Watching others play on Twitch
It’s a slow Wednesday afternoon in a time zone when most of the US is still asleep. Still, hundreds of thousands of spectators are glued to their screen watching League of Legends, Hearthstone and CS:GO. Video game spectatorship is a fascinating phenomena that has been around for quite some time, but in recent years taken a drastic leap forwards. While video game spectating initially took the form of people watching others play at arcades, this spectatorship has been taken to the mass market in the digital era of user generated content. Twitch, the largest video game streaming platform, launched in 2011 and has steadily grown to become one of the largest bandwidth hogging behemoths of Western internet culture.
Image courtesy of Diana Moon (https://www.flickr.com/photos/dianamoon/21117387213/).
As an emerging form of digital media, video game streaming is also highly relevant to us researchers, from many different perspectives. To build a better understanding of why people choose to spend their time watching others play video games, we set forth to quantify and attempt to predict the effect of these spectator motives. To do this we developed a survey scale based largely on Uses and Gratifications Theory (UGT). We piloted the initial survey and after fixing some minor inconveniences, launched it to the general public in the spring of 2015. We did not have access to absolute tracking of the survey, but estimate that the majority of our respondents came from Reddit, followed by Twitter, Facebook and some individual discussion forums. As a participatory incentive, we offered a raffle of six games (worth $50 each) from the Steam store. After removing suspected trolls and other problems from the dataset, we arrived at 1097 respondents. As part of this survey, we also collected data on eSport consumption motivations, which we have written about previously on this blog.
We measured our results through five main types of UGT motivations: affective, cognitive, tension release, personal integrative and social integrative. As our main goal was to investigate the predictive power of these motivations, we studied their correlation with four types of usage. The first, and most important, was hours watched (per week). This was quite a clear metric that showed us the raw usage of the service, something quite easily comparable to studies within other forms of media. The following two were streamers followed (total) and streamers watched (per week). Spectators may follow streamers to get notifications from them when they are broadcasting, serving as a favorites list of sorts. The number of individual streamers per week also served as a metric showing us if a spectator was more invested in just a few select streamers or if they browsed around and watched many streamers per week. The final type of usage was streamers subscribed to. Users of Twitch may pay $5 per month to subscribe to an individual streamer, with the user getting a list of benefits in return (emotes, subscriber status, access to subscriber-only content). This was an important type of usage to measure, as it is one of the few ways to spend money on video game streaming. The other notable type being direct donations to streamers, also an interesting phenomenon in itself.
Image courtesy of artubr (https://www.flickr.com/photos/48006568@N05/11345856073/).
As mentioned, we looked at how the various types of motivations could help us predict usage of the service, through the four different types of usage described above. The results are presented in both table and graph form below.
How to read these results: For people driven by a particular type of motivation (for example drama), an increase of drama by X will increase the predicted behavior (consumption frequency) in relation to X. So for example, a β-value of 0.50 would mean that for each 1 unit drama increases, spectating increases by 0.50 units. Effect sizes are commonly considered small when around 0.1, medium when around 0.3 and large when over 0.5.
Note: * indicates a result with p < 0.05, ** p < 0.01, *** p < 0.001
If the table is cut off scroll to the right to see more.
|Hours watched||Streamers watched||Streamers followed||Subscriber|
|Affective||0.144**||0.001||0.058 – 0.224||0.134**||0.002||0.048 – 0.216||0.152***||0.000||0.068 – 0.235||0.045||0.298||-0.041 – 0.130|
|Cognitive||0.089**||0.007||0.023 – 0.155||0.075*||0.037||0.007 – 0.147||0.007||0.851||-0.062 – 0.074||-0.028||0.453||-0.099 – 0.042|
|Personal integrative||-0.177***||0.000||-0.248 – -0.107||-0.105**||0.008||-0.180 – -0.030||0.091*||0.021||0.016 – 0.169||0.033||0.407||-0.044 – 0.112|
|Social integrative||0.132**||0.004||0.047 – 0.220||0.120*||0.020||0.021 – 0.222||0.213***||0.000||0.126 – 0.302||0.150**||0.002||0.055 – 0.242|
|Tension release||0.319***||0.000||0.240 – 0.400||0.217***||0.000||0.128 – 0.300||0.080*||0.048||-0.001 – 0.153||-0.001||0.984||-0.095 – 0.093|
As we can see from the results, we discovered quite a lot of relations that were statistically significant. In this post I will cover some of these results, but as going in depth with every single one would make this post way too long, that is what the forthcoming article is for.
Noteworthy was the difficultness to predict subscriber behavior, as displayed by our low R2 value (=0.037). Additionally, only one type of motivation showed a significant correlation with subscribing, and that was social interaction. That this was the single motivation that counted when it came to subscription was to be expected if you take a moment and think about the tangible benefits for subscribing. As most of the content itself is freely available, choosing to subscribe is more akin to choosing to belong to the cool kids club. As subscribers get access to exclusive emoticons and a subscriber icon next to their name in chat, it is a way for people to differentiate themselves from others and show that they are the real fans. Naturally, subscribing is not just about this, but the social aspect seems to be one of the more important factors.
For the full research paper, please visit the Social Sciences Research Network: http://ssrn.com/abstract=2779543.