April 25, 2008

What's Your Twitter Noise Ratio?

The many thousands of people who use Twitter do so in wildly different ways. Some use it to deliver minute by minute updates of their daily activity. Others use it to hold conversations with friends and peers using the service. And still, a good percentage of people use Twitter as a broadcast medium to announce items, but not necessarily to engage. Meanwhile, as Twitter has grown, its not uncommon to see people either following, or being followed, by thousands of other users. Some do so reciprocally, while others are more discerning.

I feel there are different categories of Twitter users, from those who have a listening audience, measured by a high "followers" to "updates" ratio, those who are engaging, seen with near equal "followers" and "updates", and those who are more noisy, with a lot more "updates" than actual "followers".

Taking a look at 48 Twitter users I either follow or engage with, I found the average number of "tweets" per "follower" was almost exactly 1, measuring at 1.02. But the ratio of updates to followers varied widely, from the sleepy 0.06 (@om) to the firehose-like 9.75 (@corvida). And surprisingly, those Twitterers best known for creating a lot of noise, like Robert Scoble and Jason Calacanis, were quite in line with their number of followers, measuring in with ratios of 0.50 and 0.18 respectively, making their perceived noise to be in fact, a consequence of their engagement.

Download the Microsoft Excel data file

One of the informal guidelines I've used since opening my Twitter account a little over three months ago was to maintain an updates/followers ratio of less than one. I feel if I "tweet" too often, those following will opt out or gain in annoyance. As of today, my ratio is at 0.49, with 318 updates for 644 followers, putting me on the quiet side in comparison to the others I looked at.

A Twitter "Noise" Chart for 48 Users

(Click for Larger Image)

Of note, this was done by hand, via Excel, without fancy algorithms, so it can be assumed to recognize a point in time from Friday, April 25th.

Twitter's Listeners (Ratio of Updates to Followers of Less than 1)

Twitter's Middle Ground (Ratio of Updates to Followers of 1 to 2.0)

Twitter's Conversationalists (Ratio of Updates to Followers of 2.0 to 5.0)

Twitter's Megaphones (Ratio of Updates to Followers of more than 5.0)

This is, of course, a simplistic analysis of a select number of Twitter users. An argument could be made that those with thousands of updates are flat-out noisy, regardless of how many followers they have, but I also believe that being selective in one's tweeting habits can lead to an increasing audience for further conversations. If there's an imbalance between how often somebody is tweeting and how many people are choosing to follow them, it could be the noise has grown too loud.

Have any better examples of odd ratios between total number of Twitter updates and total Twitter followers? With thousands and thousands of users, there's no way this 48-person list gets everybody. What's your Twitter noise ratio?


  1. Hi Louis - I was thinking this would be about noise (links) to conversation - not messages per follower. I'm not sure that metric matters. I think personal links/total messages is a better indication of noise. For example, Calacanis, Arrington and Cashmore seem to have very high noise levels. Calacanis has said that twitter is his marketing vehicle for mahalo and Mike uses it nearly exclusively to push TC stories. Not suggesting there is anything wrong with it, you know what you are getting when you subscribe to either of them.

    The way you have done the math, the more followers you have, the more tweets you can send without being noisey? That seems odd. Because clearly anyone with 10k plus followers will get more room to breathe when we should all be on the same playing field. Noise is noise whether one person hears it or 5 million.

    Thanks for putting this together in any event - it's a great conversation starter.

  2. Allen, links vs. conversations would be a good metric. I'd probably do much worse there, thanks to posting blog links with high frequency.

    If somebody can run that data, I'd like to see it.

    The math I did does give more leeway to those with more followers, and what should be here (but I didn't do yet) is posts per those they are following. In theory if they have 10,000 followers and have one note to each, that's 10,000 updates. If I have 10,000 updates to my 600 or so, that's 16 per person.

    So what does somebody with a very high tweets to followers ratio (5.0+ in this case) mean to you?

  3. Hi Louis - I agree with the logic of your broad categorization of Twitters, but think your attempt to look at data and slot people into the appropriate categories fails.
    You shouldn't really be using the "noise" metaphor.
    The meaningful relationship is not between # of posts and # of followers, since the number of followers one has is a result of variables like fame and the quality of your Tweets, that is to say more (or less) tweets may or may not lead to more (or less) followers.
    Hmm. Maybe I'm getting tangled here. Let me try and lay it out:
    Listeners - would be a low tweet, high follower ratio
    Balanced - would be someone who's tweets contain an approximately equal blend of simple statements (or links or status updates or whatever) and those employing @- i.e. conversations.
    Megaphones - are your broadcasters, people whose pronouncement tweets make up a disproportionate amount of all their tweets. Additionally their "@" tweets will be small in comparison to the number of people they follow - indicating a relatively low level of engagement with their followers.
    This puts the Scobles and Calcanis's of the world right where they belong - they're broadcasters, using Twitter as a straight forward microblogging to subscribers tool. (Although Robert engages in conversations more frequently).
    Noise in this model would be a function of quality - something that resists easy quantification.
    At leasts, that's how I see it...

