Beguiled
by
Bananas:
A
retrospective
study
of
the
usage
&
breadth
of
patron
vs.
librarian
acquired
ebook
collections
Jason
Price,
Head
of
Collections
&
Acquisitions
John
McDonald,
Director,
Information
&
Bibliographic
Management
and
Faculty
Relations
Claremont
Colleges
Library,
Claremont
University
Consortium
Charleston
Conference
2009
Abstract
Library
acquisitions
lore
contains
a
cautionary
tale
of
a
patron
in
a
demand-‐ driven
environment
who
spent
a
huge
chunk
of
the
library
budget
on
ebooks
about
bananas.
This
story
and
others
like
it
have
been
used
to
perpetuate
the
argument
that
demand-‐ driven
acquisition
will
result
in
collections
that
don’t
appeal
to
a
broad
audience
or
are
otherwise
unbalanced.
We
apply
post-‐ acquisition
usage
data
from
multiple
libraries
to
test
the
hypothesis
that
patron-‐ acquired
versus
librarian-‐ acquired
collections
have
different
usage
profiles.
In
addition,
we
analyze
their
subject
profiles
to
evaluate
collection
breadth
and
balance.
Our
results
will
help
libraries
to
anticipate
the
effect
of
adding
a
demand-‐ driven
component
to
their
ebook
acquisition
strategy.
2
Bananas
tipped
the
boat:
a
cautionary
tale
The
genesis
of
this
study
was
the
sizeable
impact
of
a
story
that
was
told
during
a
debate
over
patron-‐
initiated
selection
at
the
Charleston
conference
in
20081.
It
goes
more
or
less
as
follows:
One
of
the
first
ebook
vendors
to
market
ebooks
to
libraries
set
up
an
experimental
patron-‐
driven
purchasing
system
for
a
large
academic
library
in
Colorado.
Soon
after
its
inception,
a
business
professor
at
the
university
assigned
a
paper
related
to
the
economics
of
the
banana
industry.
One
or
more
industrious
students
found
the
ebook
platform
and
clicked
through
to
every
ebook
in
the
collection
matching
a
search
for
the
keyword
‘ banana’.
Thanks
to
these
students,
and
an
early
patron-‐ driven
model
that
lead
to
ownership
after
two
full
text
click
to-‐ s,
the
library
found
itself
to
be
the
no-‐ so-‐ proud
owner
of
every
banana
related
ebook
in
the
vendor’s
collection.
Almost
certainly,
only
a
select
few
of
these
were
relevant
to
the
assignment,
so
the
library
‘ returned’
most
of
these
books
for
a
refund.
The
ultimate
impact
of
this
event
on
the
Colorado
library
was
small— yet
the
story
lives
on
in
library
circles
as
an
oft-‐ repeated
cautionary
tale.
Librarians
and
ebook
vendors
alike
have
used
it
as
‘ evidence’
that
patron-‐ driven
selection
is
a
bad
idea.
The
influence
of
this
anecdote
should
not
be
underestimated— eight
years
later
it
has
entered
the
realm
of
library
lore,
and
is
still
circulating
today.
Sales
staff
of
one
major
ebook
vendor
used
it
to
justify
their
company’s
decision
( recently
reversed)
not
to
offer
a
patron-‐ driven
pricing
model.
In
addition,
the
banana
legend
clearly
won
converts
in
last
year’s
patron-‐ initiated
purchasing
lively
lunch
debate1.
There
was
a
major
shortcoming
in
the
arguments
expressed
on
both
sides,
however:
neither
had
data
to
support
their
conclusions.
Frustrated
that
an
anecdote
had
won
the
day,
I
resolved
to
rectify
this
deficiency,
posthaste.
As
such,
our
study
was
designed
to
address
the
conclusion
that
seems
to
flow
naturally
from
the
banana
book
example:
i. e.
that
patron-‐ driven
selection
inevitably
results
in
purchasing
of
ebooks
that
no
one
( or
no
one
else)
is
interested
in.
