Measuring Influence - Now It Gets Tricky

Influence is a key issue in the blogosphere. Clients of social media analysis firms (such as BuzzMetrics, my employer) are keen to learn who the influencers in their markets are, bloggers are eager to climb from the z-list to the a-list, and pundits with a new theory of influence abound. However, without any other reliable signal, the inlink count for a blog - that is to say, the number of other posts or blogs which link to a blog - is taken as the measure.

Influence is a tricky issue. A model that I prefer - and one that avoids any real definition - is that which described influence opportunities. A post that is not read can never influence anyone. A post that is read might influence someone. A result of that influence may be recorded online in the form of another post which may or may not contain a link, or it may not. However, it seems clear that an accurate measure of the number of influence opportunities is to be found in some measure of the readership. One important, and under researched, channel is that of the feed. How many people read your content in a feed reader?

It seems reasonable to assume that if there were a strong correlation between the number of inlinks and the number of readers, then we could happily continue to use the inlink count as a proxy.

In order to explore this idea, I pulled the subscriber statistics for a number of blogs (approximately 800) that use feedburner via the feedburner awareness api. I then graphed these against the number of inlinks from BlogPulse. The resulting graph - shown below - doesn't seem to show a particularly strong correlation between the number of subscribers (on the x-axis) and the number of inlinks (on the y-axis). Note that the data is presented in log-log space.

These results are preliminary.

Comments

[Dibyo]

Just wanted to say that I enjoy your blog.... (it also feeds into my google reader).

Also, I thought I'd mention that wouldn't you count feeds as a stronger measure of influence than just reading? i.e. if you had to make a metric of how 'strong' or 'popular' a blog is, you need to count the number of people who feed it, and the number of people who read it, but give a heavier weight to the feeders than the readers since feeds are a measure of loyalty.

Did I get my point across?

[Moshe Koppel]

Matt,
That actually looks to my naked eye like a pretty decent correlation. What's the R2 of a linear regression?

Moshe

[Sean Tierney]

Maybe I'm missing something but this scatterplot seems to be the expected result to me... it shows correlation between subscribers and inlinks, just not a strong one. But think about the people you read regularly, do you link to them proportionate to how often you read them? Probably not. The people I follow regularly are consistently on point with their insights but I tend to write more about the stuff I run across from my own daily experiences in working and observing or from the "needle in the haystack" I find on del.icio.us. This is a great idea for an experiment but this is about what I would expect. Can you elaborate as to what you were expecting to see?

sean

[Matthew Hurst]

Moshe - R2 is 0.4861, not that good.

[Rama]

That looks like a great correlation to me. R-sq of 0.48 is pretty good considering the nature of data you are trying to measure. And especially if you want to compare influences rather than absolute measures, you have found a cheap proxy. It does not cost much to measure inlinks and that as a surrogate of the influence looks valid, IMHO.

Good luck on that and I would love to see more analysis on that one.

Rama

[Matthew Hurst]

Put it this way, in the range of 100-110 inlinks, feed subscription ranges from 5 to 13, 939. I just pulled this range randomly (and I'm sure the 5 is anomolous in some way). Perhaps one way to measure this would be to count the number of different positions if one used inlinks as a rank to using feed subscribers. One could also measure the obvious distribution of deltas in between the two rankings.