The Valuation of Songwriting Techniques:

An Analysis of How Song Elements Affect Song Value

by Frank Merlock

A thesis presented to the Honors College of Middle Tennessee State University in

partial fulfillment of the requirements for graduation from
the University Honors College

Spring 2020
 The Valuation of Songwriting Techniques:
An Analysis of How Song Elements Affect Song Value

by
Frank Merlock

APPROVED:

Dr. Anne Anderson

Weatherford Chair of Finance

Dr. Trevor de Clercq

Recording Industry

Dr. Philip Phillips, Associate Dean

University Honors College
 Acknowledgments

First, I would like to thank my family for their constant love and support. I would

like to thank Dr. Anderson for all of her fantastic guidance throughout this project. Dr.

Anderson was vital in helping me overcome several setbacks, as well as learning the

process of putting together an academic paper. I would also like to thank Dr. de Clercq

for pointing me in the right direction on a topic with which he is rather familiar. I would

like to thank my roommate Oscar Fernandez for listening to hundreds of songs with me

because he didn’t have a choice. I would also like to thank Cedric Gilmer, Zack Medic,

Maxwell Jamerson, and Stephen Borders for their friendship despite many hours working

on this project. I would like to thank Kyle Mackulak for telling me to cut the skits from

the introductions of my own songs. He was right all along. I would like to thank J.K.

Simmons for teaching me about keeping rhythm. I would also like to thank Austin

Rochez, Laura Les, Dylan Brady, David Duchovny and my cats for providing me endless

inspiration.

iii
 Table of Contents

Abbreviations & Definitions v

Tables & Figures viii

Abstract x

Background 1

Literature Review 9

Thesis Statement 12

Hypotheses 13

Methodology 16

Results 22

Discussion 33

Conclusion 36

Works Cited 37

Appendices 41

iv
 Abbreviations & Definitions

Buzz Angle Parameter Definitions

1. Audio Streams: Audio streams are through online interactive services (which allow

for customers to choose specific songs) like Spotify and Apple Music, as well as

video streams through services such as YouTube. According to BuzzAngle, “Streams

will count on a 1:1 basis on all streams charts. There will not be any weighting for

any of the various types of streams: on-demand audio, on-demand video,

programmed, ad-supported or subscription streams" (“Methodology” 2020).

Concerning video streams, BuzzAngle states that, "Music Video rankings reflect the

sales of full-length music video projects, be they documentaries, clip compilations.

Music-intensive films will be eligible in cases where digital and/or physical track

such titles as music videos, but some that are released theatrically will not be

reported” (“Methodology” 2020).

2. Song Sales: Song sales are actual sales through online services such as iTunes or

Google Music. According to BuzzAngle, “Songs that are priced under $0.49 will not

be counted within the first three months after the song's release” (“Methodology”

2020).

3. Spins: Spins are plays of songs on AM or FM radio.

Database Abbreviation Meanings

1. NOST - Number of Song Titles: How many times the name of the song appears

within the lyrics of the song.

v
2. LTFCA - Length to First Chorus Average: The average of recorded lengths of

time in seconds to the first chorus of the song.

3. LOIA – Length of Intro Average: The average of recorded lengths of time in

seconds of the song’s intro.

4. LOSA – Length of Song Average: The average of recorded length of time in

seconds of the entire song.

5. MCWOC – Max Consecutive Weeks on Chart: Maximum consecutive number of

weeks on the song’s related chart.

6. P – Peak: The highest chart number the song received from release to 8/30/2019,

the end of the sample period.

7. MCS – Multiple Charts: If the song appears on multiple charts simultaneously, it

receives a 1. If the song doesn’t appear on multiple charts, it receives a 0.

8. MSFSA – Multiple Songs from the Same Album: If the song has multiple songs

from the same album on the charts simultaneously, it receives a 1. If the song

doesn’t have other songs from the same album charting, it receives a 0.

9. STO – Sales Total, On: Total sales while the song was on the chart.

10. SWAO – Sales Weekly Average, On: Weekly average of sales while the song was

on the chart.

11. RTO - Radio Total, On: Total radio spins while song was on the chart.

12. STTO – Streams Total, On: Total streams while the song was on the chart.

13. STOB - Sales Total, Off Before: Total sales before the song charted.

14. RTOB – Radio Total, On Before: Total radio spins before the song was on the

chart.

vi
15. STTOB – Streams Total, Off Before: Total streams before the song appeared on

the chart.

16. STOA – Sales Total, Off After: Total streams of a song while not on the chart

after charting.

17. RTOA – Radio Total, Off After: Total radio spins of a song while not on the chart

after charting.

18. STTOA – Stream Total, Off After: Total streams of a song while not on the chart

after charting.

vii
 Tables & Figures

Tables

1. Table 1: Song Breakdown of Observations for Regression 20

2. Table 2: Univariate Analysis of Song Elements 22

3. Tables 3: Regression Analysis: Billboard Hot 100 – Streams 26

4. Table 4: Regression Analysis: All Songs – Streams 27

5. Table 5: Regression Analysis: All Songs – Sales 27

6. Table 6: Regression Analysis: Country – Sales 28

7. Table 7: Regression Analysis: Country – Streams 29

8. Table 8: Regression Analysis: Billboard Hot 100 – Sales 30

9. Table 9: Regression Analysis: Mainstream Rock – Sales 31

Appendix Tables

1. Table A.1: Hit Song’s Deconstructed vs. Personal Analysis 41

2. Table B.1: Songs Analyzed in Study – Complete List 42

3. Table C.1: Regression Analysis: All Songs – Radio 46

4. Table C.2: Regression Analysis: Billboard Hot 100 – Radio 46

10. Table C.3: Regression Analysis: Billboard Hot 100 – Streams 47

11. Table C.4: Regression Analysis: Hip-Hop/R&B – Sales 47

12. Table C.5: Regression Analysis: Hip-Hop/R&B – Radio 48

13. Table C.6: Regression Analysis: Hip-Hop/R&B – Streams 48

14. Table C.7: Regression Analysis: Mainstream Rock – Radio 49

15. Table C.8: Regression Analysis: Mainstream Rock – Streams 49

viii
 16. Table C.9: Regression Analysis: Pop/Contemporary – Sales 50

17. Table C.10: Regression Analysis: Pop/Contemporary – Radio 50

18. Table C.11: Regression Analysis: Pop/Contemporary – Streams 51

Figures

1. Figure 1: Traditional R-squared and McFadden’s pseudo R-squared 25

ix
 Abstract

Although the music industry continues to capitalize on the power of big data and

analytics, the job of predicting a song's future value is left to Artists and Repertoire

(A&R) representatives who must trust their experiences and use their gut instinct. There

remains an opportunity for analytics to unearth the science behind what gives popular

music value. This paper analyzes four quantitative structural elements of a song to

determine how they impact a song's value. Using a systematic method of listening and

data mining, each element was measured and tested for a relationship with the song's

sales, radio spins, and streams. These are songs that made appearances on various

Billboard charts between 2015 and 2018. The difficulty of data cleansing, data

accessibility, and data collecting artistic products is emphasized. Certain elements,

including the repeated lyrics and length of the intro, did show some relationship with

song value, and the extent to which this is true is also emphasized. While the model does

not explain all elements that impact value, this paper could serve to start the discussion

on using big data and analytics to guide music labels on predicting a song's value.

x
 Background

Many aspiring musical artists want to write or perform a song that will catapult

their careers to financial stability. However, a clear path to financial sustainability and

success is unclear since many aspects of music’s value are intangible, such as creating

societal connectivity and a community of belief (Reimer et al. 2002, 3). Perhaps resulting

from this understanding of intangible value, the music industry uses an above-average

amount of intuition on the corporate side of the business (Schrieber and Rieple 2018).

