16.7%
2 530 17.7% 94 8.2%
3 342 11.7% 40 3.5%
4 206 9.7% 20 1.7%
99 807 100.0% 807 70.0%



1,153

Take a look at the first row.  Users who last participated the week ending November 4 had a 38.7% chance of engaging (i.e. issuing at least one tweet with the #blogchat hashtag) during the week ending November 11.  496 users had a 38.7% engagement rate, yielding 192 users who participated during the week ending November 11.


This is a pretty low rate.  And look what happens as participation becomes more "distant" ... 17.7% of those who last participated two weeks ago engaged ... 11.7% of those who last participated three weeks ago engaged ... and 9.7% of those who last participated four weeks ago engaged.


In other words, it is really important to keep the user "engaged".  If the user takes a week off, or two weeks off, the user becomes less and less likely to engage in the future.


Look at the column labeled totals.  This is the number of users by segment.  In total, there were 1,153 users who participated in #blogchat during the week ending November 11.  Most important, 807 of the 1,153 users, 70% in total, had not participated in the past four weeks.


This is sooooooooo important.  The success of the #blogchat event for the week ending November 11 was largely due to the fact that 70% of those who tweeted something had not tweeted anything in the prior four weeks!


In Direct Marketing, we know that New Customer Acquisition means EVERYTHING.


In Social Media Communities, we have the first piece of evidence that New User Acquisition means A LOT!


Tomorrow, we'll continue to analyze engagement rates across various segments.

2 comments:

  1. Hey Kevin! Love these posts, and you always make my head hurt (in a good way).

    One thing I would mention in analyzing the behavior of the #Blogchat participants. There really are 3 'types' of #Blogchats:

    1 - A #Blogchat that has a co-host
    2 - A #Blogchat where I am driving the discussion
    3 - The OPEN MIC #Blogchats

    In general, the co-hosts are on the 1st Sunday of the month, and the OPEN MIC sessions are the last Sunday of the month.

    I can't prove this with numbers, but my sense is that the #Blogchat 'regulars' will make up the highest concentration of participants during the sessions where I am leading the discussion. After that it would be for the OPEN MIC sessions, and the #Blogchats with the most 'first time' #Blogchatters would be the ones with co-hosts.

    I think it's important to consider this when you look at the behavior of the participants, especially when you are considering engagement from week to week.

    For example, when @ProBlogger co-hosted a few weeks ago, I noticed a TON of 1st timers, obviously people that were fans of Darren, and wanted to check out #Blogchat cause he was co-hosting. Obviously, a significant portion of those people would probably not return the next week, since Darren wasn't co-hosting. I see the same thing for all the co-hosts.

    Just wanted to point this out as I think each time of #Blogchat has a slightly different makeup, and as a result, will probably have different behavior and engagement characteristics.

    Thanks again for writing this, and have a Happy Thanksgiving!

    ReplyDelete
  2. I agree that each #blogchat has a slightly different makeup, with different behavior and engagement characteristics.

    When I've analyzed prior weeks, I find that the metrics vary by +/- 30%, meaning that a 15% engagement one week becomes an 10% to 20% rate the next week ... yielding very similar overall behavior.

    The thing you and your community do best is that you are appealing to new users. The week I am describing had a disproportionate number of newbies ... but overall, across all weeks I analyzed, about 60% of those who participate are "new".

    Another 6-10 posts coming!

    ReplyDelete