How do videos go viral? How do people share them through social networks? And what are the dynamics of ‘virality’?
Following the success of our Gangnam Style vs. Harlem Shake study (May 2013), Francesco D’Orazio and I have been working with Twitter UK to explore four more big viral phenomena. The stories we selected have all been driven by video, and have been chosen to represent various types of video content:
- Commander Hadfield singing Bowie’s “Space Oddity” on the International Space Station (music)
- Dove “Real Beauty Sketches” for advertising (the most-watched advert EVER on YouTube)
- Ryan Gosling Won’t Eat His Cereal series of Vine videos, for serialised narrative content and mobile
- A grass-roots video of June’s protests in Izmir, Turkey, to provide an international and news dimension
Turns out there’s not a single model of virality. Instead, different types of videos spread in different ways. Different types of content appeal to different audiences and the structure of these audiences is what shapes the viral diffusion.
Understanding the dynamics of that spread – quantifying it using metrics, and digging into the influencers and demographics to understand some of the “how”, is what we’re going to talk about in this series of blog posts. But first, take a look at the diffusion maps below, which show the pattern of tweets and retweets for each video (click to embiggen).
(Blue nodes = tweeters. Yellow nodes = retweeters. Size = author visibility, i.e. estimated reach).
It’s immediately clear that there’s something different going on for each. Some, like Commander Hadfield, have one big hub (Hadfield himself) driving half or more of the sharing. Others like Dove Real Beauty and the Turkish protest video show a constellation of many smaller influencers, each being reblogged by smaller groups. Read on and we’ll explain why.
What we did
We used Pulsar’s content-tracking technology to collect and analyse any tweet containing a link to the videos we were tracking. What we’re analysing is content diffusion and content discovery – the way videos are shared, recommended, and retweeted until they become viral phenomena. Of course people share content in other ways too – not least on Facebook – and YouTube search is in fact the second biggest search engine in the world (after Google). But Twitter provides the strongest dataset for analysis, and its role as a “hub” for curating content from across the whole social web makes it an apt case study.
Metric #1: Size
At this point the results are straightforward – the Canadian astronaut wins, with an audience of 75,000 sharing his video on Twitter. Space oddity has a wide, even global appeal – in contrast to our smallest video, the Turkish protests, which was shared by just under 12,000 people on Twitter, very largely within Turkey.
What’s interesting is how this contrasts with the YouTube view counts, shown in the chart below. Commander Hadfield may have got more Twitter sharing – but Dove Real Beauty Sketches got nearly 4x as many YouTube views. Now, there are a number of reasons for this, not least sharing on platforms other than Twitter. Is the affirmative, personal message of the Dove advert something people prefer to share with their Facebook communities of “real world” friends and family? Whereas Hadfield has a science and news-y angle that is more suited to Twitter? Quite possibly.
But we also believe the different viral patterns of the two videos can explain this discrepancy. Read on for details…
Metric #2: Sharing over time
The following charts tell the story of how each video was shared over time. Note the similarities and differences: while all of them essentially have sharp “spikes” in sharing (it’s that buzz of thousands of shares per day that made them viral in the first place), there are different patterns.
So there seems to be two patterns:
- Two videos (Commander Hadfield and Turkish protests) peak immediately, in the first 24 hours from launch
- The Dove and Ryan Gosling videos, by contrast, show a more sustained level of buzz over 20+ days. There are still spikes, sure – but interest is much less ‘front-loaded’ than it is for Hadfield or Turkey.
Metric #3: Days to Peak
So we might bring in another metric, Days To Peak. How many days does it take each video to hit its maximum sharing rate? This splits our set of videos in two: two of them peak on launch day (first 24 hours), whereas the other two take several days to get to maximum velocity:
It’s worth “zooming in” on this peak day to understand this maximum rate of sharing better.
Metric #4: Velocity
The chart below shows the Twitter shares per hour for each video, and the results are really interesting: The 2 videos that peak on Day 1 (Hadfield & Turkish protests) don’t just peak on Day 1, they actually peak on Hour 1 or Hour 2. This shows just how “viral” this content really is – it gets thousands of people’s attention instantaneously, and is sufficiently powerful for them not just to watch the video, but for thousands of people to tweet and share it as fast as they can.
It’s also interesting that the Turkish protest video gained only about 6% of Commander Hadfield’s YouTube audience, but nonetheless reached the same peak sharing rate: 6,000 shares per hour. This shows how much of an impact timely, relevant news stories can have in a smaller community. So if you’re seeing what we’re seeing, there are now clearly two patterns:
- Spike: Where a video explodes into social with a big bang, getting attention immediately but then burning out quickly
- Growth: The slower-growing version of virality, where a video gets picked up by influencers and introduced into new communities over many days
This pattern for viral video is of course one we first observed in our Gangnam Style vs. Harlem Shake study back in May. The Gangnam phenomenon was a “grower” that kept running for over six months, whereas the more celeb-driven Harlem Shake showed a “spike” pattern, bursting up and then dying down again quickly. It’s all very well to describe these patterns visually, but what if we wanted to quantify this so we could compare it objectively? Introducing our next metric: variability.
Metric #5: Variability
This was a metric that took a bit of thought. We wanted to find a way to quantify the “spikiness” of our video distributions. Which ones have the most extreme spikes on their peak day? And which ones stay show a steadier pattern of interest?
Stats-heads among you will be familiar with “standard deviation”, the measure of how much deviation or “difference” there is within a series of numbers. Our variability metric is a normalised version of this: the coefficient of variation, aka the standard deviation of Twitter shares per day, divided by the mean (average) number of daily shares. This gives us a % figure.