First
we
describe
the
EBL
demand-‐ driven
system,
which
was
built
to
ensure
that
purchases
were
based
on
more
than
browsing
or
casual
interest,
and
allowed
us
to
distinguish
specious
post-‐ purchase
use
from
more
meaningful
varieties.
Then,
we
use
post-‐ acquisition
ebook
usage
data
from
multiple
libraries’
EBL
collections
to
address
the
hypothesis
that
patron-‐ selected
collections
are
inferior
to
librarian-‐ selected
collections.
Finally,
we
discuss
key
questions
that
arose
during
our
presentation.
Terminology
&
Background
The
library
industry
has
used
various
terms
to
describe
use-‐ based
ebook
acquisition:
from
patron-‐
initiated
to
patron-‐ driven
to
demand-‐ driven
to
user-‐ driven.
Although
we
billed
our
talk
as
comparing
patron-‐ acquired
versus
librarian-‐ acquired
collections,
we
quickly
realized
that
the
nature
of
these
1
Polanka
et.
al.
Tossing
Traditional
Collection
Development
Practices
for
Patron
Initiated
Purchasing:
A
Debate.
Charleston
Conference
2008.
3
collections
was
not
quite
that
clear
cut.
Some
‘ patron-‐ acquired’
purchases
were
certainly
made
by
EBL
users
that
were
librarians,
and
some
‘ librarian-‐ acquired’
purchases
were
undoubtedly
made
on
behalf
of
faculty
patrons
who
selected
them
for
teaching
or
research.
The
most
accurate
terms
we
could
divine
to
describe
these
collections
were
user-‐ selected
and
pre-‐ selected.
These
terms
more
precisely
reflect
the
fact
that
the
user-‐ selected
(‘ patron-‐ acquired’)
books
were
purchased
AFTER
they
were
used
one
or
more
times
by
a
local
library
user,
whereas
the
pre-‐ selected
(‘ librarian-‐ acquired’)
books
were
purchased
BEFORE
they
were
used.
Henceforth
we
use
these
two
terms
to
refer
to
the
‘ patron-‐ acquired’
and
‘ librarian
acquired‘
books
or
collections.
EBL-‐ hosted
ebooks
were
used
for
this
study
because
they
have
the
most
sophisticated
demand-‐ driven
system
currently
available,
both
in
terms
of
purchase
triggers
and
usage
reporting.
Key
features
of
this
system
include:
1. There
is
a
browse
period
for
every
use
of
every
book
a. This
allows
browsing/ evaluative
use
to
be
separated
from
meaningful
use
–
when
a
user
clicks
into
a
page
of
an
ebook,
s/ he
may
decide
it
is
not
useful— such
usage
is
reported
separately,
does
NOT
count
toward
purchase,
and
was
ignored
for
this
study
( whether
it
occurred
before
or
after
purchase)
b. The
browse
period
results
in
a
recorded
use
when
content
in
the
book
is
copied,
printed,
downloaded,
or
when
a
user
clicks
through
to
keep
the
page
open
for
more
than
5-‐ 10
minutes
c. As
a
result,
every
use
we
included
in
the
analysis
indicated
meaningful,
post-‐ purchase
use
( Demand-‐ driven
models
that
lack
this
feature
can
be
likened
to
a
physical
bookstore
suggesting
that
if
you
touch
a
book,
you’ve
bought
it!)
2. Usage-‐ Type
categories
for
every
transaction
made
it
easy
to
eliminate
pre-‐ purchase
use— ALL
of
the
usage
analyzed
for
this
study
occurred
after
the
book
had
been
purchased.
3. Unique
but
persistent
user
IDs
for
every
transaction
available
from
a
proxy-‐ based
user
authentication
system
made
it
possible
to
separate
repeat
use
by
the
same
person
( even
over
a
period
of
months
or
years)
from
use
by
a
variety
of
people,
enabling
assessment
of
the
breadth
of
the
audience
for
each
book
within
each
institution,
as
well
as
its
depth
in
terms
of
total
number
of
meaningful
uses
4. Detailed
electronic
invoice
data
available
for
every
purchase
transaction
distinguish
between
user-‐ selected
and
pre-‐ selected
purchases
and
provide
an
exact
date
of
purchase,
so
that
usage
could
be
analyzed
in
terms
of
uses
per
time
to
account
for
the
fact
that
some
books
were
owned
much
longer
than
others
an
thus
had
much
greater
opportunity
for
post-‐ purchase
usage.
