For thirty-two years, a group of researchers at Indiana University led by Kate Hevner Mueller and John Henry Mueller collected data for each performance by a major American Orchestra from 1842 through the 1969/1970 season. In addition to recording the year of each performance, the Muellers also noted each piece’s length and the composer’s nationality. The creation of this repertoire index, which was done by hand, required tracking down orchestra programs from ensemble archives, public libraries, and collectors across the country.
The resulting dataset covers the establishment of orchestral institutions in the United States and over 100 years of music-making. In total, the dataset covers the performances of twenty-seven American major orchestras (defined as a symphony with expenditures of more than $500,000 in 1972): Atlanta, Baltimore, Boston, Buffalo, Chicago, Cleveland, Cincinnati, Dallas, Denver, Detroit, Houston, Kansas City, Los Angeles, Milwaukee, Minneapolis, Indianapolis, New Orleans, New York (Philharmonic and Symphony), Philadelphia, Pittsburgh, Rochester, St. Louis, San Francisco, Seattle, Utah, and Washington D.C.
For this lab, I wanted to ask questions about trends in orchestral programming: Which composers were most frequently performed? How does composer popularity change over time? Can these changes be contextualized with larger cultural trends?
Inspiration
In her 1973 report on the repertoire list, Kate Hevner Mueller included several visualizations of the dataset. However, Meuller’s visualization options were limited since the data had not yet been made machine-readable and needed to be compatible with black-and-white printing. Nevertheless, her work provides a valuable starting point for creating modern, information-rich visualizations.
I took Mueller’s visualization of the “Most Played Composers” as my inspiration for this lab. This graph charts the proportionate representation of the top six performed composers by the amount of playing time. The data is binned by five-year periods and normalized to the percent of playing time in that period.
Within the context of her limitations, I think that this visualization gets a lot right. By normalizing the representation of composers to percent-per-year, you can see trends in the programming of popular composers. Additionally, when charted together, you can see their relative popularity compared to one another. However, there is some crucial information missing when the chart is isolated, such as the fact that the data is binned and the representation is calculated by playing time.
My goal for this lab was to build upon the best aspects of Mueller’s original visualization, using Tableau to develop charts tracking trends in popular composers that are engaging, information-rich, visually pleasing, and encourage further exploration.
Pre-Processing Data
As mentioned above, this dataset was originally published as a printed list of performances organized by composer and piece. Additional information included the piece’s run time and the composer’s nationality and birth/death years.
Sometime after the index was published, it was scanned and converted to machine-readable formats by Document Solutions Inc. in Oakland, CA. Notably, the nationality and orchestra names (which are frequently repeated) were stored as codes (e.g. 1–27) – likely to reduce file size. To properly work with this data in Tableau, this required replacing the values of the codes with their string representations. This was done using Python and pandas.
Visualization
One of the first choices I had to make was the period of the data to visualize. As time went on, the number of performances documented increased dramatically, and I did not want to over-represent composers in periods with little data. While I tried out several options – namely, the last 100 years of data (1870-1970) and data after 1900 – I ended up using Mueller’s date range, 1890–1970 for consistency. When looking at a line graph of the number of performances per year, this seems like a natural breakpoint.
In order to think about how to iterate upon Mueller’s visualization from above, I found it valuable to first recreate her chart in Tableau.
While this chart provides a solid overview of trends, there are aspects of the dataset that it obscures. First, by binning data over five-year periods, some of the natural variation is lost. This is valuable for seeing long-term trends, but there are sometimes interesting fluctuations within single-year periods. Second, I wanted to explore ways to show more than six composers at once.
To build upon these ideas, the first chart I created was a stacked area chart representing the proportion of all composers per year by playing time. When filtered down to the top ten composers, this chart gives us a striking visual representation of the proportion of playing time. Compared to representing proportion with a line chart, the stacked area chart gives a stronger visualization of each area as a proportion of the whole – it is not just an abstract percentage. When comparing these proportions to the “other” category, we can better see how limited in scope most orchestral programming is. These ten composers dominate the stage. As Prof. Sula also noted, the fluctuations of the area chart also mimic the look of sound waves, which provides a nice visual motif for a visualization on the topic of music.
Another interesting aspect of the area chart is that by displaying all of the yearly data together, your eye is drawn to outliers and anomalies in the data. For example, following you can see a sharp decline in the performance of Wagner – a composer associated with German nationalism and later appropriated by the Nazis – around World War I (1914–1918) and World War II (1939–1945). While this is visible in the first graph, the effect that this decline has on the other areas makes it particularly visible in this visualization.
In fact, if we create a similar chart by nationality instead of composer, we can see that there was a sizable decline in the performance of all German composers during this period.
Reflections and Future Directions
One of the downsides of the approach that I’ve taken is its focus on the “most performed” composers. In many ways, we expect their performance to be relatively stable. One area of interest for Mueller is the full lifecycle of composers, where they are introduced, reach peak popularity, and then decline. By combining all of the less-performed composers into an “other” category, information about the other nearly 2,000 composers is excluded. Importantly, this includes individuals who do not fall within the dominate archetype of the European, male composer.
While some information will always have to be left out to create comprehendible visualizations, I would like to explore options for surfacing the trends of other less-performed composers. This led me to feature the visualization at the top of this post, which does not create an “other” category. While it is hard to extract overall trends from the chart, my finding the label of an individual composer, their general placement from the top or bottom of the graph provides a general estimation of their ranking during this period.
The topic of expanding orchestral programming beyond this cannon of primarily European composers was taken up at length by Kremp (2010), who uses the Mueller dataset for statistical analysis of the programming habits of individual orchestras. One downside to this approach is that it continues to obfuscate composer-level trends.
In the future, I would be interested in developing a dashboard that allowed users to explore data for specific composers or groups of composers. “Small multiples” visualizations could potentially lend themselves well to this approach, as they allow the trends of several composers to be compared without needing to plot them on the same graph. They also have an additional benefit over stacked area charts, as they don’t have the same challenges with estimating the exact proportion of an individual group.
Sources
Kremp, P.-A. (2010). Innovation and Selection: Symphony Orchestras and the Construction of the Musical Canon in the United States (1879-1959). Social Forces, 88(3), 1051–1082. https://doi.org/10.1353/sof.0.0314
Mueller, K. H. (2015). American Symphony Orchestra Repertoires 1842-1970 [United States]: Version 1 [Dataset]. ICPSR – Interuniversity Consortium for Political and Social Research. https://doi.org/10.3886/ICPSR35235.V1