Information Technology Reference
In-Depth Information
Table 11.10
Variety and
level of an attribute
Attribute
Level 1
Level 2
Level 3
Store
Yes
No
-
Saturation
High
Low
-
Procedure
Easy
Difficult
-
Postage
Free
500 yen
-
Price
3,000 yen
4,000 yen
5,000 yen
e
mV
ni
j2C
P
ni
¼
(11.3)
e
mV
nj
However, CL is the model achieved under two assumptions,
homogeneous
preference and Independence from Irrelevant Alternatives
:
IIA
. Although CL is
easy to analyze, it has the problem of weak model interpretability. Revelt and Train
(
1998
) advocated the mixed logit model ML, which eliminates these two
assumptions. ML is a model that has a preference from which an individual differs.
When a certain respondent
n
chooses item
i
, utility is set to
U
ni
, denoted as follows.
X
M
m¼
1
b
m
n
x
ni
þ e
ni
U
ni
¼
V
ni
þ e
ni
¼
(11.4)
e
ni
has an independent and identical type I extreme value
distribution, and the probability that the respondent
n
will choose
i
is formulized as
follows.
It is assumed that
ð
Y
T
V
i
Þ
P
j
¼
1
exp
exp
ð
P
ni
¼
f
ðbj
O
Þ
d
b
(11.5)
ð
V
j
Þ
t
1
T indicates the number of occurrences of the choice experiment, and several
repetitive questions are presented to the same respondent in the usual choice
experiment.
f
is the probability density function of
b
, and
O
indicates parameters
such as the average and variance of
.
In this study, the choice experiment was conducted using an orthogonal array
design; the variety of attributes and the level were set up as shown in Table
11.10
.
Eight profiles were created using the orthogonal array design from the level of
each attribute. Two profiles were combined at random, and a choice set with the
added option “Using neither online shop” was created. Each respondent answered
eight choice sets per questionnaire. An example of a choice set is shown in
Table
11.11
.
An alternative specific constant (ASC) was added to the analysis; ASC3 was
introduced into “Using neither online shop.” It can be interpreted as negative for the
purchase of goods from an online shop if the ASC is significantly estimated as
b