While the limited bandwidth of FM radio facilitated widespread adoption of mainstream music preferences and spending habits, new digital music technologies recommend and feature music based on personalized user profile data. Whether this includes tracking purchase history, song “likes”, users’ emotions, or otherwise, the shift from majority-based music recommendation to individual-based is a recent and relatively unexplored development in the music industry. The purpose of this study is two-fold: to determine the most influential factors shaping users’ choice of music technology, and the extent to which these new technologies affect music preferences, spending and engagement. Focusing on iTunes, Pandora, and YouTube, purpose-built surveys examine the reasons users choose each service and how they perceive the technologies have affected their music consumption. Additional survey questions seek patterns and correlations between demographics, musical experience, music preferences, and music listening environment. 125 college students voluntarily completed the survey, revealing strong correlations between variables currently ignored by music recommendation technology. By enhancing our understanding of how new music technologies impact individual users, this study may guide how music applications can improve user profiling, personalization, and the user’s music-listening experience as a whole.
North Avenue Call Girls Services, Hire Now for Full Fun
The Longer Tails of iTunes, Pandora, and YouTube
1. The Longer Tails of
iTunes, Pandora, and YouTube:
New Technology Shaping
Music Preference and Spending
Andrew
D.
Penrose
Program
in
Science,
Technology,
and
Society
Stanford
University
The
author
wishes
to
thank
his
advisors
Professor
Robert
McGinn
and
Professor
David
Voelker
at
Stanford
for
their
valued
feedback
and
guidance
on
this
project.
Additional
thanks
to
all
survey
respondents,
interview
volunteers,
Professor
Fred
Turner
for
the
lecture
that
inspired
this
study,
and
all
others
who
contributed
support.
Correspondence
concerning
this
paper
can
be
sent
to
Andrew
Penrose,
675
Lomita
Drive,
Stanford,
CA
94305.
Address
email
to
apenrose@stanford.edu.
2. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Abstract
While
the
limited
bandwidth
of
FM
radio
facilitated
widespread
adoption
of
mainstream
music
preferences
and
spending
habits,
new
digital
music
technologies
recommend
and
feature
music
based
on
personalized
user
profile
data.
Whether
this
includes
tracking
purchase
history,
song
“likes”,
users’
emotions,
or
otherwise,
the
shift
from
majority-‐based
music
recommendation
to
individual-‐based
is
a
recent
and
relatively
unexplored
development
in
the
music
industry.
The
purpose
of
this
study
is
two-‐fold:
to
determine
the
most
influential
factors
shaping
users’
choice
of
music
technology,
and
the
extent
to
which
these
new
technologies
affect
music
preferences,
spending
and
engagement.
Focusing
on
iTunes,
Pandora,
and
YouTube,
purpose-‐built
surveys
examine
the
reasons
users
choose
each
service
and
how
they
perceive
the
technologies
have
affected
their
music
consumption.
Additional
survey
questions
seek
patterns
and
correlations
between
demographics,
musical
experience,
music
preferences,
and
music
listening
environment.
125
college
students
voluntarily
completed
the
survey,
revealing
strong
correlations
between
variables
currently
ignored
by
music
recommendation
technology.
By
enhancing
our
understanding
of
how
new
music
technologies
impact
individual
users,
this
study
may
guide
how
music
applications
can
improve
user
profiling,
personalization,
and
the
user’s
music-‐listening
experience
as
a
whole.
Keywords:
digital
music
technology,
the
Long
Tail,
music
preferences,
profiling,
multivariate
music
recommendation,
iTunes,
Pandora,
YouTube,
internet
radio
1
3. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Table
of
Contents
INTRODUCTION
(3-4)
LITERATURE
REVIEW
(4-10)
THE
LONG
TAIL:
NEW
TECHNOLOGY
REVEALING
NICHE
PREFERENCES
MUSIC
PREFERENCE
STUDIES
METHODS
(10-16)
SAMPLING
CONCEPTS
Demographics
Musical
Experience
Music
Preferences
Listening
Environment
Music
Service
Features
and
Effects
CODING
AND
DATA
ANALYSIS
RESPONDENTS
RESULTS
(16-60)
MUSIC
PREFERENCE
AND
DETERMINING
FACTORS
Song
Preference
Genre
Preference
Determining
Factors
in
Music
Preference
Correlations
Between
Genres
Demographics
and
Genre
Preferences
Musical
Experience
and
Genre
Preferences
Listening
Environment
and
Genre
Preferences
MUSIC
TECHNOLOGY
PREFERENCE
AND
DETERMINING
FACTORS
Factors
in
Music
Technology
Preference
Favorite
Feature
Demographics
and
Music
Technology
Preference
Musical
Experience
and
Music
Technology
Preference
Music
Preference
and
Music
Technology
Preference
MUSIC
TECHNOLOGY
INFLUENCING
PREFERENCE
AND
SPENDING
Effects
on
Music
Preference
Listen
More
Wider
Range
of
Genres
Deeper
Within
Familiar
Genres
More
Sharing
Music
Effects
on
Spending
More
Buying
Buying
Different
Music
Buying
Concert
Tickets
Music
is
Bigger
DISCUSSION
(61-65)
REFERENCES
(66)
APPENDIX
(67-75)
2
4. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Introduction
It’s
your
last
high
school
gym
class
before
moving
to
college.
You
return
to
your
locker
to
find
the
lock
broken
and
someone
has
stolen
your
iPod
and
entire
music
collection
out
of
your
backpack.
Fearing
you
may
turn
to
a
life
of
digital
piracy
and
cyber
crime,
your
parents
purchase
the
new
32GB
iPod
Touch
that
holds
7,000
songs
and
connects
to
the
Internet.
Having
lost
all
of
your
music,
you
research
some
popular
music
applications.
It’s
May
20,
2012
and
as
of
April
30th,
the
iTunes
store
offered
over
28
million
songs.
How
do
you
choose
which
0.025%
to
buy?
Do
you
instead
rely
on
the
endless
stream
of
YouTube
videos
your
friends
share
on
Facebook,
or
do
you
create
a
Pandora
station
like
the
150
million
other
Americans
that
enjoy
personalized
music
recommendations?
The
limited
bandwidth
of
AM/FM
radio
necessitated
a
popularity
contest
for
songs,
but
the
technical
constraints
of
terrestrial
radio
don’t
apply
to
digital
music.
The
combination
of
nearly
unlimited
music
choice
and
a
wide
variety
of
music
sites
make
modern
music
experiences
vastly
more
personal
than
terrestrial
radio.