That said, some companies eschew the use of intuition in the decision-making process

and instead leverage big data and analytic research. The industry for business analytics

has grown by $20 billion between 2013 and 2016 (Grover 2018, 390). A report in 2015

showed that companies across all industries that implemented big data and analytics were

showing a 15%-20% increase in annual ROI (“Marketing & Sales Big Data…” 2015, 25).

This is evidence of big data’s effectiveness in decision making, as well as its growing

usage in the corporate world.

As the music industry continues to expand, pulling in an additional two billion

dollars of revenue in music sales alone between 2015 and 2017, one can expect the same

usage of big data and analytics within the music industry (Friedlander 2018, 1). Several

companies have emphasized the advancement of technology and analytics to increase

business performance. For example, in 2016, Warner Music Group organized a new

business structure and data analytics department with the hiring of an official Chief

Information Officer and Chief Data Officer (Schneider 2016). Their press release

emphasized the firm’s mission of implementing big data to improve performance in artist

1
payouts and finances (Schneider 2016). Alternatively, in November 2019, Sony Music

introduced a new listening system called 360 Reality Audio, which is a reinvention of

stereo imaging (Introducing Sony’s 360… 2019). The company plans to rerelease older

songs remixed in this new listening format.

However, major music companies and distributors have not publicized any

attempts to use big data to analyze musical structure and how it is connected to a song’s

value. The purpose of this study is to analyze trends in a specific set of structural

elements in American popular music to determine how they contribute to the song’s

commercial value (i.e., its “popularity”), how these relationships evolve through time,

and how they have changed during the rising popularity of music streaming. Focusing on

the years between 2015 and 2018, this study will use sales data from BuzzAngle to assess

the value of elements for the industry’s most popular genres (11-12). Furthermore, by

focusing on elements of songs that are quantitative, one can determine whether the use of

big data and trend analysis can be used effectively in the music industry to determine the

possible value of specific song structures and therefore lead to making impactful business

decisions. While subjectivity plagues the usefulness of many musical studies in the

business sense, this study aims to determine whether quantitative elements can be of use

to artists and recording companies.

Music has existed for centuries; however, it was not until the invention of the

phonograph and reproducible musical recordings that companies could profit from music

beyond the revenue gained from live performances. While there will always be niche

markets for styles and genres of music, it is evident that some types of music, through

sales and other mediums, are more profitable than others. The terms used to describe

2
different types of music can be easily confused, though. “Popular music,” as described by

Philip Tagg, is recorded music that is funded by the free enterprise for mass consumption

(1982, 42). This is not to be confused with “pop music,” which is a specific genre derived

from the rock-n-roll revolution beginning in the late 1950s (Middleton et al. 2019). While

other genres of music may be produced and released for many reasons, popular music’s

worth is found through maximizing value for those who release it.

As popular music entered into the recording era, music in the mainstream grew to

fit an incredibly specific mold of structural elements. The first elements are “sectional

elements,” which are elements or themes that are repeated in the song. This is in contrast

to music which is “through-composed,” a term which James Webster defines as “based

on run-on movements without internal repetitions,” a technique associated mostly with

classical music (2004, 7). Sectional elements in popular music include verses, choruses,

and bridges. “Verses” are the details and story of the song, and the lyrics typically vary

with each iteration. The “chorus” or “refrain” is normally a repeated, catchy segment of

the song. The “bridge” is a transitional element within the song and can tie the

storytelling of the verse with the theme of the chorus (Davidson 1997).

Songwriter and Canadian Country Music Hall of Fame member Ralph Murphy

noted that there are only a handful of combinations that make up the larger portion of

popular music. While some song structures can be defined through the lyrical content

alone, several songs and genres use a linked combination of the above-stated elements.

For example, Intro–Verse–Chorus–Verse– Chorus–Instrumental–Chorus–Outro is a

popular combination of structural elements in country and rock music. Ralph Murphy

goes on to say that the combination Intro–Verse–Pre-chorus–Chorus–Bridge–Pre-chorus–

3
Chorus is seen statistically as the most profitable structure (“Ralph Murphy Lecture,”

2011). These elements of song structure are difficult to determine, as the subjectivity

inherent to music means these elements can have equal standing even if people perceive

them differently. For example, in a study done by Soundfly on the chord progressions of

the most popular songs of 2017, the author admits that many songs can be interpreted

differently. In the song “That’s What I Like” by Bruno Mars, researcher Dean Olivet

states that depending on which chords were focused on, the song’s tonal center could be

interpreted as either minor or Lydian (Olivet et al. 2018).

From a business perspective, the technology involved with how music reaches the

consumer has evolved systemically in the last 120 years. As would be expected, these

differences have created changes in the financial landscape of the music industry. At one

point, music could only be accessed by attending live performances. This changed with

the invention of the phonograph by Thomas Edison in 1877 and the improvement of flat

disc technology with the gramophone by Emile Berliner. By 1902, these flat discs could

be mass-produced, and by 1910, they were the best-selling medium of commercial music

(Starr and Waterman 2018, 65). This trend continued with the introduction of 8-track

tapes, cassettes, CDs, and streaming services.

Every new technological development brought a change in the landscape to the

music industry. For example, when cassettes surged in popularity during the 1980s,

people could record “mixtapes,” or a compilation of specific songs (Starr and Waterman

2018, 466). This meant that some people could not only bootleg versions of albums, they

could also create cassettes with only the songs they wanted. Another example of this

market disruption comes with the introduction of the CD. When record companies would

4
produce vinyl singles, they would include only one popular song by an artist. However,

the majority of CDs would include an entire album. Music consumers were being

prompted to purchase entire albums instead of having the freedom to purchase single

songs. This would change again when streaming services such as Spotify or Apple Music

gave the power of choice back to consumers (Starr and Waterman 2018, 562–563).

Regardless of the period, every technological innovation within the music world has

directly impacted the financial elements of the industry.

Beginning in 2016, on-demand streaming services created the majority of revenue

in recorded music sales. CDs, which at their peak generated $13.2 billion in revenue in

2000, have been waning in popularity (RIAA 2016). Recently, vinyl has seen a

resurgence in popularity due to collectability. According to Giantsteps Media Technology

Strategies, vinyl will easily outpace CD sales if both new and used markets are

considered (Rosenblatt 2019). Download purchases from iTunes and other paid download

music providers are also trending downwards. Streaming services currently make up 75-

80% of music sales, which is an estimated $8.9 billion according to The International

Federation of the Phonographic Industry (IFPI). Similar to the popularity of cassette

mixtapes and single-song vinyl records, people have more control over the songs they

choose to listen to. This is in contrast to CDs, which require one to buy a collection of

songs without the ability to customize one’s selection. Streaming services also have an

element of convenience, as they can be played from mobile devices. These are all

contributing factors to the growth of streaming services.