  4. louis,

    i think that one characteristic that you might want to cover is that the individuals you mention that you are on par with, scobler, techcrunch and the like are already established in the community.

    some of us are trying to use twitter as a faster way to get our thoughts out. For instance my ratio (@notronwest) is higher (+8 to 1) because i don't have the legions of blog readers you have. most of your twitter followers (myself included) like the "insider" nature of following established bloggers.

    i guess you didn't say that it was necessarily bad to have higher ratios but i can see why some people do.

  5. Good piece, I enjoyed it. I don't think it quite covers what you're aiming for - I have put a much longer post on my thoughts over here:


  6. This is why tracking in GTalk works so well to cut the noise. Track the terms you really care about -- like your name, town, (for me 'earthquake' is also a tracked term), site,, etc. and don't send follower's updates to IM, or only send specific ones.

    I use Twhirl at points where I want to see what everyone I follow is saying, when I'm at a point where it's okay to be distracted.

    It is really quite efficient, if inelegant.

  7. Messages per follower isn't a very good metric because it doesn't factor time into it.

    Average tweets per day for the last month would be a much more interesting statistic.

  8. I'm a multimegaphone, I guess, with 6,896 tweets and only 433 followers, yet I still have people who continue to follow me despite my multimegaphone-like tendencies.

    A more interesting statistic, but much more difficult to calculate, could be achieved by dividing tweets into three categories:

    Tweets which include "http" but do not include "@" (broadcasts)

    Tweets which do not include "http" or "@" (general conversations)

    Tweets which include "@" (specific conversations)

  9. I'm happy to be in the conversationalists arena myself :)

    However, on that note, I've just started using...as of right now...tweetlater's auto-follow/auto-welcome script. I'm going to see how that goes. Might pick up a few bots, but I'd like to have more people to engage with so this should help. And if they're following me, shouldn't I see what they have to say too?

  10. I'm not sure what this ratio tells us, but it's interesting! And I seem to be smack in the middle... which is good, I guess?

    I think the numbers as well as ratio of friends / followers + number of posts tells you quite a bit: how much the person participates, how active they are in building a presence, how "popular" they are, etc. I think Twitter does a particularly great job of letting you get a very quick read on a profile -- 1-2 second scan tells you quite a lot.

  11. Maybe I'm missing something here, but I don't see the value in a metric based on # of tweets / # of followers. "Noise" is very subjective. Something that is noisy to one person won't necessarily be noisy to another. One example would be using Tweets to promote blog posts. Some may find this noisy since they may already be subscribed to an RSS feed and don't want to see it sent as a tweet. While others are fine being notified of items by whatever means available to them.

    You point out that Twitter is used by people in wildly different ways. I'm in total agreement. Here'some of which you eluded to:

    - a meetup tool for parties
    - a promotional vehicle for driving traffic to blogs
    - a way to engage followers for feedback
    - replying to questions tweets posed by followers
    - for sending out photos
    - & a multitude of other ways including many which are driven by external Twitter apps

    use of each of these methods will require a totally different frequency of tweets to accomplish their goals and none of them are necessarily tied to # of follwers.

    Anyways, the point I'm trying to make is that Twitter is a very complex and organic environment with so many variables to consider that there is no way to provide a simple metric to globally label people as noisy. Especially when it's hard to define noise.

  12. @Mark, @Ron and @EngTech, there are undoubtedly many different ways to calculate "noise". A number of ways are subjective. Was a tweet "useful"? Did it hype a post or a comment somewhere? Did it add to a conversation?

    What I was trying to get across here were essentially two things:

    1) By following and being followed by many people, it is expected that your volume in total tweets could go up, just due to natural conversation, which in a vacuum could look "noisy".

    2) That some people, even without this established Twitter network, make a lot of tweets, and they use it more differently than those who do so less.

    It's just one way of measurement. If it's not the end-all, be-all measure for noise, we could simply call it (Twitter Updates/Followers) or TUF. :-)

  13. Suppose you post a lot of tweets that are only read by a few select friends. That makes you a megaphone, according to this analysis. Why is that bad?

  14. @agalvin, I tried not to say what was "good" vs. "bad". Your example is absolutely valid.

  15. Wow, It's probably true that those who talk less write even more...It's also true those who write and employ scripts do neither... Thanx for putting my small name in your BIG blog...

  16. Interesting subjects.Maybe more intriguing one is update/reply ratio.

  17. It's an interesting study.

    But it seems inherently flawed. If you use your calculations, with nearly 9k tweets and only 1.2k followers, I'm "noisy" - but when you look over time, when I was twittering the "average" number or fewer tweets per day, I had far fewer followers.
    As my volume of tweeting has increased, so has my following list.

    The third vector of time is completely left out here.

    Scoble et al have been on Twitter how long? How long since Calacanis went on his 'add me win a macbook air' campaign to increase follower numbers?

    It's fun to look at, but without factoring in time - and growth or loss of followers in relation to increase or decrease of tweets, it's pretty much irrelevant.

  18. I'm with GeekMommy down the line; entertaining, but otherwise completely useless. Louis is a self professed master baiter though it looks like TechMeme hasn't yet taken the bait. But the weekend is young!

