Thursday, June 23, 2011

Tit for Tat Tweeting -- Learn to be Generous

As always, these are my own opinions and musings and not official Intel positions or recommendations.

I love Mathematical Game Theory and one of my favorite areas is the "iterated prisoner's dilemma" which models the human behavior to cooperate and to cheat (defect). Variations on the game and strategies give us concrete ways of describing numerous social behaviors, including con-men, free-riders, and reputation based punishment systems.

The basic principle of the game is simple and played between two players, if both cooperate with each other, they win a reward. If only one of the players defects, that player wins an even better reward and the other player gets less of a reward. If both players defect, both players get a reduced reward. Thus, it is advantageous to a player, if the other player can be convinced to cooperate, but advantageous to themselves to defect. The game is designed so the advantage for both cooperating is the best long term reward

One effective strategy for the iterated version is called tit-for-tat. In this strategy, the player begins by cooperating as that gives the optimal payout for both players [summed together] over the long run. If the other player attempts to cheat and raise their score at the expense of the player by defecting, on the next round the player defects and punishes the other player. That gives a disincentive for the other player to defect, because the other player knows they will be punished for defecting on the round after they defect. If the other player returns to cooperating, the player stops punishing them. Each time the other player defects (tats), the player responds by punishing them (tits) and thus the name of the strategy.

Twitter as a form of word-of-mouth communication is a cooperative system. If you like something that someone else tweets, you can amplify that message by adding your own voice to the message. The normal way of doing this in twitter is called a retweet (RT). The idea is that some people follow what you are saying who might not have followed (or just not have seen) the message the original author wrote. Thus, by retweeting the message, you are emphasizing it.

As a cooperative system, twitter can be described as a gift economy. In a gift economy, a positive form of tit-for-tat is a very effective strategy. It is often called "give to get" or quid pro quo. In give to get, you perform a favor (cooperate) for someone in hopes that they will return it. In twitter, this can be retweeting someone's message, answering their question, or introducing them to someone who has similar interests. In return, you hope they will do a similar favor for you.

Of course, one of the things about a gift economy is that it is based upon "gifts", things one gives freely away. That is a key aspect. If one gives with too much expectation of receiving a precise, fair, equivalent amount back, one will tend to be disappointed. Only by giving generously without keeping tally can one reap the benefit of the strategy.

The nice thing is that the mathematical model of game theory shows us just that effect. In games, where one of the players is playing tit-for-tat and the other has a propensity to cheat, the total score for both players will be lower than if the other player is totally generous.

Generosity is its own reward, especially in systems where cooperation plays off.

Sunday, June 5, 2011

Measuring Activity, Influence, and Klout

As always, these are my own thoughts on this topic and not official Intel positions. Intel does permit me to blog and tweet, but not to misrepresent that I speak for the company on any topic. In addition, Intel does measure the impact of our writings, but I have no control over how those measurements are taken nor what tools or sites are used to do so. These thoughts are based solely on my own personal use of the tools that I mention. I have no affiliation with any of these tool makers other than as a user. I have only used the free versions of these tools and not the paid versions. Finally, note I am not a professional statistician nor market researcher and you should do you own analysis before making any decisions.

Recently, I have read a fair amount of discussion on whether one's Klout is related to one's Social Media influence. You may want to read Online Influence Is More Than Just Social Media Activity and 5 Reasons Why Klout is Total BS. Both articles are fairly negative on calling what Klout measures "social media influence". While I have no specific allegiance to Klout's specific measure, I'd like to address some of the criticism.

To do so, I first need to relate a story. About 30 years ago I was in college math and computer science. At the same time my Father was in graduate school studying creative writing. To finance us he wrote some articles, one of which was about a hunger study. The statisticians who did the study wanted to identify the causes of hunger and they measured a bunch of variables they believed that were related to hunger (such as the number of low-birth weight babies) and then grouped the data into a matrix on which they performed a "principle components analysis". Out of that analysis came a list of numbers that were measures of how the data correlated. They labeled the most significant of those numbers "hunger". Now, if you accept their assumptions on whether those numbers were related to hunger and that hunger was the most important factor, then calling that number "hunger" is valid. However, if you don't, then it isn't.