The
proliferation
of
song
recommendations,
shared
playlists,
and
music
blogs
attest
to
the
power
of
the
digital
music
experience.
After
a
particularly
inspiring
lecture
on
digital
media
by
Professor
Fred
Turner
last
year,
I
designed
and
conducted
a
survey
on
Pandora
use
for
a
Communication
course
at
Stanford.
Asking
112
respondents
if
they
had
ever
bought
an
unfamiliar
song
after
hearing
it
only
once
on
Pandora,
59
students
equaling
53%
of
the
sample
indicated
that
they
had.
Even
more
surprising,
15
students
(13%)
indicated
they
had
bought
an
entire
album
after
3
5. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
hearing
a
song
from
it
for
the
first
time
on
Pandora.
Although
I
recognized
that
the
Stanford
students
that
made
up
the
survey
sample
were
not
representative
of
Pandora’s
entire
user
base,
it
seemed
likely
that
a
significant
percentage
of
Pandora’s
users
were
purchasing
unfamiliar
music
as
well.
I
wanted
to
know
how
these
new
transactions
would
affect
the
music
industry,
amidst
declining
sales
and
a
torrent
of
illegal
filesharing
applications.
Literature
Review
The
Long
Tail
of
Digital
Music:
New
Technology
Revealing
Niche
Preferences
After
Chris
Anderson
published
his
article
“The
Long
Tail”
in
Wired
magazine
in
October
of
2004,
it
quickly
became
the
most
cited
article
in
Wired’s
history,
and
his
book
became
one
of
the
most
influential
business
books
of
the
decade
(Anderson).
Using
e-‐
commerce
data
that
had
been
historically
restricted
to
executives,
the
book
outlines
Anderson’s
theory
that
the
Internet
has
expanded
the
range
of
effective
inventory
from
a
limited
number
of
“hits”,
as
seen
on
WalMart
and
Blockbuster
shelves,
to
nearly
infinity.
Since
the
post-‐WWII
era
of
TV
and
radio,
businesses
have
traditionally
capitalized
on
the
power
of
the
top
100
or
even
100,000
mainstream
products,
ignoring
all
the
books,
songs,
and
goods
that
didn’t
make
the
charts
(Figure
1).
But
as
both
Figure
1
4
6. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Anderson
and
Lessig
point
out,
the
recent
success
stories
of
Amazon,
Netflix,
and
iTunes
prove
that
the
companies
providing
customers
with
the
most
choices
and
the
most
effective
ways
to
navigate
them
can
earn
as
much
as
40%
of
their
revenue
from
products
along
“the
long
tail.”
During
the
age
of
FM
radio
and
record
stores,
limited
inventory
and
constrained
choice
contributed
to
widespread
adoption
of
popularized
music
preferences.
Retailers
optimizing
limited
shelf
space
and
FM
radio
DJs
seeking
to
maximize
listenership
reinforced
a
culture-‐wide
fascination
with
top
charts
and
superstars.
However,
as
digital
music
technologies
continue
to
proliferate,
the
seemingly
unlimited
number
of
musical
choices
and
their
innovative
recommendation
systems
are
shaping
listeners’
preferences
and
consumption
patterns
in
new
ways.
Although
the
possible
ramifications
of
unlimited
choice
and
user
profiling
are
numerous,
I
expect
these
technologies
to
both
widen
and
deepen
the
music
preferences
of
their
users.
In
other
words,
the
unique
features
of
new
music
services
will
not
only
enable
the
tracking
of
the
Long
Tail,
but
also
shift
demand
to
make
it
even
longer.
The
purpose
of
this
study
is
two-‐fold:
to
determine
the
most
salient
factors
that
shape
listeners’
music
preferences
and
choice
of
music
service,
and
to
enhance
our
understanding
of
new
music
technologies’
impact
on
users.
Throughout
history,
from
Mary
Shelley’s
Frankenstein
to
George
Orwell’s
Nineteen
Eighty-Four,
the
idea
of
technological
determinism
has
caused
society
to
irrationally
view
and
fear
technology
as
an
autonomous
juggernaut,
sometimes
causing
the
restriction
of
tools
that
extend
humanity’s
potential
(McGinn).
A
technological
determinist
might
use
phrases
like
identity
theft,
violation
of
privacy,
and
entertainment
piracy
to
describe
the
5
7. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Internet’s
relationship
with
its
users.
In
his
book
Remix,
Lessig
argues
that
the
digitization
of
culture
and
the
economy
is
a
positive
change
to
be
embraced
and
understood,
rather
than
restricted
and
criminalized.
He
protests
against
outdated
copyright
laws
now
criminalizing
creative
actions,
calling
for
copyright
law
reform
to
realize
the
full
potential
of
the
new
hybrid
of
commercial
and
sharing
economies.
After
detailing
both
economies
individually,
he
argues
that
the
Internet’s
new
hybrid
economy
is
a
fusion
of
both
voluntary
collaboration
and
traditional
commerce.
He
provides
several
examples
of
companies
—
including
Netflix,
Amazon,
Google,
YouTube,
and
Second
Life
—
and
mechanisms,
such
as
user
reviews
and
recommendations,
crowdsourcing,
and
Anderson’s
Long
Tail
principle,
that
support
his
argument
that
the
new
hybrid
economy
is
“a
model
of
success,
not
a
compromise
of
profit.”
McGinn
also
testifies
to
the
vital
importance
of
resisting
technological
determinism,
acknowledging
technology
and
society
as
interdependent
and
co-‐evolutionary,
and
monitoring
the
unique
powers
associated
with
each.
These
ideas
guided
this
study
throughout
the
various
stages
of
literature
review,
data
collection,
and
analysis.
Contrary
to
technological
deterministic
perspectives,
more
and
more
IT-‐based
media
channels
and
corporations
are
capitalizing
on
their
control
over
technology
to
shape
user
interactions
online.
Amazon’s
book
recommendation
feature
is
one
example
of
a
navigational
tool
intended
to
both
maximize
profit
and
cater
to
users’
preferences.
As
Anderson
points
out
in
the
first
chapter
of
The
Long
Tail,
Amazon’s
pairing
of
the
best
seller
Into
Thin
Air
with
the
lesser-‐known
Touching
the
Void
via
its
recommendation
feature
created
a
powerful
positive
feedback
loop
of
both
interest
and
revenue.
By
categorizing
media
based
on
similarity,
rather
than
—
or
in
addition
to
—
listing
them
by
popularity,
6
8. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
these
institutions
and
corporations
better
serve
both
the
user
and
the
long
tail
of
the
market.