Specifically,
for
those
familiar
with
EBL
terminology,
we
treated
instances
of
‘ read
online’
and
‘ download’
usage
as
equivalent,
and
ignored
all
‘ browse’
usage.
This
ability
to
distinguish
and
ignore
casual
usage
distinguishes
the
EBL
system
and
this
study
from
all
other
e-‐ resource
usage
evaluation
that
we
know
of,
and
may
be
a
major
explanatory
factor
of
its
success
as
a
demand-‐ driven
system.
We
used
data
from
the
EBL
system
to
address
each
of
the
following
questions
in
turn:
4
1) Are
user-‐ selected
ebooks
used
less
often
than
pre-‐ selected
ebooks?
2) Do
user-‐ selected
ebooks
have
a
narrower
audience
than
pre-‐ selected
ebooks?
3) Are
user-‐ selected
ebooks
less
likely
to
be
used
than
pre-‐ selected
ebooks?
4) Are
user-‐ selected
collections
less
comprehensive
( or
more
skewed)
than
pre-‐ selected
collections?
Scope
and
Design
The
Claremont
Colleges
Library
has
not
purchased
any
books
on
the
EBL
platform— user-‐ selected
or
otherwise:
the
data
used
in
this
study
were
from
other
libraries.
EBL
was
willing
to
share
data
from
eleven
libraries
on
the
condition
that
their
specific
identities
remain
unknown
to
us2.
These
datasets
included
28,327
books
purchased
over
a
four
year
period,
2006-‐
2009
( Table
1).
In
the
aggregate,
portions
of
these
owned
books
were
‘ used’
( downloaded,
printed,
copied,
or
browsed
for
more
than
5
minutes)
nearly
a
quarter
of
a
million
times
( 213,887)
during
this
period.
Purchasing
type
varied
greatly
among
the
eleven
libraries.
We
divided
them
into
three
categories
( user-‐ selected,
pre-‐ selected
or
mixed)
based
on
the
total
number
of
books
purchased
each
way
by
each
library
( Table
1).
We
selected
only
the
mixed
type
libraries
for
this
initial
study
to
allow
us
take
differences
in
user
communities
into
account.
This
approach
allowed
us
to
compare
user-‐
vs.
pre-‐
selected
collections
within
each
library
as
well
as
overall.
( Initial
exploratory
analysis
showed
that
overall
results
including
data
from
all
eleven
libraries
were
consistent
those
from
our
subsample;
data
not
shown).
Thus
the
results
we
present
are
from
five
libraries:
A,
B,
E,
I
and
K-‐-‐ including
15,673
ebooks
that
were
used
151,491
times
( Table
1,
in
bold).
Within
this
sample,
the
libraries
bought
and
average
of
twice
as
many
user-‐ selected
books,
but
this
does
not
affect
the
results
because
comparisons
were
made
on
an
average-‐ per-‐ book
basis
and
all
10
selection
sets
had
a
large
enough
sample
( n
≥ 147)
to
avoid
sampling
bias.
EBL's
model
allows
for
a
great
deal
of
customization
of
demand-‐ driven
( user-‐ selected)
purchasing.
