This article aims to provide information to streamers to aid them in the decision of whether or not streaming for 24 hours is worth the time and effort required. This article does not propose to be definitive (there is probably a Phd thesis in this subject), merely additional information to help a streamers decision. I would like to stress that, while I have experience working with large datasets I am not a statistician.
Streaming non stop for a 24 hour period has become more commonplace on Twitch, often a channel will use a 24 hour stream as a reward for viewers when certain milestones have been reached. Listening and reading comments by streamers, there is a noticeable trend, particularly among smaller channels to see a 24 hour stream as a way to grow their following. Comments from larger channels tend to suggest that their desire to do 24 streams decreases over time. It would be interesting to crunch the numbers to see if this is the case but it is beyond the scope of this article.
Stream data was collected using the Twitch API between the start of August 2015 and the end of January 2016. Not all streams had data collected, coverage in August/September was around 60-80% of streams rising to 90%+ from October onwards. Generally speaking, if a stream had more than one viewer it will been included in the collected data. The streams viewership and channels following were recorded four times an hour during the course of the stream.
The data required sanitization in order to be useful, channels with very unusual follower activity were removed. For example, a channel with < 100 followers which gained more 1000 follow in a single hour but nothing during any other hour were removed. Channels with very large follower drops were also excluded (one channel has 70,000 followers removed in a single hour). As this article is interested in individual streamers, large event channels were also removed (for example Bob Ross, tournament channels etc.).
For the purposes of this article any stream which was 23.5 hours or longer will be classed as a 24 hour stream (this is due to the nature of the data collection, to ensure that a stream which is exactly 24 hours is included). Between August 1st 2015 and January 31st there were at least 17,797 streams which were 24 hours or longer in length.
References below to "stream" will be streams that are at least 23.5 hours long.
Yes, the majority (79.18%) of streams gained at least one follower during the course of the stream, a significant percentage (17.58%) saw no change in followers and a small portion of streams lost followers (3.24%).
This isn’t particular surprising as the longer a channel is live, the more likely they are to be found and followed. The number of channels which are prepared to stream for no follower gain is interesting but again not surprising given the passion of the community for Twitch & Gaming.
The above chart is for the entirety of Twitch, Twitch however has a massive spread of a channel sizes, from no following to millions. So the question is, does the size of a channel change the probability that it will experience growth during the course of a 24 hour stream?
Streams were grouped by channel size, with the first group being between 0 and 5 followers, then each subsequent group doubled the follower range.
Its evident that the size of a channel has an impact on the probability that the channel will gain followers during the course of a 24 hour stream. For smaller channels (those 10 or fewer followers), the majority of channels will not see any change in followership during the course of a 24 hour stream. Again this isn’t particularly surprising as a smaller channel will have less exposure which will decrease the chance of a viewer finding the channel and following. Conversely a smaller channel is less likely to loose followers when compared to larger channels (with the exception of the largest channels) because they do not have a large follower base to begin with.
By the time a channel reaches 320+ followers, over 90% of those who streamed for 24 hours or longer gained followers during the course of the stream.
There is a noticeable increase in channels which lost viewers in the mid range of channel sizes (between 5121 and 40,960 followers) with 5.27% and 6.88% of streams losing followers. If I was to speculate as to the cause, one possible explanation would be that channels in this size range are big enough that they will gain a portion of followers by accounts that Twitch will eventually remove for whatever reason (abuse, view botting etc.) but not large enough that their gains will always be larger than this removal, unlike the larger channels. I am most defiantly not suggesting that channels in this range are more likely to have view botters and an abusive community, merely that channels within this range maybe suffer more adverse effect when compared to smaller or larger channels.
The following chart shows the average stream viewership during the course of a 24 hour stream (many streams exceed this length but for comparison purposes we'll restict the range to the 24 hours).
The raw numbers are not particularly interesting as they will be skewed by the largest channels but they do highlight the pattern of viewership during the course of the stream. Viewership steadily increases at the beginning of the stream, peaks around hours 4 to 10, declines until hour 16 and then holds steady until the last hour which sees a small spike in viewership.
This pattern can be explained by the initial hype of the stream, followed by viewers needing to go to bed/work etc and then jumping back in towards the end to see the streamer complete the 24 hours. I would have thought the drop in viewership towards the second half of the stream would have been larger given that Twitch peak viewership on Twitch is generally 2-3 times that lowest viewership during the course of a day. This could highlight that members of a streamers community wish to be part of the event, the “I was there” feeling.
Follower gain aligns pretty closely to viewership, the biggest difference being that follower gain peaks and declines earlier than peak viewership. The easiest explanation for this is when a viewer enters a channel they make the decision to follow pretty soon after entering (i.e. they do no wait several hours before deciding to follow).
For purposes of comparison, data in the following charts is normalised to a range of 0-100. I have also split the channel groups into four seperate charts so that they remain readable. I would urge caution with the larger groups (16-18) as the sample size is relatively small (less than 100 streams for these groups).
Channels of all size follow the same starting patterning – they all see large gains in viewership in the first three hours of the steam. Smaller channels (< 80 followers) and the larger channels (10,241+), follow a pattern similar to the average, large gains in viewership at the beginning, a peak followed by a decline in viewership over time and perhaps a spike at the end.
What’s really interesting are channels in the 80-10,241 follower range, again they see rapid growth within the first few hours of the steam. However, after this, these channels (on average) do not see the same decline in viewership that the smallest and largest channels see, rather they see a slow but steady increase in viewership for the duration of the stream.
Does this mean this channels within this range should start doing 24 hour streams all the time to gain viewership? Probably not, if I was to suggest a reason for this it would be to do with positioning in whatever game they are playing. Channels of this size will most likely be pushed down the page by larger channels but have enough of a viewership to still beat the vast majority of other channels. As these larger channels go offline, some of their viewers will look for streamers playing the same game, will see the mid range channel and watch, leading to small increases in viewership over time.
Larger channels will already be at the top and the smallest channels will still be too far down the page to get noticed. It would be really interesting to calculate this correlation to see if this is true or not!Channel viewership and growth during 24 hour streams (Part 2)