In
his
book
The
Wisdom
of
the
Crowds,
author
James
Surowiecki
explores
the
notion
that
a
large
group
of
people
is
more
innovative
and
better
at
problem
solving
than
a
small
elite
creative
team,
concluding
that
this
technique
of
“crowdsourcing”
has
enormous
potential
and
has
already
begun
to
shape
online
interaction
(Surowiecki).
Taking
Surowiecki’s
advice,
the
popular
online
DVD
rental
service
Netflix
conducted
a
nearly
three-‐year-‐long
public
competition
for
an
improved
Netflix
recommendation
algorithm,
making
Netflix
usage
data
freely
available
in
an
effort
“to
substantially
improve
the
accuracy
of
predictions
about
how
much
someone
is
going
to
enjoy
a
movie
based
on
their
movie
preferences”
(http://www.netflixprize.com).
The
winning
team’s
algorithm
is
yet
another
user-‐centered
tool
used
to
connect
niche
market
products
and
media
to
their
customers,
directly
facilitating
the
expansion
of
the
long
tail.
Studying
Music
Preference
Many
researchers
have
conducted
studies
revealing
correlations
between
demographical
information,
such
as
age,
gender
and
education,
and
music
preferences.
LeBlanc
et
al.
created
an
overall
music
preference
index
to
measure
subjects’
total
preferences
across
genres
and
compared
responses
between
different
age
groups.
After
surveying
2,262
respondents,
the
researchers
found
that
the
music
preference
index
declined
in
elementary
students,
rose
from
high
school
to
college,
and
declined
after
college
(LeBlanc
et
al.,
1996).
While
these
findings
may
not
provide
a
means
to
improve
music
recommendation
algorithms,
statistically
significant
correlations
between
age
and
7
9. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
preferences
for
specific
songs,
artists,
and
genres
would
certainly
help
predict
listener
reactions.
Although
surveys
may
be
able
to
determine
linkages
between
age
and
genre
preferences
(as
this
study
will
show),
this
method
is
obviously
not
feasible
for
collecting
larger
data
sets
regarding
artist
or
song
preference.
However,
this
is
one
of
many
examples
of
how
digital
media’s
growing
trend
of
“thumb”
or
“like”
feedback
could
be
utilized
by
companies
like
iTunes,
Pandora,
and
YouTube.
While
most
music
recommendation
sites
focus
on
users’
preferences
and
musical
similarities
between
songs,
several
Taiwanese
researchers
(Suh-‐Yin
Lee
et
al.,
2009)
investigated
the
use
of
emotion-‐based
music
discovery
within
the
context
of
motion
picture
scores.
Constructing
an
original
algorithm
called
the
Music
Affinity
Graph-‐Plus,
Suh-‐Yin
Lee
et
al.
achieved
an
impressive
85%
accuracy
in
matching
queried
emotions
with
music
of
the
same
emotions.
While
these
results
and
the
growth
of
music
recommendation
sites
like
Stereomood
and
Music
for
Emotion
prove
the
potential
of
emotion-‐based
song
sorting
and
recommendation,
such
an
approach
has
yet
to
draw
a
fraction
of
the
audience
of
iTunes,
Pandora
or
YouTube.
In
acknowledgement
of
its
potential,
this
study
will
also
survey
respondents
on
their
level
of
demand
for
emotion-‐based
music
recommendation.
In
2009,
Gaffney
and
Rafferty
conducted
a
study
investigating
users’
knowledge
and
use
of
social
networking
sites
and
folksonomies
(user-‐generated
taxonomies),
focusing
on
the
potential
of
social
tagging
to
aid
in
the
discovery
of
independent
music.
Examining
the
four
music
discovery
sites
MySpace,
Lastfm,
Pandora
and
Allmusic
through
user
surveys
and
interviews,
they
found
that
although
respondents
use
social
networking
sites
for
music
discovery,
they
are
generally
unaware
of
folksonomic
approaches
to
music
discovery.
8
10. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Furthermore,
those
who
do
use
and
contribute
to
folksonomies
are
mostly
self-‐serving
in
their
motives
(Gaffney
and
Rafferty,
2009).
While
Gaffney
and
Rafferty
state
that
their
study
rests
upon
the
assumption
that
music
recommendation
and
social
networking
sites
push
users
and
revenue
toward
the
Long
Tail,
they
make
no
attempt
to
quantify
the
impact
of
any
particular
site
on
the
time
or
money
users
spend
on
Long
Tail
songs.
Additionally,
the
landscape
of
music
discovery
sites
has
changed
dramatically
since
they
conducted
the
study,
especially
in
the
case
of
Pandora’s
rapid
growth.
Unfortunately,
the
vast
majority
of
studies
involving
music
preferences
use
a
nomothetic
approach
to
choose
one
or
two
particular
factors
to
test,
whether
for
simplicity
or
convenience.
Christenson
and
Peterson
built
upon
earlier
studies
of
gender
and
music
genre
preferences
by
including
many
“metagenres”
previously
disregarded
by
social
scientists.
Consistent
with
similar
studies,
they
found
convincing
evidence
that
gender
predisposes
people
to
certain
music
preferences;
for
example,
that
females
gravitate
toward
popular
music
and
males
gravitate
away
from
it.
While
this
study
contributes
a
piece
of
the
music
preference-‐mapping
puzzle,
Christenson
and
Peterson
admit,
“the
underlying
structure
of
music
preference
cannot
be
accounted
for
by
reference
to
two
or
three
factors,
but
is
multivariate”
(Christenson
et
al,
1988).
At
this
point,
the
need
for
an
idiographic
approach
to
music
preferences
is
clear.
This
study
is
partially
driven
by
the
lack
of
a
multivariate
or
idiographic
study
comparing
the
relative
impacts
of
age,
emotion,
social
network,
choice
of
digital
music
service,
and
more
factors,
on
music
preference.
iTunes,
Pandora
and
YouTube
certainly
have
a
wealth
of
data
on
their
services’
use
and
users,
but
data
points
like
relative
9
11. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
preference
between
services,
musical
education
and
experience
of
users,
and
listening
environment
are
often
ignored.
Not
only
do
I
find
this
information
intriguing,
I
suspect
it
could
prove
incredibly
relevant
to
both
music
marketing
strategies
and
music
recommendation
technology.
In
addition
to
enabling
the
examination
of
underlying
patterns
between
these
variables,
the
collected
surveys
provide
a
basis
for
predicting
economic
shifts
in
the
music
industry.
It
is
expected
that
by
aligning
recommendations
with
each
unique
users’
profile
rather
than
the
most
popular
songs,
new
music
technologies
like
iTunes,
Pandora,
and
YouTube
both
please
users
and
support
more
artists
further
down
the
Long
Tail.
