Antiochus
Member
A few months some researchers in Austria did a study massively mining the entire Discogs database for music from 1955 to 2011 to see if there's any relationships between musical/instrumental complexity and the underlying sales for them. Here are some of the interesting tidbits they came up with:
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0115255
The last sentence is especially cogent. Granted the researchers were not able to cover additional musical territory after 2011 (one may guess their results may be even more reinforced), but in short, the larger the album sales of a particular style, the more correlated it is with simplicity of the music.
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0115255
The Discogs database is one of the largest online user-built music database specialized on music albums or discographies. Users can upload information about music albums. A group of moderators assures correctness of the information. Discogs is an open source database and publicly accessible via API or XML dump file released every month. We use the dump file of November 2011 containing more than 500,000 artist and more than 500,000 albums assigned to 374 styles. The data spans more than fifty years of music history, from 1955–2011. Discogs uses a music taxonomy based on two levels, music genres and styles. There are fifteen different genres, such as ‘rock’, ‘blues’, or ‘Latin’. On the second level genres are divided into styles, for instance ‘rock’ has 57 styles including ‘art rock’, ‘classic rock‘, ‘grunge’, etc. ‘Latin’ contains different music styles such as ‘cumbia’, ‘cubano’, ‘danzon’, etc. S1 Fig. in S1 File shows the histogram of the distribution of music styles per genres. For each music album we extract information on the instruments played by artists, the release date of the record, and the music genres and styles assigned to the album. The data is grouped into time windows of seven years, e.g. the last time-step contains data on albums released between 2004–2010, and so on. S1 Table in S1 File provides some descriptive statistics of the dataset.
To measure the average sales numbers of music styles we use a dataset that contains information on the Amazon SalesRank of music albums as of 2006 [27]. The Amazon SalesRank can be thought of as a ranking of all records by the time-span since an item last sold [28]. Albums in the Discogs dataset are assigned their Amazon SalesRank by matching album titles between the two datasets. As the Amazon SalesRank dataset only contains information on album titles, it was matched to entries in the Discogs dataset by choosing only albums whose title appears only once in both datasets.
In this work we quantify the variety and uniformity of music styles in terms of instrumentation that is typically used for their production. We employ a user-generated music taxonomy where albums are classified as belonging to one of fifteen different music genres that contain 374 different music styles as subcategories. Styles belonging to the same music genre are characterized by similar instrumentation, a fact that has already been exploited in the context of automatic genre detection [25]. We construct a similarity network of styles, whose branches are identified as music genres. We characterize the instrumentational complexity of each music style by its instrumentational variety and uniformity and show (i) that there is a remarkable relationship between instrumentational varieties and uniformities of music styles, (ii) that the instrumentational complexity of individual styles may exhibit dramatic changes across the past fifty years, and (iii) that these changes in instrumentational complexity are related to the typical sales numbers of the music style.
The relationship between instrumental variety and uniformity of styles is remarkably stable over time. Variety and uniformity have been computed for six time-windows of seven years, starting with t = 1969–1975. For each time period and show a negative relation in Fig. 4. Values of are normalized by to make them comparable across time. Although this relation is stable over time, the position of individual styles within the plane can change dramatically, as can be seen in the highlighted trajectories of several styles. The evolution of music styles is also shown in the S2 Fig. in S1 File where the trajectory of is shown for each style that ranks among the top 20 high instrumentational complexity styles. For example, the style ‘new wave’ sharply increased in complexity rapidly and was popular from the mid-70's to the mid-80's, after which it decreased again. Similar patterns of rise and fall in complexity are found for ‘disco’ and ‘synth-pop’ music. ‘Indie rock’ gained complexity steadily from the 60s to the 80s and remained on high complexity levels ever since. Styles losing instrumentational complexity over time include ‘soul’, ‘funk’, ‘classic rock’, and ‘jazz-funk’. However, other styles such as ‘folk’, ‘folk rock’, ‘folk world’, or ‘country music’ remain practically at the same level of complexity.
To understand the mechanisms leading to an increase or decrease in instrumentational variety and uniformity we compute the change in the number of albums for each style between two seven-year windows, = 1997–2003 and = 2004–2010. The change in number of albums is compared with changes in instrumentational complexity , see Fig. 5A. We find that increasing complexity is typically related to an increasing number of albums within that time-span with a correlation coefficient and p-value . This suggests that styles with increasing complexity attract an increasing number of artists that release albums within that style.
We found a negative correlation between variety and uniformity of music styles that was remarkably stable over the last fifty years. This finding reveals an intriguing relation between local and global properties of the music production network. Styles with low instrumentational variety are characterized by instruments that are typically associated with a large number of other styles. While the overall distribution of instrumentational complexity over music styles is robust, the complexity of individual styles showed dramatic changes during that period. Some styles like ‘new wave’ or ‘disco’ quickly climbed towards higher complexity and shortly afterwards fell back, other styles like ‘folk rock’ stayed highly complex over the entire time period. We finally showed that these changes in the instrumentational complexity of a style are typically linked to the sales numbers of the style and to how many artists the style attracts. As a style increases its number of albums, i.e. attracts a growing number of artists, its variety also increases. At the same time the style's uniformity becomes smaller, i.e. a unique stylistic and complex expression pattern emerges. Album sales numbers of a style, however, typically increase with decreasing complexity, see Fig. 5B. This can be interpreted as music becoming increasingly formulaic in terms of instrumentation under increasing sales numbers due to a tendency to popularize music styles with low variety and musicians with similar skills. Only a small number of styles in popular music manage to sustain a high level of instrumentational complexity over an extended period of time.
The last sentence is especially cogent. Granted the researchers were not able to cover additional musical territory after 2011 (one may guess their results may be even more reinforced), but in short, the larger the album sales of a particular style, the more correlated it is with simplicity of the music.