The business model for streaming services is not complex. For example, Spotify

offers two accounts for its service: a free account and a premium account. Free Spotify

5
accounts do not require a subscription fee but offer limited features on the platform. For

example, users are not able to pick individual songs and are subject to advertisements

every thirty minutes. Premium Spotify accounts charge users a monthly fee, but they

allow for unlimited online and offline streaming options. In 2019, Spotify charged $9.99

a month and included a subscription to the video streaming service Hulu. Spotify's main

expense is paying music labels and artists for using their music. Generally, payouts are

made based on how much the artist is listened to compared to all other artist. For

example, if a major artist was 10% of the world’s streams, and Spotify generated $1

million from streams, the artist would receive $100,000. While most artists are paid on an

individual basis through aggregators (services that help put music on streaming services),

Spotify has reportedly secured blanket licenses from some major labels in an attempt to

reduce their payout amounts. However, Spotify has faced some lawsuits concerning

illegal practices on paying out mechanical licenses, so how effective these systems will

be in the future is in question (Jacobsen 2017).

The financial viability of music streaming services is an important area of

discussion. There is insufficient evidence to prove that Apple Music and Spotify, the two

major interactive streaming services, use profitable business models. Apple Music is a

subsidiary of Apple, and therefore information is not available on the streaming services’

finances. Apple’s CEO Tim Cook said in an interview with Fast Company that Apple is

“…not in [music streaming] for the money.” Some observers, including Bobby

Owsinoski from Forbes, have stated that Apple considers Apple Music a loss leader.

Apple's main objective is to keep users within their interface and make their mobile

operating system an all-encompassing experience (Owsinoski 2017). In June 2019, it was

6
reported that Apple Music had 60 million current paying subscribers, although the

company later confirmed this number also included people who were on free trial

subscriptions. It may be concluded that Apple Music's role as a loss leader has been

effective. However, this does not mean that music streaming currently has a viable

business model.

Spotify, the other major streaming service, strictly does business in music

streaming. However, Spotify's financial reporting has shown a very concerning trend in

the music streaming industry. Spotify has recorded net losses every single year between

2008 and 2018, and it continues to tell investors that these loses are due to attempts to

increase membership (Spotify 2020, 30). Between 2016 and 2017, Spotify recorded a net

operating income of almost one billion dollars. In 2019, Spotify saw its first positive net

operating income, but only in the first two quarters. Spotify had a net operating loss

overall in 2019, and firms such as Bernstein Bank and Wells Fargo both believe the

company is underperforming (Cohen 2020). Therefore, given the performance of both

Apple Music and Spotify, it is clear that the long-term viability of music streaming

services is in question. The importance of sustainable online revenue cannot be

understated, especially as concerts and live events are being cancelled in the wake of the

COVID-19 pandemic. Overall, live performances are the largest revenue stream for

musical artists. A study by the Music Industry Research Association showed that 80% of

artist revenue came from live shows and performances (Krueger et al. 2018). It can be

expected that the use of big data and analytics will grow in importance as businesses and

individuals look to maximize profits and achieve sustainability in an evolving consumer

landscape.

7
 Generally speaking, the two ways a business can increase operating profits are to

increase sales or reduce costs. Streaming services are already under fire for their payout

statistics. Streaming services offer little insight into the mathematics behind their per

song payout. Multiple sources (The Trichodist, Soundcharts, etc.) will report different

per-stream payouts for the same period. Regardless, artists have noticed that even as

revenue for Spotify and other services have increased, payouts have decreased.

According to The Trichodist, artists would need approximately 333,334 Spotify streams

every month to make minimum wage, based on the Federal minimum wage of $7.25 and

a 40-hour workweek (“2019-2020 Streaming Price Bible…” 2020). If an artist sold

physical CDs and made a net profit of ten dollars per CD (Christman 11, 2017), they

would only need to sell 147 units to make minimum wage. To put this into perspective,

an individual would have to listen to a ten-track album 229 times to equal the time spent

in a streaming service to make the artist the same amount. Streaming services run the risk

of dissuading artists from using their services if these numbers get any lower. However,

SoundExchange reported that in 2018 there were 51 million paying customers of the

music streaming business (Glanz 2018). Therefore, looking at how to increase sales and

optimize products seems to be the only solution.

8
 Literature Review

Many studies have been conducted on measuring musical elements; however,

most focus on subjective elements and are not related to the song’s financial value. A

number of studies on how specific song elements affect a song’s popularity or value state

there is no clear determinant of what makes a song popular. Salganik and Dodes suggest

that popular trends may be due to the cultural phenomenon that is beyond the ability to

predict (2016). Jonah Berger and Grant Packard suggest that songs with lyrics that are

atypical of the genre tend to be the most popular (2018). Their study was done on 1,879

high-charting songs between the years of 2014 and 2016 and focused on buzzwords and

phrases, which attempted to remove the subjectivity of interpreting a song’s theme.

Similarly, in 2014 an analysis in the Journal of Advertising Research was able to

predict if a song would make the Billboard Hot 100 chart with 73.4% accuracy (Hernard

and Rossetti 2014). The study showed that the most common themes in the Billboard Hot

100 songs from 1950 to 2009 were loss, desire, aspiration, breakup, pain, inspiration, and

nostalgia. Breakup songs were found to be the most prevalent theme throughout every

decade. It is difficult to make business decisions based on interpreting lyrical themes due

to the subjectivity involved. To address this, my study focuses on making objective

observations about song elements that would allow the music industry to make decisions

based on the content of their product.

Many elements can be identified in songs. Some of these include song length,

number of choruses, lyric content, and tempo. Many other factors have been covered in

other studies with specific focuses. For example, a paper published in Royal Society Open

9
Science analyzed pop songs that charted in the United Kingdom between 1985 and 2015

and put them in categories depending on the song’s mood. It was found that songs have

become sadder in terms of content, yet the instrumentals were more danceable (Interiano

et al. 2018). In another study entitled “What Has America Been Singing About?”

researchers suggested that music has become more materialistic, whereas older pop music

was more romantic and sexual in nature (Christenson 2018). A study by Michael Tauburg

found that streaming services like Spotify are reinventing classic song titles, with newer

songs either having 1 or 2 word titles, or as many as 7 or more word titles, with less

relevance to the song’s subject matter (“Spotify Is Killing Song Titles” 2018). The major

element of these studies, however, is a personal assessment by the author on what makes

music “materialistic” or “sad” in nature. This highlights the major difficulty in music

analysis: capturing the subjective nature of the art.

One of the most thorough studies on the topic is “Instrumentational Complexity of

Music Genres and Why Simplicity Sells” by Gamaliel Percino, Peter Klimek and Stefan

Thurner published in 2014 by PLOS One. By using the Discogs database of Amazon

sales rankings, they attempt to see how the complexity of instrumentation affected a

song’s popularity. While “instrumental complexity” is a very subjective idea at the

surface, the study strictly defined this through information found on the Discogs

database. The authors assumed that the instrumentational complexity of a style “is related

to the set of specialized skills that are typically required of musicians to play that style"

and therefore only took into account how many and what variety of instruments were

played by contributors of the project (Percino et al. 2014, 3). They found that

10
 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
(Percino et al. 2014, 13).