Libraries
can
choose
the
number
of
loans
before
purchase,
the
range
of
books
available
for
purchase
( by
date,
publisher,
subject,
price
range,
etc.),
and
can
even
choose
to
approve
some
or
all
user-‐ selected
2
We
are
greatly
indebted
to
Alison
Morin,
Kari
Paulson
and
Sally
TerBeck
of
EBL,
who
provided
the
datasets
that
met
our
criteria
and
engaged
in
a
number
of
detailed
discussions
to
help
us
understand
and
take
into
account
all
of
the
implications
and
nuances
that
affect
the
interpretation
of
these
data
Table
1:
EBL
data
available,
11
Libraries
Library
Purchasing
Type
User-‐
selected
Pre-‐
selected
Usage
-‐
download
Usage
-‐
read
online
A
MIX
1131
552
6773
9888
B
MIX
5246
2612
42880
38329
C
USER
2198
102
0
11801
D
USER
3010
48
697
15126
E
MIX
4159
909
17396
25604
F
PRE
0
1451
4905
3082
G
PRE
31
2154
7001
4459
H
USER
801
0
556
415
I
MIX
305
336
3334
2568
J
USER
2799
53
5
13349
K
MIX
147
276
2436
2283
TOTAL
19,831
8,496
85,983
126,904
5
purchases
before
they
are
made.
All
five
libraries’
EBL
platforms
were
customized
in
one
or
more
of
these
ways
to
varying
extents
at
different
points
throughout
the
study
period.
In
spite
of
this
variation,
our
results
show
that
differences
proved
to
be
extraordinarily
consistent
across
these
five
libraries.
They
are
consistent
despite
the
underlying
variation
in
the
type,
degree,
and
duration
of
user-‐ selected
purchase
model
customization.
We
feel
this
adds
to
the
weight
to
the
evidence
we
present,
rather
than
detracting
from
it.
Since
these
sets
contain
books
that
were
bought
continuously
over
the
4-‐ year
period,
books
within
each
set
had
greatly
varying
opportunity
for
post-‐ purchase
usage
( Table
2).
Exploratory
analysis
showed
that
purchase
date
had
a
statistically
significant
effect
on
post-‐ purchase
usage
and
number
of
unique
users
( data
not
shown).
For
simplicity
and
ease
of
interpretation,
we
designed
our
usage
metrics
to
take
length
of
ownership
of
each
book
into
account,
rather
than
including
it
as
an
additional
causal
( independent)
variable.
Total
usage
was
measured
as
uses
per
year
owned
and
breadth
was
measured
in
unique
users
per
year
owned.
Because
we
knew
the
exact
date
that
each
book
was
bought,
we
could
calculate
usage
and
users
per
day
X
365
with
no
error
in
the
denominator,
avoiding
the
hazard
of
introducing
an
extra
source
of
error
due
to
a
ratio-‐ based
dependent
variable.
In
order
to
minimize
the
chance
of
over-‐ estimating
the
effect
of
usage
of
books
owned
for
a
short
time,
we
explored
exclusion
of
books
owned
for
less
than
six
months
and
less
than
one
year.
For
instance,
if
a
book
owned
for
just
1
day
was
used
5
times
that
day,
multiplying
those
uses
by
365
days
would
drastically
inflate
its
estimated
usage
per
year).
Neither
the
direction
of
the
differences
nor
the
significance
of
the
model
changed
with
either
of
these
subsamples,
but
the
differences
observed
were
less
extreme
when
books
owned
six
months
or
less
( n= 2350)
were
removed
( data
not
shown).
No
additional
material
change
was
observed
when
books
owned
6
to
12
months
were
also
excluded.
As
such,
in
order
to
err
on
the
conservative
side,
we
excluded
all
books
owned
less
than
183
days
( 6
months)
from
the
analysis.
Analysis
As
outlined
earlier,
we
were
interested
in
understanding
if
user-‐ selected
collections
would
have
the
same
depth,
breadth,
initial
use
and
subject
distribution
as
pre-‐ selected
collections.
To
test
the
first
two
questions,
we
developed
a
model
that
included
three
predictor
variables:
the
purchase
type,
the
library,
and
an
interaction
effect
between
purchase
type
and
library.
Our
analysis
of
variance
( ANOVA)
tests
addressed
the
significance
of
these
predictors
on
two
response
variables:
usage
per
year
and
unique
users
per
year
of
each
ebook.
These
variables
address
the
extent
and
breadth
of
usage
of
each
collection.
The
latter
two
questions
were
addressed
more
informally,
based
on
observed
differences
in
the
patterns
of
non-‐ use
and
LC
class
distribution.
