My musing about Twitter accounts that I posted a week ago started a bunch of conversations and got me looking at it a bit more closely. Now fair warning – that post was the beginning of a look at the diversity of a denomination by thinking about how many different “voices” there are coming from that branch. Ultimately I want to find a way to categorize those voices on a diversity spectrum but a couple of metrics I have tried already did not pan out. However, in casting the net a bit wider, that is in bringing more denominations into the data set, an interesting relationship appeared.
As we drill into that data a brief reminder about the data set. I was looking for official Twitter accounts from a denomination. My original list from the PC(USA) included the primary account, agencies, committees, periodicals and news sources. It did not include what I characterized as commercial project-specific accounts – like the Glory to God Hymnal and the Feasting on the Word series – as well as not counting seminaries and conference centers. As I move on to other denominations I will stick to these same parameters even though some have seminaries and conference centers with much closer oversight by their highest governing bodies. In addition, I am choosing at the onset of this analysis to include the inactive, duplicate and periodical accounts.
In this search for denominational Twitter accounts I found one more for the PC(USA) and have added that to the list in the original post and annotated it as an update. For the rest of the usual American Presbyterian branches I have these that I found:
ARPC – 32,000 members (from current issue of The ARP)
RPCNA – 7,000 members (from current issue of The ARP)
OPC – 31,122 (from Statistician’s report to 2015 GA)
No official Twitter accounts found
PCA – 358,516 members (from Clerk’s summary of 2015 GA)
EPC – 149,527 reported (from statistical report to 2015 GA)
BPC – 3500 members (Wikipedia)
No official Twitter accounts found
ECO – 60,000 members (report from 2014 Synod meeting)
Cumberland – 72,370 members (2015 GA Minutes Statistical Reports for 2014)
CPCA – 7676 members (2014 GA Minutes Statistical Reports for 2013)
No official Twitter accounts found
So if we take these and plot Twitter accounts versus membership what do we get? Here is the graph.
That’s a pretty nice trend line there — all the data give a correlation of 0.990. Tough to beat that. But those who regularly deal with statistics will notice a couple of issues.
First and foremost the trend line is highly leveraged. That is to say that you have a lot of data on the left and then a really, really long space until you get to the PC(USA) on the right. When calculating the trend that isolated data point can dominate and pull the trend line to itself. Compared to the actual number of 39 Twitter accounts the trend line predicts 39.06 accounts. Yes, there is the clear possibility of leveraging.
Second, even the data point for the PCA is a bit isolated there away from the cluster. In a sense, we have the statistics of small numbers with three meaningful populations: the PC(USA) point on the right, the PCA point in the middle and the cluster containing everyone else on the left.
However, looking at the data and the trend line it still seems to be a decent fit. Yes, the PC(USA) has leveraged it but the predicted 9.11 accounts for the PCA is still reasonably close to the actual 10 accounts. So let’s test the leveraging.
Dropping the PC(USA) point from the linear regression and fitting only on the lower nine points, including the PCA, the correlation drops to 0.827. So there is a correlation drop indicating some leveraging but that is still a respectably strong number. But have a look at the plot…
So if the trend line is only based on the lower nine data points and then extrapolated out four times that distance to predict the PC(USA) value, it only over-estimates by 1.54. This is starting to look like a more robust relationship.
Having now had a look at the data let me tell you that what I found is significantly different than my expected outcome. You might have noticed that a bit of my bias crept into the last post regarding the PC(USA) having a high number of Twitter accounts. As I was compiling that list it seemed to me that the church had gone wild in creating accounts. Well, when viewed from the perspective of number of accounts per thousand members (that would be 0.024 accounts/member for the trend line if you care) the number is right in line with everyone else. They just happen to be four times larger than the next largest branch so the number of accounts is four times larger.
From a statistical point of view I went into this expecting that I would never be able to plot this on a linear line. I was expecting to have to fit it to a log scale on the number of accounts axis. Furthermore, from past experience I also expected the leveraging to be more dramatic and the extrapolated line to miss by a wider margin. So I share this little experiment to document something that truly surprised me when I took a close look at it. And furthermore, the decision of which accounts to include and which to exclude from the count was made at the beginning and carried through the analysis. It would of course be interesting to try this again with other subsets but I have not tried those and will leave that for another day.
Now, what we can say is that the number of accounts that the PCA and the PC(USA) have are completely in line with each other and generally with the smaller churches as well. While the smaller branches scatter a bit more around the line the trend is generally evident in that cluster.
What we can not say is whether, from an administrative and social media point of view, the PC(USA) and maybe the PCA have too many Twitter accounts. There is a statistical relationship here but that does not tell us whether the number of accounts per member helps or does not help get the message out. Furthermore, this relationship does not answer any questions about the consistency or coherence of the message in social media or the diversity of the branch as a whole.
Some of my preliminary thoughts are what this might mean for scaling relationships of institutional structure and self-similarity as a means of probing institutional development. In particular, it might be an interesting on-going study to see how accounts might be added as ECO becomes larger and how accounts might go dormant as the PC(USA) scales back its operations.
But it is a very interesting relationship and I put it out there for any social media theorists or practitioners who might be interested in this sort of thing. As I said, I was surprised by the proportionality, robustness and consistency of the relationship. I welcome any of you that are interested to continue pondering with me what possible implications there might be.