Unlike other studies, the authors were confident their model could accurately

predict popular music. However, their methodology does assume that open source data

about the subjective topic of instrumentation is accurate. The ability to apply this in a

business setting confidently remains in question.

11
 Thesis Statement

The purpose of this study is to analyze trends in a specific set of structural

elements in American popular music to determine how they contribute to the song’s

commercial value (i.e., its “popularity”), how these relationships evolve through time,

and how they change within different forms of consumption. Focusing on the years

between 2015 and 2018, this study will use data from BuzzAngle to assess the value of

elements for the industry’s most popular genres and assess the impact each has on song

value, as measured by streaming data. (11-12) in the digital age of music. Furthermore,

by focusing on elements of songs that are quantitative, one can determine whether the use

of big data and trend analysis can be used effectively in the music industry to determine

the possible value of specific song structures and therefore lead to making impactful

business decisions. While subjectivity plagues the usefulness of many musical studies in

a business sense, this study aims to determine whether quantitative elements can be of

use to a company and the artist.

12
 Hypotheses

1. The length of the song will be related to song value (Ciampaglia et al, 2015).

a) The relationship between song value and the length of the song will vary

across genres.

b) The relationship between the length of the song and specific sales data

(streams, programmed streams, etc.) will vary due to different consumption

mediums.

2. The length of the song’s intro will be related to song value (Frank 2009).

a) The relationship between song value and the length of the song’s intro will

vary across genres.

b) The relationship between the length of the song’s intro and specific sales data

(streams, programmed streams, etc.) will vary due to different consumption

mediums.

3. The number of times the song title occurs within the lyrics will show a weak

relationship with song value (Tough 2017).

a) The strength of the relationship between song value and the number of times

the song title occurs within the lyrics will vary across genres.

b) The relationship between the number of times the song title occurs within the

lyrics and specific sales data (streams, programmed streams, etc.) will vary

due to different consumption mediums.

4. The length to the first chorus will be related to song value (Murphy 2011).

13
 a) The relationship between song value and the length to the first chorus of the

song will vary across genres.

b) The relationship between the length to the first chorus and specific sales data

(streams, programmed streams, etc.) will vary due to different consumption

mediums.

The above hypotheses are based on previous studies, insight on consumption

habits or guides to music writing technique. Hypothesis 1 follows from the study “The

Production of Information in the Attention Economy,” which used internet traffic to test

how people’s attention span for product and information is getting smaller in the digital

age (Ciampaglia et al, 2015).

In his 2009 book Futurehit.DNA, Jay Frank states that the digital revolution is

pushing songs to have shorter intros to appeal to their consumers (38). Based on the

Frank study, I propose Hypothesis 2.

For Hypothesis 3, I test the findings in the study “An Analysis of Common

Songwriting and Production Practices in 2014-2015 Billboard Hot 100 Songs” by David

Tough (2017). He analyzed the relationship between the number of times a song’s title

appeared and its success on the Billboard Hot 100 chart. He compared this number to

peak date as well as the trend direction of the song by its charting number to calculate an

index number, which he used to define success. He found there was “a weak negative

non-significant relationship” between a song’s success and the number of times the song

used its title in the lyrics (102).

14
 In his book Murphy’s Laws of Songwriting: The Book, Ralph Murphy suggests

keeping the length to the song’s first chorus short and that successful songs follow this

trend. He states that the song’s chorus and catchy hook or phrase should happen before

the 60 second mark of the song (2011). In this study, I test a relationship between the

length to the song’s first chorus and value in Hypothesis 4.

15
 Methodology

The value of a financial investment is simply the present value of all future

expected cash flows. However, valuing a song is less clear cut. The study considers a

variety of measures of song value. In addition, I gather data on specific song elements to

determine what drives value.

The first phase of the study was collecting the raw data and creating a database of

songs to analyze. The songs analyzed were the top five songs per chart at the end of each

month between the years of 2015 and 2018. These songs were collected from Billboard

Pro, the official database for Billboard chart occurrences. According to a study by

BuzzAngle in 2017, the four most popular genres of music by album consumption are

Hip-Hop/R&B, Country, Rock, and Pop (11-12). Since the focus of this study is on a

song’s profitability, I will focus on these four genres. Specifically, Pop music was

analyzed through the Adult Contemporary chart, Hip-Hop and R&B music was analyzed

through the Hot R&B/Hip-Hop chart, Country was analyzed through the Hot Country

Songs chart, and Rock music was analyzed through the Mainstream Rock chart. In

addition, a fifth category of the Billboard 100 was analyzed, which accounts for each

month’s most popular songs without considering the genre. This allows insight into

which genres are seen as most popular by consumers as well as whether or not the most

popular songs are derived from the genre or elements within a song.

After the list of songs was finalized, I analyzed each song individually based on

four elements. One methodological limitation of the study is the subjective nature of

analyzing music. Specific details such as chord progressions or tempo proved too and

16
interpretive to offer clean data to analyze profitability. To address this, the four elements

of this study were chosen due to their quantitative nature, making cross-analysis easier.

While not entirely objective, the elements studied are less interpretive than others that

could have been chosen. The elements analyzed are the length of the song, the length of

the intro, the length to the first chorus, and the number of times the song title occurs

within the lyrics. All encoding was done by listening to each song individually.

The four song elements will be followed for the entirety of the study. The length

of song will be defined as the time when the music begins to the time when there is no

additional music being played. This includes fade-outs but does not include echoes or

delays. The length of intro will be defined as the time from the start of the song to the

first downbeat of the next structural element, such as verse or chorus. If a vocalist begins

singing before the first beat of the verse, for example, the length of the intro will still be

measured as extending to the first downbeat of the verse. This approach will also be used

to determine the length of the chorus. For the number of times the song’s title occurs in

the lyrics, variations from the title will not be counted. Lastly, to add consistency, the

original album release of the song will be the one used in the study. Songs that are

structurally altered through video streams were not counted.

To record these elements, I programmed a stopwatch with JavaScript that started

and stopped the time upon clicking the mouse. For consistency, all songs were listened to

using Apple Music and to avoid latency issues, all songs were downloaded through the

streaming service. To ensure accuracy, each song was listened through three times and

the average value for each element, length of intro (“LOIA”), length to first chorus

(“LTFC”) and length of song (“LOSA”), were used in the regression analysis. A fourth

17
listen of the song was to note the number of times the song’s title appeared in the chorus

(“NOST”) which was measured by identifying each occurrence, and cross-referencing the

number by searching through the song's lyrics on Genius.com.

To validate my plan and my personal analysis, I analyzed 10 songs that were also

analyzed by Hit Songs Deconstructed, a database that does song analysis and is used at

universities for reference purposes. After comparing the results of their research with my

own analysis, and discounting two unclear data inconsistencies within Hit Song’s

Deconstructed’s analysis, the two data groups were 100% synonymous after discounting

differences in methodology (see Appendix A). In the case of the one inconsistency

analyzing the song “Do I Wanna Know” by the Arctic Monkeys, I noted that the song

title was included six times within the song, whereas Hit Songs Deconstructed said it

occurred five times, with one partial occurrence. The difference is due to the fact that for

this study, I do not include partial occurrences.