Furthermore,
the
findings
presented
in
this
study
reveal
significant
relationships
between
variables
that
have
thus
far
been
excluded
from
music
recommendation
algorithms.
Methods
Sampling
Given
my
interest
in
the
college
student
demographic
and
my
immediate
network
of
friends
and
family,
I
focused
my
recruiting
efforts
on
three
different
colleges:
Stanford
University,
Glendale
Community
College
(GCC),
and
Arizona
State
University
(ASU).
Stanford
was
the
first
and
most
convenient
sampling
frame
for
me
as
a
Stanford
undergrad,
providing
38
respondents.
My
parents,
both
professors
at
Glendale
Community
College,
invited
their
students
to
take
the
survey
and
added
71
students
to
the
sample.
Last,
I
sent
a
brief
Facebook
message
to
recruit
ASU
students
from
my
high
school
network.
Response
and
completion
rates
were
lowest
at
ASU,
with
9
students
completing
the
survey.
The
10
12. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
shortage
of
both
time
and
funded
incentives
ruled
out
a
random
sampling
of
college
students,
but
I
minimized
potential
biases
by
recruiting
from
several
different
schools.
While
a
realistic
distribution
between
schools
would
have
been
preferable,
the
number
of
college
students
who
volunteered
for
my
unpaid,
20-‐minute
survey
was
significantly
higher
than
I
anticipated.
Concepts
The
first
page
of
the
survey
addressed
respondents’
demographics,
including
age,
gender,
hometown,
current
school,
and
competence
with
computers.
Free
response,
or
open-‐ended,
answer
formats
will
be
used
for
age,
hometown,
and
current
school,
while
gender
and
computer
competence
will
use
closed-‐ended
questions.
The
question
“Please
categorize
your
competence
using
computers”
will
include
the
options
“Advanced”,
“Average”,
“Basic”,
and
“NoneVery
Little.”
These
items
were
carefully
chosen
for
clarity
and
appropriateness,
to
ensure
optimal
accuracy.
The
demographic
variables
were
chosen
for
potential
to
influence
both
music
preference
and
music
technology
preference.
The
second
page
of
the
questionnaire
features
units
of
analysis
addressing
respondents’
musical
experience,
in
order
to
gauge
how
each
influences
music
preference.
Each
concept
will
contribute
to
an
index
summarizing
overall
musical
experience,
assigning
quantitative
values
to
qualitative
responses
where
appropriate.
First,
subjects
were
asked
the
open-‐ended
question
“Approximately
how
many
hours
per
week
do
you
spend
listening
to
music?”
Next,
respondents
selected
the
option,
“Which
best
describes
the
frequency
of
your
online
music
listening?”
from
the
list:
“Rarely”,
“Sometimes”,
“Often”,
and
“All
the
Time.”
Then,
using
a
check-‐all
question
format,
respondents
indicated
the
school
years
11
13. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
during
which
they
took
at
least
one
music
class,
with
the
options
“Elementary
(K-‐8th)”,
“High
School”,
“College”,
and
“None”.
Next,
subjects
indicated
how
many
years
they
have
taken
musical
instrument
lessons
(outside
of
school)
with
free
response.
Finally,
the
closed-‐ended
question
“Do
you
currently
play
an
instrument?”
was
followed
by
the
contingency
question
of
“How
many
years
have
you
played
an
instrument?”
In
order
to
maintain
both
accuracy
and
the
respondent’s
attention,
these
questions
and
question
formats
were
chosen
based
on
their
clarity,
relevance,
and
brevity
for
each
unit
of
analysis.
Both
the
index
and
individual
units
of
analysis
will
be
used
in
determining
the
most
salient
factors
in
music
preference.
The
next
page
of
the
survey
investigated
subjects’
music
preferences.
For
the
purposes
of
this
study,
music
preferences
were
defined
as
genres
that
an
individual
simply
enjoys
listening
to.
As
mentioned
earlier,
genres
are
the
most
feasible
unit
of
analysis
for
music
preferences
using
a
survey,
given
the
large
numbers
of
artists
and
songs
in
existence.
Using
a
matrix
question,
participants
were
asked,
“What
are
your
attitudes
toward
the
following
music
genres?”
In
addition
to
operationalizing
this
concept
with
multiple
levels
of
enjoyment
(dislike,
neutral,
like,
and
love),
the
list
of
genres
included
those
common
throughout
all
three
music
services
in
question
(see
Appendix
for
full
survey).
The
primary
issue
carefully
controlled
in
this
question
was
the
respondent’s
understanding
of
music
genres.
For
this
reason,
the
selected
genres
were
pragmatically
selected
for
distinctness
from
one
another.
While
this
potential
confound
has
been
minimized,
it
cannot
be
fully
eliminated
without
including
potentially
distracting
full
definitions
of
each
genre.
12
14. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Although
the
varying
levels
of
preference
within
genres,
or
the
“depth”
of
music
preference,
have
been
accounted
for
with
the
four
options
listed
on
this
genre,
the
following
questions
utilized
a
different
approach
to
measure
the
same
concept.
After
subjects
indicated
their
favorite
genre
from
the
same
list,
they
were
asked,
“Within
your
favorite
musical
genre,
approximately
what
percentage
of
artists
and
songs
that
you
know
do
you
like?”
with
the
options
“0-‐20%”,
“21-‐40%”,
“41-‐60%”,
“61-‐80%”,
and
“81-‐100%”.
Next,
the
questionnaire
asked
the
closed-‐ended
question,
“Of
all
the
“top
40”
popular
music
you’ve
heard,
you
like:”
where
subjects
chose
between
“All
or
almost
all”,
“Most”,
“About
half”,
“Some”,
“None”,
and
“I
don’t
pay
attention
to
top
40
charts”.
Finally,
the
matrix
question
format
asked
respondents
about
the
importance
of
the
following
attributes
in
determining
whether
or
not
they
like
a
song.
These
attributes
included
“familiarity”,
“popularity”,
“fits
my
mood”,
“artistic
talent”,
“lyrics”,
and
“friends’
preferences”,
and
were
classified
as
either
“Not
important”,
“Somewhat
important”,
“Very
important”,
and
“Extremely
important”.
Again,
these
closed-‐ended
questions
ensured
that
respondents
measure
their
perspectives
by
the
same
standards,
which
was
one
of
the
primary
reasons
for
using
the
online
survey
approach.
The
next
page
of
the
survey
examined
the
respondent’s
music
listening
environment.