Table
2:
Average
Users,
Usage,
and
Availability
period
Library
Ebooks
Average
Unique
Users
Average
Usage
Average
Days
Available
A
1683
5.63
9.90
475.28
B
7859
5.69
10.09
512.33
E
5068
5.00
8.49
440.30
I
641
4.56
9.21
766.38
K
423
5.28
11.13
784.81
6
Results
Total
post-‐ purchase
usage
Descriptive
statistics
for
the
overall
dataset
( Table
3)
indicate
that,
on
average,
user-‐ selected
ebooks
were
used
twice
as
often
as
pre-‐ selected
ebooks
( 8.6
vs.
4.3
times
per
year).
Individual
libraries
showed
1.75
to
4.5
times
higher
use
of
user-‐ selected
ebooks.
Additionally,
user
selected
collections
had
larger
standard
deviations,
indicating
that
the
most
heavily
used
books
were
used
very
frequently,
compared
to
pre-‐ selected
books
that
had
a
smaller
range
of
average
usage.
Results
of
the
ANOVA
for
uses
per
book
per
year
( Table
4)
indicated
that
while
purchase
type
was
a
significant
predictor,
library
was
not.
In
addition,
the
interaction
effect
of
purchase
type
and
library
had
the
highest
observed
power,
suggesting
that
the
extent
of
the
difference
differed
between
libraries
( see
Fig.
1).
Further
analysis
of
the
dataset
indicated
that
average
values
of
usage
per
year
were
higher
for
user-‐ selected
ebooks
than
for
pre-‐ selected
ebooks
across
all
five
libraries
( Fig.
2).
Four
of
the
five
libraries
had
significantly
higher
means
at
the
95%
confidence
interval.
Based
on
these
results,
we
are
confident
that
user-‐ selected
ebooks
are
not
less
likely
to
be
used
than
pre-‐ selected
ebooks,
but
in
fact,
are
used
at
a
significantly
higher
rate.
Table
4:
ANOVA
Results,
Mixed
Purchase
Model
Libraries
Source
Type
III
Sum
of
Squares
Mean
Square
F
Sig.
Partial
Eta
Squared
Observed
Power a
Intercept
Hypothesis
191873
191872
295.056
.000
.979
1.000
Error
4207
650
Purchase
Hypothesis
35509
35508
26.796
.004
.843
.982
Type
Error
6615
1325
Library
Hypothesis
3984
996
.431
.782
.301
.088
Error
9243
2310
Purchase
*
Hypothesis
9243
2310
8.092
.000
.002
.998
Library
Error
3801351
285
Table
3:
Usage,
Mixed
Purchase
Type
Libraries
Purchase
type
Library
Mean
Std.
Dev
Books
A
10.12
13.95
778
B
8.39
19.31
4723
E
8.14
15.25
3111
I
14.36
42.49
277
K
7.09
13.16
147
User-‐
selected
ebooks
Total
8.61
18.74
9036
A
3.67
5.89
498
B
4.60
11.55
2475
E
4.61
18.55
766
I
3.21
8.69
331
K
3.08
5.47
217
Pre-‐
selected
ebooks
Total
4.31
12.25
4287
A
7.60
11.92
1276
B
7.09
17.14
7198
E
7.44
16.02
3877
I
8.29
29.88
608
K
4.70
9.56
364
Total
Total
7.23
17.04
13323
7
Breadth
of
post-‐ purchase
usage
While
user-‐ selected
ebooks
are
used
more
frequently
than
pre-‐ selected
ebooks,
one
could
reasonably
assert
that
this
is
due
to
the
fact
that
if
a
user
selects
an
ebook,
they
are
certainly
more
likely
to
use
it
a
second
time
and
continue
to
use
that
item.
Stated
in
another
way,
user-‐ selected
collections
could
end
up
being
built
in
a
narrower
sense,
with
each
book
appealing
only
to
the
user
who
selected
it.
This
observation
led
to
question
2:
Do
user-‐ selected
ebooks
have
a
narrower
audience
than
pre-‐ selected
ebooks?