One aspect of this study was to see how well songs performed relative to their

time on each of our selected charts. Billboard Pro’s database provides a database that

includes charting appearances. This data shows the charting position of each song every

week. From there, I recorded the number of weeks the song was “on the chart” and “off

the chart.” Each song’s time off the chart was divided into pre and post charting

appearances. Before charting appearances were the number of weeks before the song’s

first appearance on the subsequent chart. After chart appearances are every week the song

was off the chart after the song had its first appearance on the chart.

After recording these elements, I cross referenced each song to BuzzAngle’s sales

database. BuzzAngle provides “… complete access of every single music consumption

18
transaction in its finest detail" (“Our Platform” 2020). Similar to Nielsen, this database

provided sales information for each song of the database. From this database, I was able

to get weekly data for each song in three categories: interactive online streams, and

digital sales and radio spins. These will be referred to as “audio streams”, “song sales”

and “spins.” Specific limitations of these three categories are provided in the

Abbreviations & Definitions section. These three categories were split into three

categories of charting appearances: On-Chart, Off-Chart Before, and Off-Chart After. I

recorded the sales data from the song's release date until the end of the sample period

studied, August 30th, 2019. A cut-off date was chosen when data collection began, as

songs can sell infinitely beyond their release date. From the initial recording of 1,200

chart occurrences, the number of songs used in the final database was 356. However, for

the sake of regression analysis, songs that appeared on multiple charts were recorded

twice for genre-specific analysis. Including duplicate songs because of chart

reoccurrences, the database contained 417 songs. This breakdown of the sample studied

is shown in Table 1.

19
 Table 1: Song Breakdown of Observations for Regression

Next, I calculated several values from this database. I calculated the total and

weekly averages of audio streams, song sales, and spins while they were on-chart, before

they were on the chart, and after they were on the chart. I also calculated the maximum

number of weeks the song consecutively charted. Other elements that were recorded were

the peak charting number (1, 2, 3, etc.) and the first date in which the song reached its

peak charting number. These were used as control variables. Other control variables

include "Multiple Charts" and "Multiple Songs From the Same Album." "Multiple

charts" denotes when a song appears on multiple charts simultaneously. A "1" on the

database signified if this was true, and a "0" denotes if this is false. "Multiple Songs From

The Same Album" denotes when multiple songs from the same album were charting

simultaneously, and the same process was used to signify this on the database as the

"Multiple Chart" value. To make sure the data sample was of high quality, songs were

removed if all data was not available. The criteria for having an incomplete dataset

20
consisted of any missing information in the song’s sales figures and charting time or the

lack of an official version on streaming services. A list of songs studied is provided in

Appendix B.

21
 Results

The first step was completing univariate analysis on the song database. This

included calculating the mean, median, standard deviation and the minimum and

maximum for each of the four song elements ("LOIA," "LOSA," "LTFCA," and

"NOST"). The univariate results are shown in Table 2 for the entire sample and for each

genre.

Table 2: Univariate Analysis of Song Elements

22
 Across the whole database, the average song mentioned the title in the lyrics

10.23 times, with a 12.1 second intro, 44.26 seconds to the first chorus, and an overall

length of 218.84 seconds. On average, songs that appeared on the Billboard Hot 100 list

mentioned the title in the lyrics more than any other genre with 13.14 occurrences.

Mainstream Rock showed the fewest mentions of the title in the lyrics on average (8.72

occurrences) with the smallest standard deviation of 7.85 occurrences. Mainstream Rock

also had the longest time to first chorus on average with 52.64. My analysis showed that

in comparison to other genres, Hip-Hop/R&B had the highest number of songs that began

with the chorus and subsequently had the lowest average length to first chorus with 39.52

seconds. Pop/Contemporary songs had the shortest average intro with 8.73 seconds (and

the smallest standard deviation of 6.63 seconds) with Mainstream Rock averaging the

longest intros with 19.33 seconds. Pop/Contemporary charts are based heavily on radio

play, and this may explain the desire for a short introduction. Country songs averaged the

lowest total length of songs with 205.23 seconds, with Mainstream Rock averaging the

longest songs with 234.43 seconds. The extreme nature of Mainstream Rock’s elements

may explain why the sales of the genre were lowest of the group, perhaps serving a niche

market. Differences in observation among elements in the same genre can be explained

through Tobit regression censoring.

Next, a Tobit regression method was chosen due to the nature of the database and

the sales value. Tobit regressions are the best option because the measure of value is

limited to values greater than or equal to zero. A number of regressions were run to

predict sales, streams and radio spins based on the four song elements for the songs in the

database. Due to the large numerical values of sales, streams, and radio spins (our

23
dependent variables), each value was scaled by a million. The regressions were

completed by genre and for the total sample and for three measures of value. In addition,

models were also estimated for totals as well as pre-charting and post charting values.

Additional independent variables were added when calculating the regression over the

period of time from the song's release. These include the song's peak charting number

("P"), the song's longest consecutive run on the chart ("MCWOC"), if the song was

appearing on multiple charts simultaneously ("MCS"), and if multiple songs were

charting from the same album ("MSFSA"). Due to the vast number of regressions, I focus

on important findings throughout, and my explanations for what I learned. The results are

divided based on the hypotheses made about each element and highlight significant

values that support a working model, as well as the importance of which time segment of

data is focused on. The regression results not discussed in the text are provided in

Appendix C.

Tobit regression analysis does not have an equivalent to R-squared, which is seen

in OLS regressions. This is mathematically impossible with multinomial regressions. In

OLS regression, the sum of squared regressions and sum of squared errors will always

equal the sum of squared total. This is not the case when the regressions are non-linear.

There are a number of pseudo R-squared equations that attempt to calculate a goodness-

of-fit statistic. Stata, the software used for this study calculates McFadden’s pseudo R-

squared valued. A relationship between the R-squared of OLS regressions and

McFadden’s pseudo R-squared can be found in the book Urban Travel Demand: A

Behavioral Analysis by Tom Domencich and Daniel L. McFadden (124). McFadden’s

24
pseudo R-squared is calculated as equaling one minus the constant-only log likelihood

divided by the full model log-likelihood.

Figure 1: Relationship between Traditional R-squared and McFadden’s pseudo R-squared

From this, it can be understood that larger pseudo R-squared values indicate a

stronger fit for the model, where smaller values do not. Negative values are also possible

in McFadden’s pseudo R-squared for continuous or mixed continuous regressions such as

Tobit (Sribney 2020).

25
Total vs. Before Charting vs. After Charting

Table 3: Regression Analysis: Billboard Hot 100 - Streams

The initial regression analysis was completed over the length of time between the

song’s release until the end of the sample period, August 30th, 2019. However, it became

obvious very quickly that a significant predictor of song value would not be found in total

values or after charting data. Once a song is on the chart, it is popular. Hence, there is a

redundancy to using the time on the chart to indicate value. By the time a song has left

the chart, many outside factors may be involved it what influences value. Referencing

Billboard Hot 100 streams in Table 3, NOST (p > .007), LOIA (p > .036) and LOSA (p >

.024) were all significant predictors. Longer songs with short introductions that used the

title of the song extensively throughout the lyrics correlated to a higher value. Restricting

the regressions to pre-charting data shows the strongest method of valuing a song through

these elements.