Using
the
matrix
question
format,
the
respondents
indicated
how
often
they
listen
to
music
in
each
of
the
following
environments
and
activities,
including
“At
home”,
“In
the
car”,
“At
work”,
“By
yourself”,
“With
a
few
friends”,
“At
a
party”,
“While
studying”,
and
“While
sleeping”.
Potential
responses
utilized
the
Thurstone
scale,
and
included
“Never”,
“Rarely”,
“Sometimes”,
“Often”,
and
“Always”.
While
these
activities
and
locations
may
have
overlapped
somewhat,
each
item
was
chosen
for
relevance
and
potential
to
13
15. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
influence
music
preferences.
Perhaps
the
most
pivotal
of
the
entire
survey,
the
next
matrix
question
asked
respondents
to
rank
their
“Favorite”,
“2nd
Favorite”,
and
“3rd
Favorite”
music
services
between
iTunes,
Pandora,
and
YouTube;
respondents
could
also
select
a
fourth
option,
“Never
use
it”.
These
music
services
were
selected
because
they
are
widely
used,
legal
alternatives
to
music
piracy
and
because
I
wanted
to
understand
how
they
are
reshaping
the
music
industry
from
individual
users’
perspectives.
The
following
three
pages
contained
contingency
questions
depending
on
whether
respondents
use
the
services
iTunes,
Pandora,
and
YouTube.
Using
similarly
structured
matrix
questions,
these
pages
sought
to
ascertain
the
perceived
impact
of
each
service
on
users’
music
preferences
and
spending
habits
based
on
the
Likert
scale.
For
example,
respondents
were
asked
to
indicate
various
levels
of
agreement/disagreement
with
the
statements
“As
a
result
of
using
Pandora,”
“I
listen
to
music
more
often”,
“I
listen
to
a
wider
range
of
genres”,
“I
listen
to
more
music
within
the
genres
I
like”,
“I
share
music
with
my
friends
more”,
“I
buy
more
music”,
“I
buy
different
music
than
I
would
have
otherwise”,
“I
have
bought
concert
tickets
that
I
wouldn’t
have
otherwise”,
and
finally
“music
plays
a
bigger
role
in
my
life.”
Because
these
questions
directly
apply
to
the
hypothesis
of
this
study,
they
did
not
contain
negative
answers
or
answers
that
might
bias
results,
and
there
were
several
different
units
or
elements
intending
to
measure
the
same
concept.
The
final
question
on
each
page
asked
respondents
to
choose
their
favorite
feature
of
each
service,
choosing
between
“customizability/personalization”,
“its
interface”,
“its
wide
selection
of
music”,
“playlisting
and
song
recommendation”,
and
“Other:
Please
Specify”.
The
Likert
scale
was
chosen
both
for
its
speed
and
appropriateness
in
this
case,
and
the
use
of
similar
14
16. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
questions
on
the
pages
for
all
three
music
technologies
will
ensure
a
common
standard
of
measurement
and
enable
a
closer
comparison
of
their
relative
impacts
on
users.
Additional
measures
taken
to
ensure
accuracy
of
questionnaire
responses
include
carefully
ordering
the
questions
in
ascending
order
of
difficulty,
eliminating
double-‐
barreled
questions,
providing
questionnaire
instructions,
and
pretesting
the
questionnaire
on
a
number
of
classmates.
Wherever
possible,
questions
with
similar
potential
responses
were
grouped
as
matrix
questions
to
quicken
response
times
and
maintain
a
higher
response
rate.
Furthermore,
to
improve
the
relevance
of
the
questionnaire,
the
questions
that
may
not
apply
to
all
respondents
have
been
formatted
as
contingency
questions.
Coding
and
Data
Analysis
In
order
to
analyze
the
results
of
the
online
questionnaire,
I
downloaded
the
CSV
file
of
raw
data
for
138
respondents
from
www.rationalsurvey.com
and
imported
it
into
SPSS
Statistics,
which
I
purchased
through
Stanford
Software
Licensing.
Preparing
the
survey
data
for
analysis
involved
several
steps,
the
first
of
which
was
removing
the
incomplete
and
age-‐inappropriate
cases.
After
deleting
the
few
cases
of
respondents
who
were
no
longer
in
college
or
hadn’t
completed
the
survey,
I
ended
up
with
125
total
respondents.
Next,
I
defined
each
of
the
variable
properties
by
classifying
them
as
either
ordinal,
nominal,
or
scale.
I
then
used
a
number
of
coding
techniques
to
enable
tests
of
correlation,
assigning
numeric
values
to
all
textual
responses.
For
example,
“Never”
=
1,
“Rarely”
=
2,
“Sometimes”
=
3,
and
so
on.
Next,
I
assigned
corresponding
labels
to
the
numeric
values
to
facilitate
my
interpretation
of
statistical
procedures.
Due
to
the
relatively
large
number
of
15
17. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
questions,
73
in
total,
the
various
strategies
used
to
assign
numeric
values
will
be
discussed
in
tandem
with
the
results
and
analysis
of
each
variable.
Respondents
Due
to
the
financial
and
temporal
constraints
of
this
study,
the
online
survey
was
distributed
to
a
convenient
sample.
Of
the
125
college
students
who
completed
the
survey,
66
(53%)
were
male
and
59
were
female.
Since
I
was
targeting
the
college
student
demographic,
respondents
had
an
average
age
of
21.43
with
a
standard
deviation
of
4.1.
In
response
to
the
third
question
of
hometown,
63
respondents
(50%)
indicated
they
were
from
Arizona,
40
of
which
were
from
Phoenix.
Another
26
respondents
(21%)
hail
from
various
cities
in
California,
and
the
remaining
subjects’
hometowns
included
18
states
and
4
locations
outside
the
United
States.
Although
the
survey’s
findings
may
have
a
slightly
southwest/west
coast
bias,
I
found
this
geographical
spread
acceptable
given
the
study’s
constraints.
Results
While
the
online
questionnaire
consisted
of
five
sections,
analysis
of
results
was
divided
into
three
sections:
music
preference
and
contributing
factors,
choice
of
music
technology
and
contributing
factors,
and
impacts
of
each
music
technology
on
preference
and
spending.
Each
of
the
three
sections
contains
several
different
variables
that
measure
similar
ideas
to
reinforce
findings.
Since
nearly
all
variables
were
coded
into
numeric
values
and
most
of
these
were
ordinal,
a
simple
function
in
SPSS
created
a
spreadsheet
of
all
correlations
between
variables
and
designated
those
of
significance
at
the
.05
and
the
16
18. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
.01
levels.
Because
the
survey
was
distributed
to
a
convenient
sample,
statistically
significant
correlations
cannot
be
generalized
to
larger
populations.