On
the
contrary,
the
average
number
of
unique
users
per
year
was
1.75
to
3.3
times
higher
for
the
user-‐ selected
collections
( Fig.
3).
As
was
the
case
with
overall
usage,
the
average
for
the
user-‐ selected
collections
was
significantly
greater
than
for
the
pre-‐ selected
collection
at
four
of
the
five
libraries
at
the
95%
confidence
interval.
The
ANOVA
results
( not
shown)
were
similar
to
the
results
testing
for
an
effect
on
overall
usage
( presented
above).
Figure
2:
Average
usage
by
library
and
purchase
type
Figure
1:
Purchase
type
by
library
interaction
Figure
3:
Unique
users
per
ebook
per
year
8
Figure
4:
Percentage
of
titles
unused
by
library
and
purchase
type
Number
of
unused
titles
While
we
found
that
the
average
number
of
users
per
book
was
greater
for
user-‐ selected
ebooks
than
pre-‐ selected
ebooks,
the
possibility
remained
that
user-‐ selection
could
result
in
a
larger
number
of
titles
that
went
unused
after
purchase.
This
led
to
question
3:
Are
user-‐ selected
books
less
likely
to
be
used
than
pre-‐ selected
books?
We
looked
at
descriptive
statistics
to
answer
this
question.
The
percentage
of
unused
books
was
quite
low
overall:
90%
of
the
books
in
every
collection
had
been
used
at
least
once
since
they
were
purchased
( Fig.
4).
Four
of
the
five
libraries
appeared
to
have
fewer
unused
titles
in
their
user-‐ selected
collections.
Because
this
is
a
library-‐ level
statistic,
however,
our
sample
size
( 5)
was
not
large
enough
to
test
for
significance.
Subject
coverage
While
user-‐ selected
collections
are
used
more
frequently
than
pre-‐ selected
collections
and
used
by
more
users
( apparently
with
fewer
unused
titles),
an
argument
can
be
made
that
there
is
an
additional
value
to
the
overall
library’s
collection
from
pre-‐ selected
collections
in
the
form
of
a
more
balanced
collection
across
all
subject
areas
and
research
interests
for
the
community
as
a
whole.
Our
final
question
addressed
subject
balance:
“ Are
user-‐ selected
collections
less
comprehensive
( or
more
skewed)
than
pre-‐ selected
collections?”
1.7%
10.0%
3.5%
9.7%
4.2%
5.9%
2.5%
2.0%
0.3%
6.3%
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
12.0%
USER
PRE
USER
PRE
USER
PRE
USER
PRE
USER
PRE
A
B
E
I
K
%
of
books
unused
Library
and
Purchase
( Selec ` on)
Type
9
Figure
5:
Discipline
distribution
by
library
and
purchase
type
Discipline
level
subject
distribution
appeared
to
be
remarkably
consistent
across
purchase
types
( Fig.
5).
The
only
distribution
that
is
markedly
different
is
for
Library
“ K”,
where
users
selected
twice
as
many
Arts
&
Humanities
and
Social
Sciences
books,
while
the
pre-‐ selected
ebooks
fell
more
heavily
into
Science
&
Technology.
This
may
have
been
by
design,
or
because
this
library
expected
its
EBL
users
to
be
more
science-‐ centric
than
they
proved
to
be.
Addressing
the
same
question
at
a
more
specific
level,
we
categorized
the
collections
by
LC
class
( Fig.
6).
The
relatively
similar
distributions
of
the
most
common
classes
of
the
Library
A,
B
and
E
collections
suggest
that
the
subject
distribution
of
user-‐
selected
and
pre-‐ selected
collections
is
similar
in
these
libraries.
Libraries
“ I”
and
“ K”
had
somewhat
dissimilar
user-‐
vs.
pre-‐ selected
collection
patterns.
Library
“ I”
had
an
almost
inverse
collecting
relationship
for
some
subjects,
and
library
K
pre-‐
selected
a
much
higher
proportion
of
‘ Q’
books
as
noted
above.
