Information Cascades: Music Streams and Downloads
Our discussion of information cascades centered around the principle that “when people are connected by a network, it becomes possible for them to influence each other’s behaviors and decisions” (Easley and Kleinberg, 2019). Individuals in a network usually have different types of information available to them when making decisions; the information they came up with on their own, or information based on other people’s choices. When people continuously decide to make their choices based on others’ previous choices, regardless of the information they have, this can lead to an information cascade or herding.
I have a friend who when faced with a choice between two products in a store, always chooses whichever has less stock on the shelves. She reasons that it must be low on stock because many people like the product, hence it is highly likely that it is a good product. The emptiness of the shelf likely started with one person who arbitrarily chose between the two products when they were both fully stocked. The next person might have also chosen arbitrarily but happened to choose the same one the first person did. The third person who comes will see that one product is down by two while the other is fully stocked, and they may reason that two people chose the same product over the other so it must be better. Everyone coming after that may also find themselves reasoning in a similar way, thus starting an information cascade using the signal of how many products have been taken. Of course, there are many other factors to consider but this is just one scenario that demonstrates how information cascades happen in networks.
In their paper titled “Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market”, Salganik, Dodds, and Watts outline their investigation of the paradox where experts fail to predict what media products will become hits even though the fact that hits are so much more successful than average implies that there are qualitative differences between them and the others (Salganik et. al, 2006). The authors examine how cultural markets are affected by social influence. An artificial music market was created with 48 songs and 14,341 participants. Participants were asked to listen to songs they had not heard before and then decide whether or not to download them. They were either part of a group where they decided either independently or with social influence. People in the latter group had information on previous participants’ decisions whilst those in the former did not. In the social influence groups, there were two versions; they were designed to include either a weak or a strong influence. In the weak influence version, songs were displayed with download counts but in a random order, while in the strong influence version, songs were displayed with ranked download counts from highest to lowest number of downloads. The authors’ results showed that factoring in social influence increased inequality and unpredictability of success. People were more likely to download songs that they saw had high download counts. This made it such that there were no particular types of songs that became successful in terms of download counts since the downloads were not based on song quality alone but on how much people were influenced by the information they now held.
This seemed very similar to the concept of information cascades we studied in class. In this real-world experiment, we had the chance to witness how herding may contribute to the success of a piece of media on the market. There are usually other factors that influence people’s decisions to stream a song like how familiar they are with the artists, and how popular it is within their social group amongst many others. In my opinion, I think these are also social influences and act as signals that give someone information on whether or not it will be a good idea to download a song and how well they may like it. Personally, when I like a song by an artiste I have not listened to before, I visit their Spotify page in search of similar music that I might like as well. Without much external information, my biggest signal of how well a song is liked is the number of streams as indicated by Spotify. If the streams are very high, or even just relatively higher than the others, I assume it is one of their best songs and I will give it a listen to determine if I like it myself or not. I can see how this may form a cascade, especially when it may not be an objectively high number of streams but just higher than their other stream counts. People visiting their page may use the same reasoning I do and give those songs a listen. Whether they end up liking it or not, it will still count as a stream, raising the number. If enough people do this, the number will keep going higher even though it may not necessarily have anything to do with the actual quality of the song, increasing the strength of the influence and subsequently the number of streams.
Works Cited:
Easley, David, and Jon Kleinberg. “Chapter 16: Information Cascades.” Networks, Crowds, and Markets: Reasoning about a Highly Connected World, Cambridge University Press, New York, 2019.
Salganik, Matthew J., et al. “Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market.” Science, vol. 311, no. 5762, 2006, pp. 854–856., https://doi.org/10.1126/science.1121066.