However,
these
findings
may
be
used
to
speculate
about
how
college
students
consume
music
online
and
how
technology
influences
their
preferences.
Due
to
the
length
and
comprehensiveness
of
the
survey,
the
three
results
sections
include
only
the
most
significant
and/or
surprising
results.
Music
Preference
and
Determining
Factors
Song
Preference
Perhaps
the
most
direct
question
addressing
the
factors
affecting
music
preference,
question
16
asked
respondents
to
indicate
the
importance
of
six
attributes
in
determining
whether
or
not
they
like
a
particular
song.
In
the
interest
of
saving
respondents’
time,
I
selected
attributes
that
were
highly
likely
candidates
of
influence.
Based
on
my
experience
with
music
and
friends’
preferences,
I
expected
popularity
and
friends’
preferences
to
rank
the
highest.
After
all,
it
seems
like
the
two
most
persuasive
reasons
to
check
out
a
new
song
are
that
friends
love
it
or
everybody
else
does.
I
also
speculated
that
lyrics
would
receive
polarized
ratings
of
importance,
and
that
“fitting
the
mood”
would
rank
as
more
important
than
most
of
the
other
attributes.
In
hindsight,
the
attribute
“artistic
talent”
should
have
either
been
reworded
as
“musicianship”
or
juxtaposed
with
“producer’s
talent”;
as
it
stands,
it
seems
hard
to
believe
many
respondents
would
indicate
that
they
don’t
care
if
the
artist
is
talented.
17
19. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
The
results
for
these
six
attributes
were
fairly
surprising,
and
have
tremendous
implications
for
the
improvement
of
music
recommendation.
First,
my
predictions
about
popularity
and
friends’
preferences
were
almost
completely
wrong;
respondents
rated
both
lowest,
between
“somewhat
important”
and
“not
important”,
on
average.
Furthermore,
average
responses
for
familiarity
were
positioned
just
above
“somewhat
important”,
illustrating
users’
comfort
with
music
exploration.
Next,
lyrics
ranked
third
with
an
average
response
just
above
“very
important”,
in
contrast
with
my
expectation
that
some
respondents
preferring
instrumental
music
or
songs
by
Justin
Bieber
would
consider
lyrics
of
minimal
importance.
Interestingly
enough,
importance
of
lyrics
was
negatively
correlated
with
preferences
for
electronic
music
and
positively
correlated
with
R&B/Soul,
both
of
which
make
sense.
Although
I
guessed
“fitting
the
mood”
and
“artistic
talent”
would
rank
fairly
high,
I
didn’t
expect
them
to
rank
highest
overall
with
an
average
response
between
“very
important”
and
“extremely
important.”
While
these
findings
don’t
prescribe
an
ideal
way
to
incorporate
each
attribute
into
song
recommendations,
they
do
suggest
that
the
traditional
mechanisms
of
music
discovery
are
far
less
effective
than
new
recommendation
technologies
that
utilize
this
information.
18
20. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Determining
Factors
of
Song
Preference
-
Mean
Response
4
Not
Important
Very
Important
3.5
3
2.5
2
1.5
1
0.5
0
Familiarity
Popularity
Fits
Mood
Artistic
Lyrics
Friends'
Talent
Preferences
Figure
2
Admittedly
these
findings
are
self-‐reported
and
it’s
entirely
possible
that
people
simply
don’t
want
to
recognize
how
much
a
song’s
popularity
or
their
friends’
tastes
influences
their
own
preference.
To
approach
the
question
of
how
popularity
impacts
song
preference
from
a
different
angle,
I
examined
the
frequency
of
responses
for
question
15
that
addressed
feelings
toward
top
40
music
(Figure
3).
The
average
response
was
halfway
between
“Some”
and
“About
Half”,
suggesting
that
the
previous
findings
were
correct.
Furthermore,
a
significant
portion
of
respondents,
reaching
almost
20%
of
the
sample,
state
that
they
either
don’t
pay
attention
to
top
40
charts
or
they
like
none
of
the
songs
on
them.
This
implies
that
although
many
users’
music
tastes
are
still
influenced
by
top
40
music
charts,
these
indicators
of
popularity
may
be
losing
the
power
they
once
held
over
AM/FM
radio
audiences.
19
21. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Preference
for
Top
40
Music
All
or
almost
all
Most
About
half
Some
None
I
don't
pay
attention
to
top
40s
0
5
10
15
20
25
30
35
40
Figure
3
Genre
Preference
The
survey’s
first
and
simplest
measure
of
respondents’
music
preferences
entailed
rating
fourteen
distinct
music
genres.
The
rating
scale
included
“dislike”
=
-‐1,
“neutral”
=
0,
“like”
=
1,
and
“love”
=
2.
Rock,
Alternative,
and
Hip
Hop/Rap
scored
the
highest
on
average
among
the
125
respondents,
with
Latin
and
World
ranking
lowest
(Figure
4).
Additionally,
the
ratings
for
Hip
Hop/Rap
and
Country
were
the
most
polarized,
yielding
standard
deviations
over
1.
20
22. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Genre
Preferences
-
Mean
2
1.5
1
0.5
0
Figure
4
After
seeing
how
respondents
ranked
each
genre
independently,
I
wanted
to
know
how
genres
clustered
together
based
on
these
ratings.
Using
multidimensional
scaling
in
SPSS,
I
determined
the
coordinates
for
each
genre
to
create
a
Euclidean
distance
model
that
provides
a
visualization
of
the
similarities
between
genres
based
on
the
respondents’
rankings
(Figure
5).
Though
the
interpretation
of
the
axes
is
essentially
meaningless,
this
graph
is
simply
a
way
to
visualize
perceived
similarities
between
genres
according
to
respondents.
For
the
most
part,
these
groupings
of
genres
make
sense
when
considering
musical
similarities,
probable
listening
environment,
and
several
other
characteristics.
21
23. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Derived
Stimulus
ConSiguration
-
Euclidean
Distance
Model
Rock
Alternative
Jazz
Classical
Electronic
Vocal
R&B/Soul
Hip
Hop/Rap
Reggae
World
Pop
Dance
Latin
Country
Figure
5
Next,
respondents
indicated
their
favorite
genre,
choosing
from
these
fourteen
and
“Other”
(Table
1).
Consistent
with
Christenson
and
Peterson’s
findings,
the
“Other”
category
ranked
fourth
largest
with
14
respondents,
verifying
the
importance
of
accounting
for
“metagenres”
and
subgenres
in
music
classification
and
recommendation.
However,
for
the
purposes
of
this
analysis,
metagenres
and
subgenres
were
ignored
to
facilitate
quick
and
accurate
responses.