26
Length of Song’s Intro

Table 4: Regression Analysis: All Songs - Streams

Table 5: Regression Analysis: All Songs - Sales

Hypothesis 2 predicted that the length of a song’s introduction would have a

significant relationship to value. Of the four song elements chosen, the length of the

song’s intro shows significance in several places. Referencing Table 4 and Table 5, LOIA

is significantly related to STOB (p > 0.005, with a pseudo R-squared of 2.26%) and

STTOB (p > 0.001, with a pseudo R-squared of 0.57%) for all songs included in the

27
database. In both cases, songs with short introductions result in higher streaming and

sales numbers. If we increase the LOIA by 1 standard deviation, we see decrease of 6.88

in the number of streams as measured by STOB. This is much more effective than for

streams, which sees a minute decrease of 0.044 in the number of streams as measured by

STTOB. However, the effectiveness changes dramatically depending on the genre.

Referencing Table 3, Billboard Hot 100 songs saw a 39.72 decrease in the number of

streams for every 1 standard deviation increase in LOIA (p > .05, with a pseudo R-

squared of 1.31%).

Length of Song

Table 6: Regression Analysis: Country - Sales

28
 Table 7: Regression Analysis: Country - Streams

Hypothesis 2 predicted that the length of a song would have a significant

relationship to value. Results for the Country genre are consistent with Hypothesis 2.

Referencing Table 6 and Table 7, LOSA is significantly related to STOB (p > .045, with

a pseudo R-squared of 8.12%) and STTOB (p > .025, with a pseudo R-squared of 0.75%)

for all Country songs included in the database. In both cases, the longer songs have

higher streaming and sales numbers. A 1 standard deviation change in LOSA results in

an increase of 26.50 in the number of streams as measured by STTOB. Length of song

showed far less effect on sales in the Country genre, showing only a 0.03 increase of

0.044 in the number of sales as measured by STOB. This can also be for songs that

appeared on the Billboard Hot 100. Referencing Table 3, if we increase LOSA by 1

standard deviation, we see an increase of 42.46 in the number of streams as measured by

STTOB (p > .05, with a pseudo R-squared of 1.31%). We can reject the null hypothesis

in these instances.

29
Number of Times Title of Song Appears in Lyrics

Table 8: Regression Analysis: Billboard Hot 100 - Sales

Hypothesis 3 predicted that the number of times the title of a song appears in the

lyrics would not have a strong relationship with the song’s value. Although this remained

true across all songs, songs on the Billboard Hot 100 chart showed a significant

relationship between sales and streams and NOST. Referencing Table 3 and Table 8,

NOST is significantly related to STOB (p > .017, with a pseudo R-squared of -55.11%)

and STTOB (p > .024, with a pseudo R-squared of 1.31%) for all songs in the database in

Billboard Hot 100. For every 1 standard deviation increase in NOST, there is a 49.96

increase in the number of streams as measured by STTOB. In contrast, the impact on

STOB is only a 0.03 increase for each 1 standard deviation in NOST. According to this

model, the number of times the title of a song appears in lyrics does have a relationship

with value.

30
Length to the Song’s First Chorus

Table 9: Regression Analysis: Mainstream Rock - Sales

Hypothesis 4 predicted that the length of a song’s first chorus would have a

significant relationship to value. Of the four song elements chosen, the length to the

song’s first chorus shows the least significance over the span of genre’s selected.

However, an interesting relationship is found in Mainstream Rock. MCS was removed

from the calculations on Mainstream Rock to due collinearity. Referencing Table 9,

LTFC is significantly related to STOB (p > .041, with a pseudo R-squared of 1.9%) for

all songs included in the database from the Mainstream Rock genre. The faster a song

arrived at the chorus section, the more valuable it was seeming to be connected with sales

numbers in the Mainstream Rock genre. For every 1 standard deviation increase in

LTFC, there is a decrease of 1.83 in the number of sales as measured by STOB. This is

more effective than songs across all genres, which did not show significance for this

element. According to the database, the songs with the longest time to the first chorus

were mainly in the Mainstream Rock genre, which makes these findings surprising.

31
 The critical value of p > .05 was used in my analysis. However, because multiple

tests were conducted, adjusting the alpha level may be a necessary step for future tests.

Multiple methods of to correct alpha levels exist, including the Bonferroni correction.

Bonferroni’s correction divides the alpha level by the number of tests conducted. In this

case, regressions were calculated on five genres, as well as the five genres collectively,

over three time periods. This would make the critical value, by Bonferroni’s correction,

p > .0028, or p > .01 if discounting time periods. This would eliminate the majority of

significances found throughout the regressions. However, due to the complexity and

number of regressions calculated and the conservative nature of such correction

methodologies, we assumed all the tests were independent of each other for this study to

avoid substantial occurrences of false positives. Future studies, however, may benefit

from some form of correction.

32
 Discussion

The results of how songwriting elements affect a song’s value were surprising.

Some elements proved to be more effective at creating value. For example, the length of

song was more effective at predicting value than length to the song’s first chorus across

all genres. Some genres were more impacted by certain elements than others. For

example, Mainstream Rock songs were more valuable when they had shorter choruses

compared to other genres. Billboard Hot 100 songs were more valuable when they had

shorter introductions and the title of the track was used in the lyrics frequently. However,

the four factors analyzed do not offer a straightforward answer to predicting a song’s

value. To create simple, quantitative elements of songs that could be analyzed

automatically, the model does not provide a surefire way of answering the question of

what drives value. This perhaps could be expected in the digital age. So many influences

can affect a song’s value. These include popularity through viral means (including

memes and other social media posts), movie placements, and non-quantitative artistic

elements. Trying to quantify these factors remains difficult, and these findings are similar

to other studies. For example, the 2018 study by Minna Reiman and Philippa Ornell

attempted to use machine learning with algorithms such as Gaussian Naïve Bayes and

Support Vector Machine to predict whether songs would reach the Billboard Hot 100 list.

Their model was only able to predict songs with 60% accuracy (Reiman and Ornell,

2018).

A key element of my hypotheses was that the value would vary depending on

consumption medium. This turned out to be true, as the elements overall showed to be

33
valuable indicators in streaming numbers but not sales number. This may be due to the

demographics that are purchasing music versus those who are streaming services.

However, streaming is the most valuable medium to predict, as nearly 80% of music

consumption in 2019 was via paid streaming services (Nicolaou 2020).

In 2014, a viral video by Sir Mashalot emerged that showed six chart-topping

songs could be laced together seamlessly through minor editing in Pro Tools. These six

songs were “Sure Be Cool If You Did” by Blake Shelton, “Drunk on You” by Luke

Bryan, “Chillin’ It” by Cole Swindell, “Close Your Eyes” by Parmalee, “This is How

We Roll” by Florida Georgia Line, and “Ready, Set, Roll” by Chase Rice (“Sir

Mashalot…” 2014). At the surface, certain genres are more prone to following formulaic

song structures than others. However, my regression analysis only found significance in

the length of country songs in streams before charting and streams before charting. This

may suggest that using seconds instead of bars in studies on song structure may not be

effective. However, that analysis is subjective, and the use of quantitative and non-

subjective data collection becomes more difficult. Until a more sophisticated computer-

based listening methodology is created, human subjectivity may always be necessary for

studies such as this.

