Redistributing the streaming wealth: Understanding the links between popularity, demand, and time

The purpose of this post (which will be released as a Grey Paper) is to further inform the ongoing discussion over how best to distribute the royalties from subscription-bases music services. Over the next few pages we are simply going to think and type out loud, in an effort to understand how popularity, demand, and time (i.e., listener hours) interact to determine whether and how different payouts methods — in this case, “total pool” versus “per user” methods for distributing royalties — might lead to different outcomes for different populations of artists.

Rather than advocate that either approach to royalty payouts is somehow fairer than the other we are simply going to take a calculating look at a series of scenarios that present intentional variations in artist popularity, demand, and user listening hours — in the same way we might setup a series of interactions for an experiment. Then, we are going to calculate and compare the distribution of royalty payout resulting from these varied scenarios under both the total pool and per user payout methods.

What we are going to see is that diversity in listener hours, mixed with that in demand, become the driving forces for any differences in payout outcomes only when those hours and demand are systematically linked with artist popularity. In other words, we will see differences in payouts when comparing the total pool and per user methods only when users who listen to greater amounts of music also listen to a different population of artists than that population listened to by those users who listen to lesser amounts of music.

Therein lays the rub, as far as expectations for the fair and the unfair from any switch in royalty payout method:

If the fans who listen to the greater amounts of music each month also listen to the most popular artists, then it is the less popular — in fact the least popular artists — who benefit most from a switch towards a model that pays royalties on a per-user as opposed to a total pool basis.

If, however, the users who listen to lesser amounts of music each month happen to listen to the most popular artists — a usage pattern various sources of listening data suggest — then, perhaps non-intuitively, it is these popular artists who benefit most from any switch towards a model that pays royalties on a per-user rather than total pool basis.

Furthermore, the so-called “middle class” in any scenario seems to gain the least from any shift between payout methods. Instead, the extremes in the distribution — the most and least popular or demanded artists — see the most notable gains or losses.

Beyond the plusses and the minuses, I hope we are about to see how either of these approaches to royalty payouts can lead to seemingly fair or unfair outcomes. And so, while much of the debate over the distribution of streaming royalties has focused upon the apparent winners and losers, it may well be time to begin to speak more openly about our objectives for these services, and just how we might best play the payout game given these objectives.

The Setup

Before we go any further, lets clarify these two methods for royalty payouts as well as define, at least for the purposes of this work, what we mean by popularity, demand, and listening hours.

The “total pool” approach to royalty payouts aggregates all of the dollars paid by subscribers into one big pool of money, and then pays royalties to each artist/label based upon that artist’s/label’s share of all tracks played across all users. In contrast, the “per user” approach to payouts pays royalties from each, individual user’s subscription dollars to any artist/label based upon that artist’s/label’s share of each, individual user’s listened tracks.

Importantly, both payout approaches involve a “pro rata” (i.e., “in proportion”) consideration. The total pool approach simply considers that proportion of the pool of dollars and plays from the entire population of users, while the per user approach considers that proportion of the dollars and plays from each, individual user account.

As far as the three terms that matter most to this analysis:

Popularity is a measure of the proportion of user playlists in which any artist’s music might be found. Highly popular artists would be found in a greater proportion of user playlists, while lowly popular artists would be found in lesser proportions of user playlists.

Demand is a measure of the proportion of the total tracks played, whether on a per-user or across-all-users basis. Music from high demand artists occupies a greater proportion of total tracks played, while music from low demand artists occupies a lesser proportion of the total tracks played.

Note that the definitions above mean that we can have very popular artists who are in low demand as well lowly popular artists with quite high demand, especially on a per user basis.

Listening Hours is a measure of the time any user spends listening to music. It should come as no surprise, however, that the greater the amount of time people spend listening to music the greater number of tracks they are likely to play.

Each of the following scenarios will be setup as follows:

A set of Subscribers (labeled Q through Z) listens to music from a set of Artists (labeled A through K). You might think of these Subscribers or these Artists as individuals or, ideally, as representative groups — such as deciles from some larger distribution. On average, Subscribers will spend about 111 minutes listening to music each day. The total pool of royalties available will be $7 per Subscriber. As a result, the music will play and the chips will fall where they may.

The Scenarios

Each scenario on the following pages will introduce a different interaction of Popularity, Demand, and Listening Hours, by creating variation both within and among these variables across a set of Subscribers who listen to music created by a set of Artists. Each scenario will also result two distributions of royalty payouts to these Artists, one distribution resulting from a Total Pool approach and the other resulting from a Per User approach to the calculation of these payouts.