Favorite Genre Respondents Favorite Genre Respondents
Rock 29 Pop 6
Hip Hop/Rap 18 Reggae 5
Alternative 16 Dance 4
Other 14 Classical 2
Country 13 Jazz 2
22
24. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
R&B/Soul 7 Vocal 2
Electronic 6 Latin 1
Table
1
While
the
average
genre
ratings
and
favorite
genre
for
all
respondents
are
informative
and
fairly
interesting,
these
metrics’
true
function
is
to
provide
a
basis
for
correlations
between
subgroups
of
the
college
student
sample.
These
subgroups
are
drawn
from
four
categories
of
variables:
music
preferences,
user
demographics,
musical
experience,
and
listening
environment.
Factors
in
Music
Preference
Correlations
Between
Genre
Preferences
By
having
almost
daily
conversations
about
music
preferences
with
friends
and
strangers
for
at
least
ten
years,
I
developed
a
few
theories
regarding
relationships
between
genres.
I
got
the
sense
that
people
who
listened
to
at
least
one
niche
genre
tended
to
like
almost
all
others
as
well,
and
people
who
preferred
popular
music
had
much
narrower
tastes
for
genres.
While
portions
of
the
Euclidean
distance
model
conveyed
similar
information,
the
best
way
to
test
this
claim
was
through
bivariate
correlations.
Using
the
spreadsheet
of
Spearman
correlations,
I
calculated
the
number
of
significant
correlations
between
genres
and
found
two
groups
of
genres
separating
from
one
another.
I
created
one
table
using
the
genres
with
many
positive,
significant
correlations
(Table
2)
and
another
for
those
with
fewer
positive
correlations
and
more
negative
correlations
with
other
genres
(Table
3).
23
25. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Table
2
Significant
Correlations
With
Other
Genres
Mostly
Niche
Genres
Positive
Negative
Reggae
7
0
Vocal
7
0
Latin
7
0
World
7
1
Classical
7
1
Jazz
7
1
Alternative
5
0
R&B/Soul
5
0
Table
3
Mostly
Popular
Significant
Correlations
With
Other
Genres
Genres
Positive
Negative
Hip
Hop/Rap
6
2
Dance
4
1
Electronic
3
0
Pop
3
1
Rock
2
1
Country
2
1
These
tables
provide
strong
evidence
supporting
my
claim
that
users
who
like
one
niche
genre
are
likely
to
enjoy
many
more.
Not
only
does
it
show
that
niche
genres
are
positively
correlated
with
many
others
(Table
2),
the
more
popular
genres
have
twice
as
many
negative
correlations
(Table
3).
Hip
hop/rap
was
the
one
genre
positioned
in
between
the
distinct
groups
but
was
included
in
the
second
table
because
it
had
the
most
negative
correlations.
These
findings
seem
to
confirm
my
hypothesis
that
fans
of
niche
genres
have
wider
preferences
and
fans
of
popular
genres
have
narrower
preferences.
Demographics
and
Genre
Preferences
24
26. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
I
predicted
a
few
demographical
variables
from
the
first
page
of
the
survey
would
correlate
with
genre
preferences
so
I
examined
their
Spearman
correlations.
As
I
expected
based
on
Christenson
and
Peterson’s
study
and
my
own
experience,
gender
was
negatively
correlated
with
preferences
for
Dance,
Pop,
and
Country.
Since
I
assigned
the
values
“Female”
=
1
and
“Male”
=
2,
this
means
that
females
are
more
likely
to
enjoy
these
three
genres
and
males
are
less
likely.
While
this
isn’t
an
especially
groundbreaking
conclusion,
it
both
makes
sense
and
matches
up
with
Christenson
and
Peterson’s
findings,
adding
a
degree
of
confidence
to
other
correlations
with
genre
preferences.
I
found
another
fairly
predictable
correlation
between
age
and
preference
for
jazz
and
classical
music.
Since
the
correlations
were
both
significant
and
positive,
we
can
conclude
these
two
genres
are
more
appealing
to
older
respondents.
While
this
isn’t
incredibly
surprising,
it’s
interesting
to
consider
that
the
standard
deviation
of
respondents’
age
was
only
4.1.
This
means
that
just
a
few
years
of
age
separates
the
fans
of
classical
and
jazz
from
those
who
enjoy
these
genres
much
less.
It’s
difficult
to
determine
whether
this
is
caused
by
a
generational
difference
or
perhaps
a
difference
in
maturity
levels,
but
simply
knowing
the
correlation
could
improve
song
recommendations
significantly.
On
the
other
hand,
I
found
an
unexpected
correlation
between
competence
using
computers
and
preferences
for
electronic
music
at
the
.01
level.
Put
simply,
the
more
experience
respondents
had
with
computers,
the
more
likely
they
were
to
like
electronic
music.
While
this
correlation
makes
sense
because
the
creation
of
electronic
music
requires
digital
signal
processing,
I
was
surprised
that
electronic
music
was
both
the
only
25
27. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
genre
correlated
with
computer
skills
and
it
was
significant
at
the
.01
level.
Again,
these
correlations
at
the
.01
level
can’t
be
generalized
to
the
population,
but
instead
indicate
particularly
strong
correlations
between
variables
for
the
college
students
in
the
convenient
sample.
This
particular
correlation
between
a
genre
preference
and
computer
skills,
a
characteristic
seemingly
unrelated
to
music,
begs
the
question
of
how
many
other
personality-‐based
characteristics
correlate
with
music
preference.
Although
I
was
expecting
a
greater
number
of
correlations
between
demographic
information
and
genre
preferences,
those
that
I
found
present
convincing
evidence
for
the
implementation
of
demographics
in
music
recommendation
technology.
iTunes,
Pandora
and
YouTube
already
attain
demographic
information
and
incorporate
it
to
varying
degrees
when
serving
up
recommendations.
But
the
more
personality-‐based
information
these
services
can
capture
without
annoying
users,
the
more
they
can
measure
correlations
and
target
recommendations.
Whether
this
implementation
involves
data
mining
from
public
social
media
profiles
or
building
extended
social
profiles
within
a
music
application,
it
has
potential
to
dramatically
improve
music
recommendation.
The
key
is
to
convince
users
they
are
benefitting
each
time
they
build
out
their
profile
and
use
A/B
testing
to
ensure
that
recommendations
improve.
Musical
Experience
and
Genre
Preferences
I
expected
the
survey
questions
addressing
musical
experience
to
correlate
strongly
with
genre
preferences.
I
based
this
hypothesis
on
two
observations
of
my
own
experience
with
music.