However, for songwriters, these elements are not to be completely dismissed. The

strongest correlations were found within songs on the Billboard Hot 100, which is a good

indicator of what is popular among listeners at the time. Specifically, short songs with

titles that are repeated without the lyrics were linked to an increase in value through

streams. These factors may be considered when attempting to write songs to receive

widespread popularity.

34
 Of all consumption mediums, radio spins had very few correlations throughout

the study. Unlike streaming and direct sales, the consumer has little choice in deciding

what they are consuming. As covered in a 2019 Rolling Stones article, despite formal

investigations in 2004, the practice of payola and “pay-to-play” is still rampant within the

industry (Leight 2020). While these allegations are far from proven, it does beg the

question of the usefulness of including radio spins in studies to determine the value of a

song.

Another important discussion is the difficulty of collecting clean data related to

music sales, charting data and information on song elements. As I experienced over the

course of this study, the data available on music is not easy to obtain nor extremely

accurate upon review. Individual labels and companies are in charge of reporting data.

Some songs, specific to certain labels, had large amounts of missing sales data. Billboard

has made third party collection of their data illegal under copyright infringement. Outside

of large established labels that have access to this data, the process remains tedious. With

advances in technology, the access and accuracy of data should continue to improve, so

that independent labels and lower-level artists may have the ability to use this data to the

advantage of their careers.

35
 Conclusion

This paper shows one example of how big data and analytics can be used to

predict song value. As companies continue to implement the use of big data and analytics

throughout the business world, it only seems certain that it will attempt to automate all

facets of the business. This includes A&R jobs at major labels and music conglomerates.

Although my model does not show any groundbreaking relationships between my chosen

elements, it does highlight the possibilities involved with creating a model that could

predict song value. In a market that is rewarding sustainable online revenue in the face of

pandemic, the use of big data and analytics to maximize these profits in foreseeable

future is vital. With the continued growth of these technologies, it may one day be

possible to find a science within the art of music.

36
 Works Cited

“2019-2020 Streaming Price Bible: YouTube Is STILL The #1 Problem To Solve.” The
Trichordist, March 5, 2020. https://thetrichordist.com/2020/03/05/2019-2020-
streaming-price-bible-youtube-is-still-the-1-problem-to-solve/.

Berger, Jonah, and Grant Packard. “Are Atypical Things More Popular?” Psychological
Science 29, no. 7 (July 2018): 1178–84. doi:10.1177/0956797618759465.

BuzzAngle Music. BuzzAngle Music 2017 Report US, 2017.

https://www.buzzanglemusic.com/wp-content/uploads/BuzzAngle-Music-2017-
US-Report.pdf. 28 December 2018.

Christenson, Peter G., et al. “What Has America Been Singing about? Trends in Themes
in the U.S. Top-40 Songs: 1960–2010.” Psychology of Music, (January 2018) pp.
1-19. EBSCOhost, doi:10.1177/0305735617748205.

Christman, Ed. “Life at the Margins.” Billboard, 14 Apr. 2007, pp. 11.

Ciampaglia, G. L., Flammini, A., & Menczer, F. (2015). The production of information in
the attention economy. Scientific Reports, 5(9452). doi:10.1038/srep09452

Cohen, Colin. “Bernstein Initiates Coverage of Spotify, Starts at 'Underperform'.” Digital

Music News, 13 Jan. 2020, www.digitalmusicnews.com/2020/01/13/bernstein-
initiates-coverage-spotify/.

Davidson, Miriam, and Kiya Heartwood. Songwriting: An Easy Beginning Method.

Alfred Pblishing Co., 1997.

Domencich, Thomas A., and Daniel McFadden. Urban Travel Demand: A Behavioral
Analysis. Amsterdam: North-Holland Pub. Co., 1975.

Frank, Jay L. Futurehit.DNA. Nashville: Futurehit Press, 2009.

Friedlander, Joshua. “News and Notes on 2017 RIAA Revenue Statistics.” RIAA, last
modified March 22, 2018, www.riaa.com/wp-content/uploads/2018/03/RIAA-
Year-End-2017-News-and-Notes.pdf.

Glanz, William. “Report: Streaming Has 51 Million Paying Customers in U.S.”

SoundExchange, October 18, 2018.
https://www.soundexchange.com/2018/09/13/report-streaming-has-51-million-
paying-customers-in-u-s/.

37
Grover, Varun, et al. “Creating Strategic Business Value from Big Data Analytics: A
Research Framework.” Journal of Management Information Systems 35, no. 2
(2018) pp. 388–423. EBSCOhost, doi:10.1080/07421222.2018.1451951.

Henard, David H, and Christian L Rossetti. “All You Need Is Love? Communication
Insights From Pop Music’s Number-One Hits.” Journal of Advertising Research
54 (June 1, 2014): 53–66. https://doi.org/10.2501/JAR-54-2-178-191.

Interiano, Myra et al. "Musical Trends And Predictability Of Success In Contemporary

Songs In And Out Of The Top Charts". Royal Society Open Science 5, no. 5
(2018) p. 171-274. The Royal Society, doi:10.1098/rsos.171274

International Federation of the Phonographic Industry (IFPI). "Music streaming revenue

worldwide from 2005 to 2018 (in billion U.S. dollars)." Chart. April 2, 2019.
Statista. Accessed December 12, 2019.
https://www.statista.com/statistics/587216/music-streaming-revenue/

Introducing Sony's 360 Reality Audio | The Future of Music. YouTube. YouTube, 2019.
https://www.youtube.com/watch?v=h4ZbhQgE9_k.

Jacobson, Erin M. “Spotify May Have To Pay Songwriters $345 Million.” Forbes. Forbes
Magazine, July 19, 2017.
https://www.forbes.com/sites/legalentertainment/2017/07/19/spotify-may-have-
to-pay-songwriters-345-million/#4d79c25c193d.

Krueger, Alan B., et al. “Inaugural Music Industry Research Association (MIRA) Survey
of Musicians.” 22 June 2018, https://img1.wsimg.com/blobby/go/53aaa2d4-793a-
4400-b6c9-95d6618809f9/downloads/1cgjrbs3b_761615.pdf.

Leight, Elias. “Want to Get on the Radio? Have $50,000?” Rolling Stone, January 16,
2020. https://www.rollingstone.com/music/music-features/radio-stations-hit-pay-
for-play-867825/.

“Marketing & Sales Big Data, Analytics, and the Future of Marketing & Sales.”
McKinsey & Company, March 2015,
https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/Marketin
g%20and%20Sales/Our%20Insights/EBook%20Big%20data%20analytics%20an
d%20the%20future%20of%20marketing%20sales/Big-Data-eBook.ashx

“Methodology.” BuzzAngle Music, 2020,

www.buzzanglemusic.com/charts/methodology/.