The purpose of the effort here is not to advocate for any particular approach — Total Pool or Per User — but to tease out the underlying factors that drive any difference in the distribution of payouts between the Total Pool and Per User methods.

After that, the choice is yours.

Scenario One: You have to start somewhere

Any scenario-based approach to understanding has to start somewhere. Here, we start with the simplest combination of the variables in question: No variation at all. Each Subscriber listens to the same amount of music (900 plays), and each Artist is not only equally popular (all Subscribers listen to each Artist), but also equally in demand (i.e., each Artist sees the same 90 plays from each of the Subscribers, for 900 plays in total).

Perhaps as no surpise, therefore, the difference in royalty payout outcomes for the artists in involved does not vary whether we use the Total Pool or Per User approach. Under the Total Pool method, each Artist has 10% of the total pool of plays, and is rewarded with 10% of the royalty pool. Under the Per User Approach, each Artist has 10% of the plays of each Subscriber, leading to a net 10% of the total payouts.

As we said, you have to start somewhere. Sort of like a mic check, now we know this thing is on.

Scenario Two: Equal Popularity, Varied Demand, Equal Listening Hours

In this second scenario, we will simply introduce some variation in Demand — the proportion of plays any Artist earns from Each Subscriber, as well as across all plays. Beyond that, each Artist will be equally as popular, and each Subscriber will spend an equal amount of time listening to music.

Here we find similarly reassuring results. Artist K, the most in-demand Artists in the crew receives not only the largest payout regardless of payout calculation method, but also there is not difference between these two payouts — whether in raw dollars, or general proportions of the total royalties paid. Artist A, while equally popular is not so in demand, and, therefore, receives the smallest share of the royalty dollars, regardless of method.

And so, simply enjoying a greater number of track plays as compared the plays of others is not enough to trigger a difference in payout between the Total Pool and Per User methods.

Scenario Three: Equal Popularity, Varied Demand, Varied Listening Hours

In this scenario, we are going to mix variation in demand with variation in listening hours, while keeping all Artists equally popular. Essentially, while all Subscribers will prefer some Artists over others, some Subscribers will also spend more time listening to music each week, leading to a greater number of tracks played across all Artists.

What we see when we turn to the payouts may be unexpected: even when we introduce an extreme variation in listening hours (e.g., Subscriber Z spends almost 100 times the, um, time that Subscribe Q spends listening to music), the royalty payouts across the two methods do not differ — whether in dollar amounts or relative percentages.

Apparently, the combination of demand and listening hours is not enough to shift apart the payouts from these approaches. We are going to have to alter popularity.

Scenario Four: Slightly Varied Popularity, Varied Demand, Varied Listening Hours

In our fourth scenario we are going to begin to mix variation in popularity with that in listening hours and demand. Two of the Artists, played by only nine of the Subscribers, will be only slightly less popular than the remaining eight Artists. However, zeros matter.

Turning to the payouts, we begin to see some deviation between the Total Pool and Per User methods. While slight in dollar terms, these differences are measurable in percentage terms. Artist K (the most in demand among the less popular artists) earns 13.6% less under the Per User approach, Artist A (the least in the demand among the less popular artists) earns nearly 50% more, while the Artists in “the middle” each see a small bonus of 1.76%.

Interesting.

Scenario Five: Some Variation in Popularity and Distinct Populations, Varied Demand, Varied Listening Hours

In Scenario Five, we further twist the variation introduced in the prior scenario. We have variation in demand as well as listening hours. Not to mention, we have some variation in Popularity, with some Artists having a five, six, and seven Subscribers and on with eight Subscribers. Importantly, however, the Subscriber populations who select these Artists are different: those Subscribers who spend the least amount of time listening to music each month select different artists from those Subscribers who listen to the greatest amounts of music.

When we turn to the payouts we see more clearly how listening hours, demand, and popularity interact in order to drive an difference in outcomes between the Total Pool and Per User payout methods. Artist K, the most demanded artists from the voracious listeners, sees a 54% decline in payout from the Per User approach, while Artist A, the least demanded artist by the least voracious listeners sees a 1000% increase in payout.

Whoa, that’s a redistribution of dollars.

However, all Artists aren’t equally popular in the music world in which we actually live. Not to mention, we still have to pursue additional scenarios in order to dig deeper into what might seem fair or unfair.

In the next post, we are going to introduce variation in Popularity across the Artists so that we can see how the interaction of all of these variable can lead to outcomes quite different from what might be expected.