First,
the
more
time
I
spent
listening
to
music,
the
more
I
got
bored
listening
to
the
same
few
genres
and
tended
to
explore
unfamiliar
genres.
Second,
playing
guitar
has
26
28. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
had
a
tremendous
impact
on
my
music
preferences
and
listening
habits,
and
I
expected
this
trend
to
hold
true
for
other
people
regardless
which
instrument
they
play.
The
plethora
of
studies
on
the
effect
of
music
education
on
preferences
also
motivated
me
to
include
these
measures
in
the
study
(LeBlanc).
To
my
knowledge,
the
music
recommendation
technologies
of
iTunes,
Pandora,
and
YouTube
don’t
take
users’
musical
experience
into
account
and
I
felt
this
represented
an
opportunity
for
improvement.
Although
genres
were
the
only
feasible
metric
of
music
preference
for
the
purposes
of
this
analysis,
future
studies
may
address
correlations
between
musical
experience
and
song
preferences.
Upon
examining
the
spreadsheet
of
bivariate
correlations
generated
in
SPSS,
I
found
four
variables
of
musical
experience
that
correlated
strongly
with
several
genre
preferences.
First,
listening
hours
per
week
correlated
positively
with
electronic
and
jazz
music
at
the
.05
and
.01
levels,
respectively.
Although
correlations
with
other
genres
weren’t
statistically
significant,
all
were
positive
except
country
music.
This
proves
that
listening
to
music
more
often
facilitates
a
wider
range
of
preferences
and
correlates
strongest
with
electronic
and
jazz.
Next,
I
examined
how
musical
education
in
both
schools
and
private
lessons
correlated
with
genre
preferences.
I
expected
the
two
metrics
to
have
similar
correlations
with
genre
preferences,
and
hypothesized
that
higher
levels
of
music
education
would
correlate
positively
with
preferences
for
niche
genres.
As
it
turned
out,
“musical
education
in
school”
correlated
positively
with
classical
at
the
.01
level
and
with
jazz
and
world
at
the
.05
level.
On
the
other
hand,
“years
of
private
music
lessons”
correlated
positively
with
preferences
for
classical,
world,
and
rock,
but
negatively
with
country.
While
none
of
the
27
29. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
other
genres
had
statistically
significant
correlations,
I
noticed
a
general
trend
of
negative
correlations
between
musical
education
and
preferences
for
popular
genres
like
dance,
pop,
hip
hop/rap,
and
country.
Additionally,
I
found
statistically
significant
positive
correlations
between
years
of
experience
playing
an
instrument
and
preferences
for
rock
and
classical,
with
six
more
genres
producing
positive
correlations
that
were
above
the
.05
level.
These
findings
generally
confirmed
my
hypotheses
and
show
that
music
recommendations
may
be
improved
by
accounting
for
users’
music
experience,
though
a
more
thorough
study
using
song
preference
is
necessary
to
substantiate
these
conclusions.
Listening
Environment
and
Genre
Preferences
The
final
category
of
variables
I
analyzed
in
conjunction
with
genre
preferences
was
respondents’
listening
environment.
I
examined
respondents’
views
across
eight
distinct
listening
environments
according
to
the
following
coded
indicators
of
how
often
they
listened
to
music
in
each:
“Never”
=
1,
“Rarely”
=
2,
“Sometimes”
=
3,
“Often”
=
4,
and
“Always”
=5.
The
average
responses
and
their
standard
deviations
are
represented
in
Figure
6.
28
30. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Frequency
of
Listening
in
Environments
and
Activities
6
Never
Rarely
Sometimes
Often
Always
5
4.73
4.36
4
4.07
+1
σ
3.83
3.63
3.27
Mean
3
3.07
-‐1
σ
2
1.89
1
0
At
Home
In
the
Car
At
Work
By
With
At
a
Party
Studying
Sleeping
Yourself
Friends
Figure
6
Again
I
explored
the
spreadsheet
from
SPSS
that
highlighted
significant
correlations
between
these
eight
metrics
and
genre
preferences.
In
much
the
same
way
that
correlations
between
genres
divided
the
genres
into
two
distinct
groups
(Tables
2
and
3),
the
variables
for
listening
environment
separated
into
three
separate
groups
(Table
4).
The
first
group
of
listening
environments
included
“In
the
Car”,
“Studying”,
and
“Sleeping”,
and
more
frequent
listening
in
these
environments
was
correlated
with
higher
ratings
in
several
genres,
with
no
negative
correlations.
The
next
group
consisted
of
“At
Home”,
“By
Yourself”,
and
“At
Work”,
and
had
one
or
fewer
correlations
with
genre
preferences.
The
final
group
of
environments
was
more
social
than
the
other
two,
and
had
an
equal
or
greater
number
of
negative
correlations
than
positive
correlations.
29
31. The
Longer
Tails
of
iTunes,
Pandora
and
YouTube
Penrose
Table
4
Significant
Correlations
With
Genres
Listening
Environment
Positive
Negative
In
the
Car
Rock,
Hip
Hop/Rap,
R&B/Soul,
Country
0
1
Studying
Jazz,
Latin,
Classical,
Reggae,
Rock
0
Sleeping
R&B/Soul,
Latin,
Classical
0
At
Home
Hip
Hop/Rap
0
2
By
Yourself
Pop
0
At
Work
0
0
With
Friends
Hip
Hop/Rap
Classical,
World
3
At
a
Party
Dance,
Pop,
Hip
Hop/Rap
Classical,
World,
Vocal
At
first
glance,
the
first
and
third
groups
of
Table
4
might
appear
to
be
a
DJ
guide
indicating
which
genres
should
and
shouldn’t
be
played
in
each
environment.
However,
these
are
only
correlations
between
frequency
of
listening
in
eight
environments
and
ratings
for
genres;
respondents
were
not
asked
directly
which
genres
they
listen
to
in
each
environment.
But
since
they
follow
such
a
logical
pattern,
it’s
clear
that
listening
environment
plays
a
pivotal
role
in
determining
which
genres
users
listen
to.
At
the
very
least,
these
correlations
provide
evidence
that
music
services
using
recommendation
technology
should
experiment
with
allowing
users
to
adjust
for
different
environments,
especially
in
lean-‐back
music
experiences
like
Pandora.
Choice
of
Music
Technology
and
Determining
Factors
Although
forcing
respondents
to
choose
one
favorite
service
may
have
made
for
simpler
analysis,
I
assumed
most
people
use
more
than
one
of
the
three
music
services
in
question.
So,
I
asked
respondents
to
rank
the
three
of
them
in
order
of
preference
and
30