Top Streamers on Twitch: Analysis of Success Factors
The streaming industry on platforms like Twitch has grown tremendously over the past ten years. By 2020, Twitch had become the go-to platform for streaming video games, and the next year, it enjoyed its largest spike in growth to date. The growth in viewership was mirrored by an increase in content creators, leading to a highly competitive market. Given this competitive landscape, it would be helpful for content creators to understand the key factors that contribute to a channel's success.
In this blog post, I will explore the factors that make a Twitch streamer successful and promote the growth of a channel. This information will be useful both for new streamers who are wondering what variables they should focus on to grow their channel efficiently and for established streamers who want to level up their channel.
The Data
This project uses data from two sources. The first is the Top Streamers on Twitch dataset from Kaggle, containing one year of data on the top 1000 Twitch streamers circa 2020. This dataset contains several key metrics to explore, including watch time, stream time, number of viewers, and number of followers. However, it does not include information on how much each channel earned during that period, which is why this project required the information from a second data source, as well.
The second data source is the Twitch Earnings Leaderboard. It shows the earnings of top Twitch streamers from August 2019 until October 2021. Although the earnings leaderboard is not a perfect match to the Kaggle dataset, there is a large overlap in the time period between the two datasets. Given that earnings are such a critical component of a channel's success, combining both datasets enables us to extract more significant insights than we can get from a single dataset. The Twitch earnings leaderboard was not available for direct download, so I scraped the information on the web page using a Python script. In the end, I matched earnings data for 788 of the original 1000 channels in the Kaggle dataset.
Earnings Distribution
First, I wanted to explore the distribution of earnings. As you can see, the distribution is highly skewed, although earnings are roughly normally distributed on a log scale. The median earnings of a channel were $263,500.
Factors Influencing Earnings
There were five numeric variables in the Kaggle dataset that I thought could all potentially positively influence earnings: watch time, stream time, number of followers, number of average viewers per stream, and peak viewers. Out of these five variables, watch time had the strongest correlation with earnings (rs = 0.47), followed by the number of followers (rs = 0.34) and the number of average viewers (rs = 0.32). Stream time had the weakest correlation (rs = 0.13). The correlation between watch time and earnings is moderate, while the correlations between earnings, followers, and average viewers are weak but positive. Perhaps unsurprisingly, popular channels with high watch time, viewers, and followers tend to earn more.
Visualizing the scatter plot of the three strongest correlations, we can see that the relationship between earnings with watch time, number of followers, and number of average viewers appears to be linear on a log scale. Given that the correlations are of weak to moderate strength, it seems that no single factor has a disproportionate influence on earnings, and it’s likely that a combination of many variables is at play.
Next, I wanted to visualize the relationship between earnings differently. The number of average viewers per stream is likely of particular interest to advertisers who aim to reach the largest audience possible when partnering with a channel. Below, I calculated the average earnings of channels classified into five categorical buckets. We can see that as the average viewers per stream increases, the average earnings also increase. Channels with >10,000 average viewers per stream earn over twice as much as channels with 5,000-10,000 average viewers per stream.
Similarly, I calculated the average earnings of channels according to five categorical buckets of stream time minutes. We can see that average earnings increase as stream time increases from 5,000 to 50,000 minutes per year. Interestingly, streaming over 50,000 minutes per year does not appear to influence earnings further. This suggests that the return on investment (ROI) is greatest up to 50,000 minutes a year. Beyond that point, though, quality rather than quantity becomes more critical to maximizing earnings.
Partnered vs. Non-Partnered Channels
Being a partnered channel with Twitch comes with many benefits, including the ability for viewers to subscribe to your channel, share in ad revenue, and other improvements that lead to a better experience for viewers and streamers alike. Overall, these benefits appear to lead to much higher earnings for partnered channels. The only caveat is that the number of non-partnered channels in this dataset was small (N = 7). Interestingly, the 7 non-partnered channels do meet the minimum criteria to become partnered based on their viewers and stream time, so that is not the reason for the discrepancy in earnings.
Mature vs. Non-Mature Channels
There was not a big difference in earnings between channels with mature vs. non-mature content. The median earnings and interquartile ranges between the two categories were very similar.
The Language Factor
English was by far the most popular language among the channels. Boasting 485, it was far ahead of the 77 claimed by the second-most popular language, Korean.
Russian and Spanish language channels had the most average viewers per channel, with English third. Spanish channels had the largest growth in terms of followers gained, with Turkish and Portuguese channels in second and third place.
The top three average earnings per channel were for English, German, and Italian channels, respectively, while the top three earnings per viewer per channel were for Italian, English, and German channels, respectively. Interestingly, while Russian channels were the most popular in terms of viewers per channel, they were only ninth in terms of earnings per channel. It seems like English, German, and Italian streamers are better able to monetize their channels, despite having fewer average viewers than Russian and Spanish channels.
Exploring Growth Opportunities
There are two channel growth metrics in the dataset: the number of followers gained and the number of viewers gained. Stream time was not correlated with viewers gained (rs = 0.03) and was negatively correlated with followers gained, though only weakly (rs = -0.32). It is unclear why stream time would be negatively related to followers gained. At a minimum, we can conclude that increasing stream time is not associated with channel growth.
On the other hand, the number of viewers gained was positively correlated with watch time (rs = 0.73) and the number of average viewers per stream (rs = 0.58). Watch time and number of average viewers were also positively correlated with each other (rs = 0.65). Watch time is potentially a proxy for the quality of the content or how engaging it is, with higher quality engaging content leading to higher watch time. In turn, higher watch time is associated with more average viewers, and both high watch time and average viewers are associated with more viewers gained.
Exploring how quickly channels grow, the number of followers gained per hour was positively correlated with average viewers (rs = 0.69) and total followers (rs = 0.63). Combined, these findings suggest that popular channels (that is, those with high numbers of viewers and followers) gain new followers more quickly. There seems to be a positive feedback loop where popular channels become more popular more quickly!
Conclusions and Recommendations
You can find success in different languages: English and German had the highest average earnings per channel, while Italian had the highest earnings per viewer. All three languages are great options for channels looking to maximize earnings. Spanish channels were also popular in terms of average viewers per channel and had the highest follower growth of any language, making it a great language for streamers looking to reach a large audience and grow their channel quickly.
What to focus on: Twitch streamers should focus their efforts on maximizing engagement to increase watch time, viewers, and followers. All of these had the strongest correlations with earnings and channel growth.
Quality over quantity: The quality of the content matters more than the quantity of content. Stream time was not strongly correlated with earnings or channel growth. However, new streamers should aim for at least 20,000-50,000 minutes per year because streaming too little appears to be detrimental to earnings.
Don’t go it alone: It’s worthwhile to get partnered. Being partnered provides more revenue opportunities and a better user experience, which, in turn, translates into higher average earnings.
Future Work
Future work should include some formal statistical modeling to determine the relative predictive power of each metric in the dataset for total earnings. It would also be great to obtain two variables missing from the Kaggle dataset that would enhance our analyses: chat metrics and genre. Chat metrics would allow us to explore message frequency, sentiment, and viewer interaction rates to understand engagement beyond just watch time and how that relates to earnings. Exploring how the genre of different games, as well as non-video game streams, relate to earnings could also shed more light on the key factors of success.
Quick Links
Main Photo by Rahul Mishra on Unsplash
About Author