Middleton, Richard, et al. “Pop.” Grove Music Online. (January 2001) pp. 1-54.
Accessed 9 Jan. 2019, www.oxfordmusiconline-
.com.ezproxy.mtsu.edu/grovemusic/view/10.1093/gmo/9781561592630.001.0001
/omo-9781561592630-e-0000046845

38
Murphy, Ralph. Murphy’s Laws of Songwriting: The Book. Nashville, Tennessee:
Murphy Music Consulting, 2011.

Nicolaou, Anna. “US Music Sales Set 13-Year Record After Streaming Surge.” Financial
Times. Financial Times, February 25, 2020.
https://www.ft.com/content/448e544a-57e1-11ea-a528-dd0f971febbc.

Olivet, Dean, et al. “Chartmania!! I Broke Down Every Song That Reached the Billboard
Top 5 in 2017.” Soundfly, 11 Jan. 2018, flypaper.soundfly.com/write/chartmania-
breaking-down-billboard-top-40-songs-2017/.

“Our Platform.” BuzzAngle Music, 2020. https://www.buzzanglemusic.com/platform/.

Owsinski, Bobby. “This Is Why Spotify Is Beating Apple Music.” Forbes. Forbes
Magazine, September 16, 2017.
https://www.forbes.com/sites/bobbyowsinski/2017/09/16/spotify-apple-
music/#503f01a3170b.

Percino G, Klimek P, Thurner S (2014) Instrumentational Complexity of Music Genres

and Why Simplicity Sells. pp. 1-16. PLOS ONE 9(12): e115255.
https://doi.org/10.1371/journal.pone.0115255

“Ralph Murphy Lecture - How to be Successful at Songwriting.” Filmed [July 2011].

YouTube video, 1:48:45. Posted [July 2011]. http://youtu.be/8wBOUJ5Mbrk.

Reiman, M., & Örnell, P. (2018). Predicting Hit Songs with Machine Learning
(Dissertation). http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-229705

Reimer, Bennett & Palmer, Anthony J. & Regelski, Thomas A. & Bowman, Wayne D.
"Why Do Humans Value Music?" Philosophy of Music Education Review 10, no.
1 (2002): pp. 41-41. Project MUSE, muse.jhu.edu/article/408670.

Recording Industry Association of America (RIAA). (2016). U.S. sales database.

Undefined: RIAA.

Rosenblatt, Bill. “Vinyl Is Bigger Than We Thought. Much Bigger.” Forbes. Forbes
Magazine, March 29, 2019.
https://www.forbes.com/sites/billrosenblatt/2018/09/18/vinyl-is-bigger-than-we-
thought-much-bigger/#3926d2eb1c9c.

Salganik, M. J., Dodds, P. S., Watts, D. J. (2006). Experimental study of inequality and
unpredictability in an artificial cultural market. Science, 311, 854–856.

Schneider, Marc. “Warner Music Group Hires Chief Information Officer, Adds New
Data Role.” Billboard. Prometheus Global Media, LLC, December 14, 2016.

39
 https://www.billboard.com/articles/business/7624487/warner-music-group-ralph-
munsen-vinnie-freda.

Schreiber, David, and Alison Rieple. “Uncovering the Influences on Decision Making in
the Popular Music Industry; Intuition, Networks and the Desire for Symbolic
Capital.” Creative Industries Journal 11, no. 3 (2018): pp. 245–262. EBSCOhost,
doi:10.1080/17510694.2018.1490146.

Sribney, William. “Pseudo-R2 for Tobit .” Stata. StataCorp. 2020. Accessed March 22,
2020. https://www.stata.com/support/faqs/statistics/pseudo-r2/.

“Sir Mashalot: Mind-Blowing SIX Song Country Mashup” Sir Mashalot, published on
November 4th, 2014, YouTube video, 03:55,
https://www.youtube.com/watch?v=FY8SwIvxj8o.

Spotify. "Annual Report for 2019." Spotify (2019): 1-238. Spotify. February 12, 2020.
https://s22.q4cdn.com/540910603/files/doc_financials/quarterly/2019/601c445e-
1d37-4938-b854-e5344850c3f9.pdf

Starr, Larry, and Christopher Alan Waterman. American Popular Music: from Minstrelsy
to MP3. Oxford University Press, 2018.

Tagg, Philip. “Analysing Popular Music: Theory, Method and Practice.” Popular Music,
vol. 2 (1982): pp. 37–67. JSTOR, www.jstor.org/stable/852975.

Tauberg, Michael. “Spotify Is Killing Song Titles.” Medium, last modified March 23,
2018, www.medium.com/@michaeltauberg/spotify-is-killing-song-titles-
5f48b7827653.

Tough, David T. "An Analysis of Common Songwriting and Production Practices in

2014-2015 Billboard Hot 100 Songs." MEIEA Journal 17, no. 1 (2017): 79-120.
Academic OneFile (accessed February 25, 2019).
https://link.galegroup.com/apps/doc/A520582346/AONE?u=tel_middleten&sid=
AONE&xid=9a3d9620.

Webster, James. Haydn's "Farewell" Symphony and the Idea of Classical Style: Through-
Composition and Cyclic Integration in His Instrumental Music. Cambridge Univ.
Press, 2004.

“What Are Pseudo R-Squareds?” UCLA. IDRE Stats, October 20, 2011.
https://stats.idre.ucla.edu/other/mult-pkg/faq/general/faq-what-are-pseudo-r-
squareds/.

40
 Appendices

Appendix A

Data Collection Comparison – Hit Song’s Deconstructed vs. Personal Analysis

Table A.1: Data Collection Comparison – Hit Song’s Deconstructed vs. Personal Analysis

Note: This chart presents a comparison between my analysis and that of Hit Song's
Deconstructed, a database used in universities and studying in-depth song analysis.
Due to the possible subjectivity of music analysis, it is important to show the
similarities and or differences between my analysis and that of other published
sources. The abbreviation "CNIR" means that the report had conflict information

41
Appendix B

Songs Analyzed in Study

Table B.1: Songs Analyzed in Study – Complete List

42
Table B.1: Songs Analyzed in Study – Complete List (Continued)

43
Table B.1: Songs Analyzed in Study – Complete List (Continued)

44
Table B.1: Songs Analyzed in Study – Complete List (Continued)

45
Appendix C

Regression Results – Separated by Genre and Consumption Type

All Songs – Radio

Table C.1: Regression Analysis: All Songs - Radio

Billboard Hot 100 – Radio

Table C.2: Regression Analysis: Billboard Hot 100 - Radio

46
Country – Radio

Table C.3: Regression Analysis: Billboard Hot 100 - Streams

Hip-Hop/R&B – Sales

Table C.4: Regression Analysis: Hip-Hop/R&B - Sales

47
Hip-Hop/R&B – Radio

Table C.5: Regression Analysis: Hip-Hop/R&B – Radio

Hip-Hop/R&B – Streams

Table C.6: Regression Analysis: Hip-Hop/R&B – Streams

48
Mainstream Rock – Radio

Table C.7: Regression Analysis: Mainstream Rock – Radio

Mainstream Rock – Streams

Table C.8: Regression Analysis: Mainstream Rock – Streams

49
Pop/Contemporary – Sales

Table C.9: Regression Analysis: Pop/Contemporary – Sales

Pop/Contemporary – Radio

Table C.10: Regression Analysis: Pop/Contemporary – Radio

50
Pop/Contemporary – Streams

Table C.11: Regression Analysis: Pop/Contemporary – Streams

51