Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- I was involved in some discussion concerning Youtube's new platform today, Youtube Red. The basic premise is that you can disable ads and instead support Youtube and creators through a monthly fee of $10. Now, the way this money gets distributed, amongst other things, has become cause of controversy and since it sounded like something simple but interesting I decided ... to do some math. And why not publish it as well? At the end of this post I'll also discuss some more, well, "rhetorical" thoughts, you could say.
- Let's first define some parameters here:
- m is the amount of viewers on Youtube Red.
- n is the amount of channels.
- f is the portion of the subscription fee ($10) which is directly sent to creators, be it by first collecting all these portions or by distributing this portion to some channels right away. This should be 55% of the total fee, however due to some vague wording it'll probably be a bit less. The value doesn't affect my calculations though.
- w(i, j) is the watch time (unit doesn't matter) viewer i spends on channel j.
- r(j) is the revenue channel j earns.
- I was told the way the system works is roughly like this: all fees are collected into a giant pool and then everyone gets a fraction corresponding to the fraction of watch time their channel gathered (from Youtube Red users only, of course) compared to the total watch time. I'll call this model 1. If this isn't true, most of the math is kind of irrelevant, though it's still fun hypothetical reasoning. People however seemed to be in favor of model 2: a model where every fee is distributed based on that individual's watch time and distributed to the channels this individual watches accordingly. Let's convert this to formal mathematical expressions:
- Model 1:
- r(k) = pool*(total_channel_watch_time(k)/total_youtube_watch_time)
- = m*f*(total_channel_watch_time(k)/total_youtube_watch_time)
- = m*f*(total_channel_watch_time(k)/sum(j = 1, n, total_channel_watch_time(j)))
- = m*f*(sum(i = 1, m, w(i, k))/sum(j = 1, n, sum(i = 1, m, w(i, j))))
- Model 2:
- r(k) = sum(i = 1, m, user_contribution_to_channel_revenue(i, k))
- = sum(i = 1, m, f*(channel_watch_time(i, k)/youtube_watch_time(i)))
- = sum(i = 1, m, f*w(i, k)/sum(j = 1, n, w(i, j)))
- I felt like these two systems would be equivalent, which is why I went on and looked into it. Turns out my intuition was completely wrong; I overlooked the fact that active viewers have more influence on the final distribution in model 1 than inactive viewers, even though both pay the same amount. A quick example to prove the difference:
- User 1 watches channel A for 0 hours a month, watches 100 hours a month in total.
- User 2 watches channel A for 10 hours a month, watches 50 hours a month in total.
- User 3 watches channel A for 15 hours a month, watches 150 hours a month in total.
- Using model 1, we add everything up: 3*f to distribute, 25 hours watch time, 300 hours total watch time. That means channel A gets 1/12 of the 3*f, being f/4 or 0.25*f. Using model 2 however, each user individually pays the channels they watch, so channel A receives nothing from user 1, f/5 from user 2 and f/10 from user 3. This adds up to 0.3*f. In other words, because the active viewers don't watch channel A as much as the inactive viewer, they get more say. I will discuss the merits of both models at the end of this post, as I think there's some things to be considered behind the seemingly blatant unfairness of model 1. We have some more math to do first though.
- Alright, so sometimes you will gain/lose revenue due to the difference in between model 1 and 2. But when? Well, let's try to mess with our formulas here. In the case that you earn more through model 1, we can deduce that ...