    When people unfollow someone due to noise, what % of the time is that choice directly made because of the number of followers the person had? To the degree someone winds up doing a lot of @replies because they have a lot of followers, I can see that being an issue. Also, it could be that people with 20K followers feel obligated to broadcast more.

    As for true noise, it would be more relevant to use the same people and come up with some kind of average tweets per day. It would be somewhat interesting to see if those with more followers do tweet more on average.

  19. Those that have critical mass elsewhere skew the results tremendously, as does automatic following and competitions.

    Some people all so go out of their way to add 1000s of people and get autofollowed back.

    That being said, if someone has 1000s of tweets and very few followers, it is an indication that they aren't tweeting interesting content, though that is not a hard and fast rule.

    My ratio is around 0.25 but then I have the benefit of a fair sized audience.

  20. One person has 5 friends following because their friends were trying to convince them to use twitter.

    They tweet once to say "how do I use this?"

    Ratio of 0.2.


    You would have to add one follower per tweet to keep a 1.0 ratio.

  21. @louisgray,

    First off, it was great to meet you at Web2Expo!

    Second, BTW I am @elliottng NOT @elliotng (though I've squatted that account too). Flattered to be included.

    Third, I'm a data junkie so found this super interesting. BTW did you read Moneyball? probably cause you're a baseball and stats fan.

    I think Tweets/Follower is not a right stat and would drive the wrong behavior. (Totally understand that its the only easily stat available and already appreciate the hard work of putting it into the spreadsheet)

    Problems (some mentioned above)
    1. factor of longevity not considered
    2. updates include @ tweets. So in fact if you have some good relationships your "noise" ratio would go up even though you were really directing the tweet to 1 person.
    3. some of the "listeners" actually don't follow many people. e.g. @om and @techcrunch follow a small % of their followers.

    Wrong behavior if I optimize for the "noise ratio":

    1. if I want to tweet I NEED FOLLOWERS AT ALL COST! As time passes and I continue to tweet, I need to KEEP ADDING FOLLOWERS. (plug: follow @elliottng and I'll follow you back.)
    You probably don't want to drive that.

    2. I self-censor on direct @ tweets. In fact, replying to @ tweets is the best way to build relationships.

    What are the right metrics in a dream world then:
    1. "tweets/month" - or "raw author contribution" (see Avinash Kaushik's awesome blog metrics) - # of tweets/month. More is not better but this is a good metric.
    2. "follower growth rate" - are more people following or are people unfollowing?
    3. "followed/follower ratio" - I find it fascinating the differences. Look at @aglick35 who is following a massive # of people but have 1/4 the people following him. Look at @ev who has massive followers but really doesn't follow many people (and didn't respond to my @ tweets either).

    Mash these 3 metrics into a 3 dimensional cube and now you've got some interesting segments! BTW I DID NOT volunteer to do this data collection LOL.

    Thanks for another interesting post Louis. And frankly I'd like it if you tweeted more. You have permission to Tweet-promote your blog posts to me because I live in Twitter and no longer in FriendFeed, Google Reader or even Real Life. :)

    @elliottng (who needs more followers so he tweet more!)

  22. Elliott, I updated your Twitter link. Sorry about the typo. You make some good points, and others are as well. Looks like most don't agree with the Updates/Followers ratio which I'd been tracking, but we're always eager to hear more ideas.

  23. Great conversation -- but I don't get the metric. I'm w/ others here on importance of a longitudinal measure. But more to the point IMHO wld be to distinguish signal/noise ratio. Reason being that treating posts as noise doesn't make sense.

    Using the reputation score approach at tweeterboard, for example, we might use @ replies and @ citations as a way of distinguishing signal -- because both of those show that the post has been picked up.

    The way twitter works there's too much bias in number of followers -- w/o mutually reciprocated friending followers we can't assume that # followers = attentive audience. Followers/following enhances a twitterer's presence: in fact it's best used as a presence or potential reach metric (hence it's popularity in measuring influence, tho that's problematic too).

    Distinguishing signal/noise ratio would make more sense because it indicates the audience that's tuned in. And signal reach, that is a post that is cited by several, would indicate pass along/word of mouth, and would say something about the reach of a twitterer's influence and attention level of and engagement with audience. But to treat all posts as noise makes little sense to me...

    tweeterboard's reputation score is interesting in this respect because it measures reciprocity. Again, it's compromised by fact that @ name posts are often missed if the followers aren't mutually connected.

    Post number seems to indicate noise level or volume. (spinal tap moment -- 11 tweets/day = taphead?) So what if a person talks a lot? We want to know if they're being listened to -- and that would be signal strength. As measured by who's tuning in as well as who's rebroadcasting.

    But on twitter there's always the problem of "if a tree falls in the forest"... What's noise if nobody hears it?

  24. ps -- for more on this:


  25. I was playing around with something similar taking into account the value of tweets you receive which I realize is rather subjective but in some respects a better measure. You're method assumes all posts are noise and non are signal. The Twitter Equation is an article I wrote on the subject a few weeks back.