I don't understand those charts at all. Some charts show growth (right side of line is higher than left side of line) but the growth figure is negative. Some charts show decline, but the growth figure is highly positive. I have no idea what those charts are supposed to be illustrating - seems to be random noise.
I believe it's all constant growth, spike aren't indicating the growth, just a spike in growth. If that makes any sense. So if it's flat for example it's 100/day, if it's going up it's probably a spike over an average day, say 200/day for that day. I think the give away is that people don't seem to unfollow on twitter.
The charts are showing growth over time, as in delta change on any given day from the previous day.
Imagine you had a normal chart where each point is the absolute number of followers, but you take the derivative of that function (or use the deltas between days instead).
The data is only being displayed for the past 30 days. Keep in mind, just because it's flat doesn't mean it's 0. It could be constant at a higher number of growth. My apologies for not showing that effectively.
Thanks for posting this. I've been trying to get an idea of just how useful/helpful YC is in creating a successful company but evidence in of this impact in terms of social media has been hard to come by.
any other sources you'd suggest?
I guess it depends on how you define a successful company. In the capitalist sense, something like http://yclist.com/ is probably most poignant.
I wouldn't say that YC has a huge impact in terms of social media; they definitely help with getting press mentions, but getting engaged subscribers is really up to the founders.
Twitter is not relevant to all companies, but it's obvious that many are leveraging it very effectively. I wonder if investors would be interested in a dashboard like this for all their portfolio companies. The correlation to "real growth" is probably small, but still better than nothing.
Pet peeve on sparklines: they're clean because they get rid of chartjunk, but they're also pretty uninformative without context. Right now, they add almost nothing to the presentation ("@dropbox hasn't changed <much> since <some time ago>, but this other guy's popularity is <swinging wildly/pretty small and noisy>").
Tufte originally intended them to be inserted into text and contextualized by it. As an extension, he suggested adding small indicators of scale and position as colored dots for example.
Either that or just lose them and graph the relative results of each of the contenders. It all depends on what kind of story you want to tell.
Edit: just noticed that you mention that this information is tracked over the last 30 days, but since that's not localized to the graphs, it's pretty easy to not notice it.
Fair point. I couldn't come up with a better way to display 57 charts on one page which give some kind of interesting information without it being a total mess.
The information is actually tracked indefinitely from when I started monitoring it, and I have daily changes for more than 30 days for each of these accounts at this point. I also have an interface for browsing each one of them in a more granular fashion so that the chartjunk is visible (I can give you access to this if you're interested). Here's Posterous for example: http://goo.gl/QSGp4
Do you have any ideas for how to better present such a comparison between so many data sources? Keeping in mind that scale varies hugely between the smallest and largest.
Edit: To me, the most interesting thing to look at in this showcase is the "shape" of the sparkline (which indicates stability) combined with the percent delta change next to it (click on it to get absolute numbers).
I'd merge the chart and the current followers count. Even after you click the delta to find its absolute count, it's difficult to interpret it (to me). Place a green dot at the end of the chart and a red at the start and then color code two numbers to correspond. It'll give a sense of scale and variation that's currently missing. It'll also immediately suggest a linear model for followers over the last X days which might be a good summary for some names.
I'm not sure what to draw from the scale-independent representation you've got right now since I can't tell if wide swings in the sparkline indicate something really changed in way people follow that name or if it's just noise. This is a perfect opportunity for some sort of random process model which could be used to suggest that certain spikes (such as the recent one for @reddit) are maybe more interesting that real random variation.
I'd also look for ways to investigate and highlight weird behaviors like @greplin's bimodalism. That a pretty huge.
I'm not sure I understand how the expanded interface matches to the sparklines, actually. Posterous' doesn't match up with the sparkline much at all that I can see. Are you doing linear detrending?
I suppose as always there's no magic bullet for information presentation. There are any number of questions I think you could ask of a data set like this (stability, relative growth, comparison with other metrics like investment or publicity, looking for spikes).
I imagine that comparing each different company against the others would probably be not terribly useful since they're all at different scales, but I'd be very interested in things like how well data from one source could be predicted from all the others (which just starts out as computing correlation between them all) which might help you to separate out whole market trends from successes from each particular company.
The Tuftean method, which I support, still desires a complete, untransformed view of the raw data. So don't remove the sparklines. Just make them more interpretable by giving scale to the shape using further real data. Any of these cross-company models can potentially be added as additional information to the sparklines. Charts remain interesting and interpretable so long as they're drawn mostly in data-ink and are hierarchically readable.
You should check out Tufte's personally-curated forum (he and his staff carefully filter out only the best posts to be made visible) topic on Sparklines: http://www.edwardtufte.com/bboard/q-and-a-fetch-msg?msg_id=0... There is some good discussion there.
we were able to get ours after registering the trademark and then waiting for the twitter account to be dormant for 1 year. Luckily, no one was really using it. If it was not dormant, I suspect it would have been much harder if not impossible to get.
Also, API endpoints for friendship might have a follower_since date, and you can approximate the history. In so far as unfollowing is rare compared to following, it would be accurate.
Unfollowing is not as rare as you think. Out of all follows/unfollows SocialGrapple has tracked so far (for detailed accounts), 35% are unfollows of the subject account (63% unfollows from the subject account).
I did check the API, Twitter doesn't give you historical relationship data like that. This is something that SocialGrapple was built to solve.
Scraping is something I haven't considered, though being a tertiary source of data is something I'd like to avoid (I already have nightmares of Twitter shutting me down for no reason). I'm not even sure how accurate TwitterCounter's data is, at least I can account for my own and Twitter's mistakes. I'll give it some more thought nonetheless.
Interesting data, but unfortunately nothing you can see a trend in (other than the most well known companies have the most followers). One good advice you can take from this data: If you want to have many followers, you have to tweet more than once a day.
There's no specific point or statement, it's just "interesting data" that you can draw your own conclusions from.
For one, it's a nice overview of which YC companies are active on Twitter and which are not so much. Which are recently growing a lot, and which are stagnating.
It's also interesting to compare the number of followers to the number of following/tweets. We can see that Dropbox has very disproportional numbers, turns out this is because following @Dropbox on Twitter gives you a reward on their service.
Interesting how asymmetric the ratios are for almost all YC companies: they're happy to have people follow them, but not in general interested in following back.
