Introducing The ReTweetability Metric





A topic of much recent discussion has been how do you quantify influence in social media and social networks. In the context of Twitter, much ado has been made of purely network-size-based statistics, which are easily gamed and present a shallow picture of the process of viral influence.

Extending from the work I did with the ReTweet Mapper, I’ve been exploring more intelligent metric to analyze the influence wielded by a user’s Tweets. My data has shown me that while the actual size of a user’s follower network has a positive correlation to the amount of ReTweets they get, the relationship is actually rather weak. This tells me that the actual content of a user’s Tweets may be more important to how influential that user is.

In an attempt to algorithmically study how “viral” someone’s Tweets are, I propose the following ReTweetability Metric:

This is designed to control for both the rate of Tweets the user posts and the number of followers the user has, so that this metric represents soley how ReTweetable a user’s posts are. This formula typically yields very small results so for the purposes of readability I’ve taken to multiplying it by a large constant, 1,000,000.

To demonstrate how this metric works, I’ve calculated it for the top 100 most ReTweeted users in my system which you can see below. This list is sorted by most ReTweeted, as I don’t yet have enough users’ ReTweetability metrics calculated to produce a “Most ReTweetable” list.


User ReTweetability Metric
guykawasaki 2.13909
BreakingNewsOn 2.10939
problogger 3.25695
mashable 2.45021
timoreilly 6.27744
chrisbrogan 0.382216
TechCrunch 1.71331
kevinrose 2.32571
StatSheet 1515.92
Scobleizer 0.592066
Armano 2.27349
JesseNewhart 4.07655
chrispirillo 1.85232
nytimes 0.583496
zaibatsu 0.853129
steverubel 4.18765
mayhemstudios 1.25115
codinghorror 4.69129
unmarketing 0.657318
TheOnion 7.05509
Twitter_Tips 6.81798
domestic_diva 25.6448
skydiver 5.37113
wilw 1.14165
PRsarahevans 1.15549
shortyawards 0.0536032
zappos 5.63933
twitter 14.5021
Pistachio 0.922368
davewiner 1.37601
levarburton 5.05578
Foodimentary 1.71968
tweetbomb 13.6751
themediaisdying 3.83852
perrybelcher 1.14216
AJGaza 6.19664
stephenfry 2.44443
jayrosen_nyu 5.08907
tinybuddha 98.526
lancearmstrong 1.41447
jeanlucr 4.10289
BreakingNewz 1.53622
laughingsquid 1.81378
caseywright 5.34765
mbites 4.81602
fatwallet 155.291
CNETNews 1.73988
weirdnews 3.32362
MacHeist 386.29
pleasedressme 89.7901
gapingvoid 2.33999
tferriss 16.5756
styletime 1.56265
cnn 1.10834
shelisrael 0.848424
garyvee 0.733047
BILL_ROMANOS 3.17027
darthvader 50.3311
howardlindzon 2.34331
shanselman 3.302
ColonelTribune 5.27153
redstarvip 2.33445
JasonCalacanis 0.393586
smashingmag 3.75457
ev 2.4533
barefoot_exec 0.288966
boris 15.4311
jakrose 1.02494
LeoLaporte 1.25961
manifestmmind 12.7619
mattcutts 4.91364
imjustcreative 0.891586
jowyang 0.774668
MrTweet 4.1668
sugree 1.00299
danzarrella 4.24575
OwenC 46.3504
Andrew303 10.4188
jemimakiss 3.27651
BrentSpiner 2.62709
zen_habits 13.1589
1938media 2.93775
MariSmith 0.40963
HubSpot 47.0784
cnnbrk 2.68889
copyblogger 0.807263
MarketingProfs 0.565924
Suntimes 4.33059
UstreamTV 3.67757
tamar 22.9639
BertDecker 10.9617
rww 4.83897
feliciaday 4.0208
TwitPic 26.2796
QassamCount 24.4141
gazanews 11.8672
secrettweet 0.485707
Positive_Thinkr 17.6236


If you liked this post, don't forget to subscribe to my RSS feed or my email newsletter so you never miss the science.

{ 6 comments }

Warren Sukernek January 23, 2009 at 5:10 pm

Wow, great concept! An excellent way to normalize things and focus on the actual content. Can you create a table of the top Retweetability Metrics of all people with over 1000 followers or some other reasonable number?

Dave January 23, 2009 at 5:14 pm

I think you formed/explained your ratio in a confusing way. Try something like this:

(# of retweets) / (Possible Retweets)

(Possible Retweets) = (tweets in a day) * (followers)

Then, you can more clearly state that you are effectively measuring the percentage of retweets out of retweets possible. I would leave “average” out of the formula, since it’s possible (but difficult) to gather exact data (i.e., the exact number of followers at the time of each tweet), but state that in reality it is much easier and mostly accurate (except in cases of major changes in number of followers) to use averages.

Dave January 23, 2009 at 5:18 pm

It’s also disorienting that you chose to multiply the results by one million, somewhat arbitrarily. Maybe you can show the results as a percentage (something like “.015%” for Fat Wallet) or flip the ratio (1 in 6440 opportunities results in a ReTweet, again for Fat Wallet). Keeping the numbers representative of some real world value will make them much more usable and easy to understand the importance.

Stefanie January 23, 2009 at 8:26 pm

This is a great concept, eventhough I do agree that the additions dave made improve the whole calculation. Still think it is only an indicator, and the solution has to involve more variables. However, I haven’t seen a better solution so far.

Jim Lane January 26, 2009 at 11:48 pm

With the focus being on numbers, this measurement measures popularity and misses the quality factor.

Mere retweeting is one thing, but a retweet by a quality person is much different than a retweet by a spammer.

There is also the whole issue of the quality of what is being retweeted. And, then the quality of who the retweet goes to.

As a popularity metric, perhaps this works.

Quantity is always easier to understand and measure than quality. Quality calls for judgment.

jim

Bosilytics January 27, 2009 at 7:59 am

Very nice. Thank you for doing the legwork in identifing the ratio of what is starting to be the closest to a conversion in Twitter we have so far -a “RT”. Or at least one measurable by the masses.

I do agree with Dave, the presentation is much easier to comprehend, or at least makes nice since when quering results:
COUNT(like *RT*@bosilytics*) / (Opportunities)
WHERE (Opportunities) = (tweets in a day) * (followers)

Again, brilliant due to the fact that it is easily calculated, has a fair amount of normalization and after a couple trial runs I think it is something to act upon (or at least seriously consider after a little more digging).

Couple notes:
1) I think a higher value should be added to a users 2nd, and outer social cloud. While being retweeted by a follower is always great, getting RT’ed outside the cloud gets exponentially more exposure.
Problem: presently, this is hard to measure without a nice tool budget.

2) I do think the name Opportunities is more telling. THis is also more scalable as a scoring method is derived

Jim: While you will have your spammers and their network of spammers that can easily throw off this number, the shere nature of friend/folllow weeds these guys/gals out anyway. That said, until monetary value is put on a RT, why cheat?

Again, very nice. Now, who is going to make the ReTweetability calculator usign the API? …only wish it could be I.

{ 3 trackbacks }