The first thing to note is that all the videos show a high level of variability. We’ve seen this on the Twitter shares over time charts already – all of them have substantial ‘spikes’ in interest, of varying degrees of steepness. Social virality is never an entirely evenly-dispersed phenomenon – as we’d expect, for something traversing the power-law distribution of social influence.
For comparison, the O2 brand shows 71% variability in day-to-day discussion, and Tesco 47% (October 2013 figures). So you can see the Twitter diffusion of these videos was much “spikier” than typical topic buzz.
But looking at variability, we also see our “spike vs. growth” model confirmed. The two “spike” videos, Commander Hadfield and the Turkish protest both display around twice as much variability as the “grower” videos, Dove Real Beauty and Ryan Gosling. So variability’s a really useful measure for identifying which kind of virality you’ve got on your hands.
Metric #6: Retweetablity
Virality isn’t about people seeing things, it’s about people doing things – sharing. And on Twitter that can take two forms: original tweets sharing the video URL, and retweeting other people’s messages.
We find it interesting to examine the ratio between the two, as there’s a surprising amount of variation. The chart below shows how many retweets each video-sharing post got. You could call this the “engagement rate”, but to our mind it’s only one measure of engagement. People’s original tweets sharing a video are another form of engagement, one that might even be more valuable for a brand as it carries a greater sense of personal advocacy.
So there’s some variation: The Dove video generated the most original tweets, in about a 3:2 ratio with retweets. Examining the messages themselves, we see a lot of personal comments being added –
- “wow, wow, wow…so powerful.”
- “THIS IS A MUST WATCH FOR EVERY WOMAN… and every man who loves one :)”
- “This made me cry, as women we have to be more kind to ourselves.”
Commander Hadfield also generated more original messages than retweets (1 to 0.89). The tone of people’s messages was different to Dove however –compared to the sense of “I relate” generated by Dove, here people were simply awestruck by something “out of this world”
- “Chris Hadfield is the boss of bosses. Really interested in seeing what he does after the CSA”
- “@Cmdr_Hadfield gives an amazing look from space. This time, Bowies Space Oddity. Absolutely incredible.”
- “I’m aware I’ve linked this before. I do not fucking care. It is absolutely epic.”
But the real finding of our retweetability metric is just how retweetable the Ryan Gosling Vines were. They gained fully 4.3 times as many retweets as original posts. This isn’t about maintaining attribution to the author, as most retweets were of @TheFunnyVines, not creator Ryan McHenry. So what’s going on? Across a wide range of categories we see humour getting retweet rates an order of magnitude higher than other stories. Perhaps the impulse to retweet a joke is a fleeting one, making pressing “retweet” more appealing than copy-pasting and typing out for a original post. Alternatively, with Vines being a new-ish format, perhaps people may retweet as then they know that the Vine will show up properly.
Metric #7: Social Currency
The final dimension we want to talk about in this post is social currency. We define this as Twitter shares per million YouTube views. This isn’t a measure of popularity per se – that’d be the YouTube views total, which we’ve shown at the start of this blog post. Dove won that contest, with a massive 59 million views. Instead, social currency can measure the social value of a piece of content – how far people think it’s relevant it is to their friends & followers. And of course sharing is also a representation of self: people share content that makes them look good. We discuss this in more depth on our blog post Why We Share.
So what was the social currency of our viral videos? A surprise leader: the Turkish protest video, documenting police violence in Izmir. This achieved a massive 12,900 shares per 1m YouTube views, 2.5x the Commander Hadfield performance, and 11x Dove Real Beauty Sketches.
What do we think was going on? Well, unlike the other two, the Turkish protest video was news content. It showed protests and the state clampdown spreading from beyond Istanbul into other cities (Izmir), and potentially citizens with camera-phones were able to provide the first record this event before official news crews got there. So this story spread like wildfire within Turkey, hitting a velocity of 6,000 shares/hour.
By contrast, the Commander Hadfield and the Dove videos were both a little less urgent. They were both bigger overall – meaning there’s more chance your friends would already have seen them via other people. So that factor, overexposure, could dissuade sharing and reduce social currency. We also hypothesise that the Dove video may have gained more sharing on Facebook rather than Twitter, given its more personal message. (We would love to measure this too but with a large and unknown percentage of private data on Facebook it’s less easy to do so).
This is a big post with a lot of variables and data. So let’s recap on what we’re saying overall. How do viral videos spread socially?
We can see there are 2 broad patterns of content diffusion. One model we call “spike” – the sudden ‘explosion’ of sharing activity – and the other we call “growth”, where popularity is a slower and steadier grower. The metrics we’ve discussed, such as velocity, variability and social currency, provide a way to identify which kind of virality you’re looking at:
In our next blog post, Face CIO and Pulsar creator Francesco D’Orazio will talk about the people who made these videos go viral. Who were they – which demographics did each video reach? And how does content spread through online communities? Does “Spike” virality travel through communities differently to slower-burning “Growth” virality? Watch this space!
(Or, as a preview, watch our videos of how this content diffused through influencer hubs, over on the Twitter blog.)
How Videos Go Viral Part 2 on how audience networks shape viral dynamics is now published – read it here.
Found this interesting? Got viral content of your own that you want to understand? Check out the tool we used for this study, Pulsar (PulsarPlatform.com) and contact us to arrange a demo – send an email to James.Cuthbertson@Facegroup.com and we’ll be in touch in no time.