As promised back in my initial post on game thread habits among NL Central blogs, I've got a full paper available on my analysis of the full 2012 season at SBN. Some caveats:
- This initial draft is only the NL. My RA has a new baby this semester, so it's taken him longer than expected to do all the archiving. The AL should be done pretty soon, and adding those sites into the analysis will be a fairly trivial process and will be part of the final conference version of this paper assuming it's accepted. I wouldn't expect this to change the results significantly.
- If you're interested in the literature I cite, there's a good amount of interesting stuff in these areas that I could have included, but left out because it was either redundant or more focused on individual psychological processes that I wasn't testing here. But if you have access to journal subscriptions or a campus library, the literature on fan-team identification is worth exploring. I will probably expand on this a bit in my revisions, as I think I could probably be clearer in supporting my hypotheses, but I had a submission deadline to hit.
The analysis was done using multilevel modeling (with team blogs as the group level and game threads as the subject level), and if any of our stat people have suggestions on how I use it, I'm all ears. I haven't done this type of analysis in about seven years, and spent a lot of time trying to refamiliarize myself with it, but almost every text I found was both too abstract to be helpful, and dealt with too narrow examples to apply to what I was trying to do. The big thing that I need to add is random effects for a number of the predictors that are currently only examined as fixed. There are also some additional interactions that the literature suggests that I didn't have time to include before I decided I ought to write the draft instead of spending all my time on analysis. But for now, the main thing to understand is that all the findings below are true notwithstanding the differences across teams, which are quite strong.
- The number of participants in any given game thread is strongly influenced by factors relating to short- and long-term team success, as well as the nature of the relationship with the opponent. Games following wins, divisional games, and games when the team has better records all get more participants.
- Schedule isn't that important. Weekday games get more participants, but day vs. night doesn't matter, and the two factors don't interact. Surprisingly, the more time zones the game's venue and the team's home are, the more participants there are. Not surprisingly, the number of participants declines throughout the season, which is probably a combination of fatigue, teams falling out of contention, and competition from football.
- I hypothesize pre-game factors influencing the number of participants, but two in-game factors play a role as well, perhaps by getting lurkers to chime in. Total runs scored and the team's winning (or losing) margin positively relate to participants, but the game's margin (regardless of who wins) doesn't.
- The number of comments per participant (i.e., how talkative people are in the thread) is strongly influenced by in-game factors -- total runs, team's margin, overall margin -- that add to the general and team-specific excitement level. But pre-game factors have almost no impact. People are more talkative on weekends (but, c.f. above, there are fewer total people on weekends, and probably it's the quieter people who don't show up then), at night, and there is an interaction. But there's no season-long decline, and team-related pre-game factors (team success, opponent) have no impact.
- I think there's probably a complicated relationship between number of participants and number of comments per, but it's going to take more thinking through the theory and how to analyze it to get that unwound. It will probably end up in the version of this that gets sent out to a journal.