Therein lies the point. Mathematics is like that. Everything is determined by the assumptions one makes and how one labels them. If one accepts the assumptions, the conclusions can be drawn (and proven). If one doesn't, the conclusions don't hold. The numbers Klout puts up are no different.

Social media influence is intangible. No one can actually measure it in a way that everyone agrees with. In fact, I think you would find it would be hard to point to an example of social media influence and get broad consensus as to whether it was or wasn't a good example. Thus, any measure of social media influence is likely to be controversial. A certain amount of this controversy can be applied to any definition of social media influence.

Moreover, there are many people who object to any measuring of these intangibles. The answer to them is simply that it is our labeling that makes these intangibles real. Many abstract and intangible terms are simply our agreed upon convention of what it means. The fact that they mean slightly different things to different people does not prevent us from using them. This holds true for social media influence. We don't need universal precise consensus to make it real. It is real and measurable because we choose to make it so.

The end result of this, if Klout measures as it claims to the number of users who take action after reading a tweet or status update, then it is measuring something related to social media influence. You might quibble about details, but in the end that is what you are doing, quibbling.

Social media influence is also ephemeral. Although a tweet or FaceBook status update may last indefinitely, its ability to cause action is distinctly temporary. I have seen some measurements of how long after a tweet and how long after a status update, users clicked on links. For tweets, the fall off of activity was measured in hours, with essentially no actions taken after a day. For status updates, the fall off of activity was measured in days with essentially no actions taken after a week. You can easily verify these measurements. The reason for this fall off is obvious, if you tweet something, it is a very short time before that information disappears off the recipients' timelines. This is particularly true for the most influential readers who are the ones who most likely have the most connections and thus the fastest moving timelines.

Given that fall of of activity rate, any time longer than a couple of weeks should capture nearly all activity related to a specific tweet or status update. Since Klout appears to use a 30 day window, that definitely captures responses to all recent activity. Moreover, if someone hasn't responded to something you have written in the past month, how influential can you be over that person who responds at most 12 times per year?

Another way of looking at the ephemeral nature of social media is to see the effect of absences. I had two recent ones that are pretty indicative.

Ouch, your Klout score has been falling lately. Share more content and engage with your network and your Klout score will rise!

Most recently I took a short trip that kept me off twitter for a few days and which resulted in me tweeting slightly less for about a week total including the time I didn't tweet at all. The result was pretty clear, not tweeting took me out of the consciousness of some of my followers. Even when I started tweeting again, it took awhile before I was getting responses. Some of the people I had been regularly conversing with me took awhile to return to conversing and some of them have not returned yet. The level of conversation grew as my number of tweets grew.

Before that for much of last year, I had a longer absence, where I was tweeting only sporadically, perhaps as infrequently as once a month. During that absence, only a few of my most committed followers kept in conversation with me. And, similarly, to what happened after the short break, only as I began to tweet more consistently and more intensely did people return to engaging with me. Moreover, some of the people with whom I conversed before that long break have never done so since.

So, while two incidents do not make a scientific survey, they do provide some evidence that taking a break from tweeting does reduce one's influence, at least if one equates influence with the ability to generate responses. It seems clear to me that consistent and frequent engagement are the keys to keeping one's audience's attention. Moreover, it seems quite likely that audience like trust is difficult to build and easy to lose.

Thus, if Klout really emphasizes one's level of activity, it is not clear that activity is not related to influence. Additionally, if Klout is actually able to measure response to that activity, then calling that measurement influence is not a major stretch of the term. It is fair to want to distinguish activity from influence, but it is not clear that one can divorce them entirely and if one wants to be influential one needs to participate in some activity that drives that influence.