This
analysis
provided
good
evidence
that
user-‐ selected
collections
are
no
more
narrow,
skewed,
or
individually
focused
than
those
chosen
by
a
pre-‐ selection.
And
in
fact,
for
most
institutions
in
the
study,
the
collecting
pattern
of
users
mirrored
those
of
pre-‐ selection.
Figure
6:
Collection
distribution
by
LC
class,
library
and
purchase
type
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
A
B
E
I
K
Propor ` on
of
collec ` on
Library
H
R
Q
T
L
H
R
Q
T
L
1
0
Discussion
These
results
clearly
and
repeatedly
demonstrate
that
EBL’s
demand-‐ driven
acquisition
model
builds
collections
of
ebooks
that
are
used
more
often
and
have
a
wider
audience
than
their
pre-‐ selected
counterparts.
As
such,
ownership
of
obscure,
unwanted
books
is
NOT
an
inevitable
outcome
of
use-‐
based
ebook
selection,
as
the
patron-‐ driven
banana
book
legend
seems
to
imply.
Furthermore,
we
present
preliminary
analyses
that
suggest
that
gaining
this
use-‐ based
advantage
does
not
require
libraries
to
sacrifice
subject
breadth.
The
usage
and
breadth
advantage
conferred
by
user-‐ selected
acquisition
appears
to
exist
in
every
library
we
tested.
Library
K
was
the
only
one
that
did
not
show
a
statistically
significant
effect
of
purchase
type,
probably
because
it
had
the
smallest
sample
sizes.
The
broad
agreement
in
this
advantage
across
libraries
suggests
that
it
does
not
depend
on
the
degree
of
customization
of
the
demand-‐ driven
model.
Libraries
user-‐ selected
collections
had
greater
use
by
a
broader
range
of
users
regardless
of
whether
they
mediated
the
purchases,
or
restricted
the
catalog
of
books
available
for
purchase,
or
had
a
higher
threshold
of
initial
use
before
purchase,
or
bought
their
pre-‐ selected
books
up
front
or
during
the
entire
duration
of
the
study.
This
consistency
should
assure
libraries
that
are
concerned
about
the
initial
set
up
of
their
plan
that
it
will
have
a
minimal
affect
on
the
demand
driven
usage
advantage.
For
some,
higher
post-‐ purchase
usage
is
sufficient
reason
to
implement
a
demand-‐ driven
model.
For
others,
questions
of
collection
quality
remain.
Does
greater
popularity
necessarily
mean
that
the
collection
is
better?
Will
it
result
in
purchases
the
library
would
not
otherwise
have
made?
Future
work
could
use
independent
book
profile
definitions
to
compare
collection
quality
and
address
these
questions.
YBP,
for
instance,
profiles
books
as
Basic
Essential/ Recommended,
Research
Essential/ Recommended,
Specialized,
Supplementary,
or
Unlisted.
Comparison
of
the
proportion
of
books
in
each
category
for
both
purchase
types
would
be
enlightening.
Publisher
and
cost
breakdowns
would
also
be
of
interest.
Another
important
question
is
whether
the
higher
usage
and
larger
audience
we
show
for
EBL-‐ based
demand-‐ driven
collections
would
also
occur
with
other
demand-‐ driven
platforms.
More
specifically,
is
the
ability
to
ignore
casual
click-‐ to
use
critical
to
the
success
of
user-‐ selected
collections?
This
question
could
be
addressed
within
the
EBL
data
by
comparing
the
real
life
demand-‐ driven
collections
with
those
that
would
have
resulted
had
the
browse
usage
been
included.
Alternatively,
a
comparative
analysis
between
patron-‐ driven
collections
on
multiple
platforms
( i. e.
EBL
vs.
MyILibrary
or
NetLibrary)
could
also
provide
insight.
Either
of
these
approaches
would
have
significant
methodological
challenges,
but
it
seems
crucial
to
take
the
next
step
to
determine
whether
a
more
sophisticated
platform
is
essential
for
success
in
demand-‐ driven
ebook
acquisition.
Library Research, Publications and Events at The Claremont Colleges