- m*f*(sum(i = 1, m, w(i, k))/sum(j = 1, n, sum(i = 1, m, w(i, j)))) > sum(i = 1, m, fee*w(i, k)/sum(j = 1, n, w(i, j)))
- [move m*f/sum(j = 1, n, sum(i = 1, m, w(i, j))) to the right-hand side]
- <=> sum(i = 1, m, w(i, k)) > 1/(m*f) * sum(j = 1, n, sum(i = 1, m, w(i, j))) * sum(i = 1, m, f*w(i, k)/sum(j = 1, n, w(i, j)))
- [remove f in numerator and denominator on the right-hand side]
- <=> sum(i = 1, m, w(i, k)) > 1/m * sum(j = 1, n, sum(i = 1, m, w(i, j))) * sum(i = 1, m, w(i, k)/sum(j = 1, n, w(i, j)))
- [sum(j = 1, n, sum(i = 1, m, w(i, j)))/m is total watch time across the platform divided by users, so average watch time per viewer]
- <=> sum(i = 1, m, w(i, k)) > average_youtube_watch_time * sum(i = 1, m, w(i, k)/sum(j = 1, n, w(i, j)))
- <=> sum(i = 1, m, w(i, k)) > sum(i = 1, m, average_youtube_watch_time * w(i, k)/sum(j = 1, n, w(i, j)))
- [w(i, k)/sum(j = 1, n, w(i, j)) is how much watch time channel k receives from user i, relative to user's i Youtubew-wide watch time]
- <=> sum(i = 1, m, w(i, k)) > sum(i = 1, m, average_youtube_watch_time * relative_watch_time)
- That's basically where I got. So, if all Youtube Red users were to have the same total watch time, while keeping their preferences, and your channel's watch time drops because of that, you earn more with model 1 than model 2. The reverse can be proven in an analogue way. Yeah, it's not the kind of concise statement I wanted to reach, but it at least has some non-mathematical meaning.
- So long for math. I also said I had some opinionated stuff to talk about. This being why I can see reasons both for adapting model 1 and 2. On one hand, you may agree with this: you pay to take away ads from people's videos, so you reimburse them based on the ad revenue they'd lose otherwise. On the other hand, you may agree with this: videos with twice the watch time (loss of ad revenue) should get twice the reimbursement. You agreed with both? Then don't be angry at Youtube/people who support model 2 anymore, because that's kind of the reasoning behind model 2 and 1 respectively. And no, reason 2 does not apply to model 2, contrary to what you may think. To prove this, another example:
- User 1 watches video A for 10 minutes, video B for 5 minutes and other videos for 85 minutes.
- User 2 watches video A for 0 minutes, video B for 15 minutes and other videos for 285 minutes.
- Total watch time: 10 minutes (video A), 20 minutes (video B) and 370 minutes (other videos).
- In model 1, video A earns 1/40*2f = 0.05f and video B earns 2/40*2f = 0.1f, upholding condition from reason 2. In model 2, video A earns 1/10*f = 0.1f, video B earns 1/20*f + 1/20*f = 0.1f, not upholding the condition. Video B earns as much as a video with half the watch time simply because video B's viewer also watches a lot of other stuff.
- The solution? I don't think there is any. The two conditions I mentioned above are probably mutually exclusive ... if you add one assumption, which I didn't mention in this entire post. A constant fee. You see, I didn't do math here, but to me it seems all these bad representations are caused by the simple fact all Youtube Red users pay same monthly fee, yet some of them watch half as much as others and therefore wouldn't generate the same ad revenue on regular Youtube. If you let people pay in function of the ad revenue they have to compensate for, everything is fixed. However, then Youtube Red becomes some kind of bill you have to pay off, watching videos becomes costly, ... (and to be honest the fee would probably be way lower than the current $10 one, looking at CPM rates). From a business perspective, there'd be a lot of downsides to a variable fee.
- Got some thoughts? Discuss them here on Twitter: https://twitter.com/Wout123456/status/659548542789357568 Mind you, I'm not saying this makes Youtube Red all great or something. This is just a specific aspect of the system, which I discussed today, learned a bit about and thought had some objective and subjective theories surrounding it. There's plenty of other issues, like possible division of paid/free communities, Youtubers moving away from the regular website, ... However, there doesn't seem to be much math in these and since I don't have a strong opinion on them and there's already plenty of ranting by others about them, I didn't mention them here. If you feel like it though, discuss it on Twitter. Chances are I don't have some point of view with solid arguments for/against a given stance though.
- Also, as a final note, I didn't re-read this, so please inform me of any spelling/grammatical errors I've made.
Advertisement
Add Comment
Please, Sign In to add comment