Back of the Envelope Twitter Metrics

The Pareto principle — 80% of the effects/outputs come from 20% of the causes/inputs — works in a variety of situations. The 80:20 rule is useful because it allows us to make decisions based on rough data and a simple calculation. If we could get 80% of the right answers from just 20% of the inputs we would have a great Back-of-the-Envelope [BoE] metric!

Twitter has been swamped with many new metrics in the last year, most of them trying to figure out who the influential tweeters are. I wonder…

Are these scores significantly better than simple BoE metrics?

How accurate are these scores in predicting actual behavior?

How much sociological rigor do they contain? The math is easy, the sociology is hard.

Let's examine a progression of BoE twitter metrics.

The first quick influence metric was "number of followers." The thinking went, the more followers you have the more influential you are. On the surface this appears too simplistic and it is easily gamed and distorted. For a while many early adopters on Twitter where in a mad rush to get more followers — quantity over quality. Some vendors still make money by selling "followers" to fools. But most people on Twitter understand that having random people, who do not know you, follow you, does neither party any good. Followers are a much better estimate of popularity rather than interpersonal influence — Hollywood stars have many followers.

The next BoE Twitter influence measure that gathered some interest was the ratio of Followers to Following. This measure had some background in the various prestige measures in social network analysis. A person in a network is prestigious if they are sought out by many others, while at the same time they do not seek out many others themselves. We divide the number of Followers by the number of Following. Let's call this the FFR ratio. The higher the FFR number, the higher the prestige. A ratio [FFR number] greater than 5 is starting to show some real prestige.

This ratio is a step forward, but it can also be deceiving. Often when people first join Twitter they start by following many people -- those they recognize and those who are recommended by friends. Newbies on Twitter often have the FFR ratio reversed. They follow 5 people for everyone that follows them — until people get used to having the newbie on-line and start following them back. This ratio is a good indicator of not just real new users of Twitter, but can also indicate spambots. Spambots follow many -- often at random -- but are not followed back by any except those who automatically follow everyone who follows them -- a bad idea.

The FFR can be gamed, by keeping the Following count artificially low, though there are often good reasons for being selective in who you follow and keeping that group from getting too large. FFR also gets in the way of people who are very social on Twitter. Some people follow many, but not all, of their followers so that they can carry on direct conversations [DM] with them — only people that both follow each other[have a symmetric tie] can DM reach other. The Following number for this strategy can get high quickly, resulting in their FFR rarely exceeding 5, even though these very social folks may be influential.

Once Twitter Lists started, many believed that List membership would be what number of Followers never was -- a better indicator of influence, of those we listen to. Lists are much harder to create, and people that make them usually put some thought into who belongs in what category. Lists do make the transition from popularity to influence, but they do not eliminate the problem of pure popularity. We known popular tweeters still end up on more Lists than their less well known brethren. How do we eliminate/reduce this overbearing popularity factor?

We can not eliminate the popularity factor, but we can minimize it by using it as the divisor in a ratio — the greater the popularity, the less affect it has on your Listed score. Dividing the number of Lists one is on, by the number of followers gives us a pretty decent BoE that anyone can calculate -- all of the numbers are under each person's profile. Let's call this the Lists to Follower's Ratio[LFR].


LFR = Lists / Followers

LFR gives us a number less than one, I use the first four digits to the right of the decimal for the LFR score. Using my numbers from the graphic above we calculate an LFR of 1056.

Let's take a look at several people I know on Twitter. We will look at Clay Shirky [@cshirky], Venessa Miemis [@VenessaMiemis] and myself [@valdiskrebs]. Looking at Followers only, Clay wins easily, followed by Venessa, a distant second, and then me. Next we look at the FFR. Here again the number of Followers dominates the calculation and again Clay gets the highest score by a great margin. Clay is a popular author and blogger and is very popular on Twitter also. Many newcomers to Twitter probably automatically follow Clay. Several newbie Twitter apps recommend to follow the popular people on Twitter such as @scobelizer, @timoreilly, @cshirky @techcrunch and others. This just increases the popularity of already popular people. Maybe most newbies don't know why they are following these people, nor know who these people are? They just follow them.

Next we will look at List membership by itself. Here again popularity skews our number. Same results as before -- Clay is first with Venessa and Valdis a distant second and third.

I have talked about hidden and local influencers before. They do not have the reach of an Oprah in recommending books, but they have a smaller, focused, audience that seeks their advice and opinions on one or more topics. How do we find them? I think the LFR BoE metric gives a decent indicator of who is one of these hidden/focused influencers. Running the LFR metric [looks at Lists and reduces for Popularity] we get a totally different result! With popularity diminished, we get a different view into who is looked to advice, ideas, opinions and expertise. Now Venessa is number one with an LFR [as of August 21, 2010] of 1679 [first four numbers after the decimal]. Valdis is second with an LFR of 1056 and Clay is last, for the first time, with an LFR of 0614.

Clay is popular, but Venessa and Valdis are listed as topic experts/advisors/influencers beyond their generic popularity.

Both Venessa and Valdis are more focused on topics they address and they attract a more focused, yet much smaller, audience. People put them on many Lists according to their speciality. People get to see Venessa's and Valdis' opinions and ideas — their following grows more slowly. It is an active choice by their followers — they know why they chose to follow/list either of these two. The decision to follow an already popular person is often a passive choice... "everyone else is doing it, they must know something I don't" goes the thinking. An active decision to listen/follow someone is probably a much stronger basis for the transfer of ideas and influence.

The LFR is also a good metric in uncovering spammers or tweeters pushing an agenda — whether it is ads or only their own content. The difference between Followers and Lists will result in a very low LFR, with two or more leading zeroes.

A very quick way to gauge your LFR is to drop the last digit from your followers [i.e. 1234 followers becomes 123] and compare that Follower proxy your Listed count [i.e. 321]. Is the new Follower proxy LESS THAN your Listed count? If so, you are probably more influential, than popular. Anyone can do this calculation while staring at a person's Twitter Home page and deciding whether to follow, or not.

Another quick way is to calculate a "batting average" like in American baseball. Take two people, Rita and Ralph, they each appear on 200 lists. Rita has 1000 followers while Ralph has 4000. Rita's batting average is 200/1000 or .200, while Ralph's is 200/4000 or .050. Rita get's a "hit" more often than Ralph.

There is a new Twitter metric that looks promising — PeerIndex. I know the person behind it, Azeem Azhar. He has been thinking about social networks since the mid 1990s. He and I had a wonderful discussion on this issue of Release 1.0 [large PDF] that I wrote about social networks in, and between, organizations in 1996. I like the way that PeerIndex divides one's influence by topic. Look at the "key authorities" in social media on PeerIndex — they are not the usual suspects that other Twitter metrics seem to echo. Azeem understands that in addition to the technology, he has to get the sociology right!

BTW, almost all of those listed as highly influential by PeerIndex for the topic "social media" [as of 8/21/2010] also score high in the BoE metric LFR — 4/5 or 80%, just like the Pareto Principal predicted!

Update: Here is another good use of lists! Create a Twitter List of those who have listed you and check that regularly for serendipitous opportunities! Hat tip to social media guru @arturs in Latvia.

Update #2: Further thinking on LFR and a simplification of the scale -- Who gets Attention on Twitter?  Here is a Twitter List of people with high LFR scores.