Databases Reference
In-Depth Information
One can theoretically model consumer search as a process where the searcher is
in a state of deciding to seek out additional information (i.e., searching again) or not
seek additional information (i.e., stop searching). This decision process is a function
of the expected benefit of any additional information [
59
], with the benefit of search-
ing being a reduction in uncertainty.
As the searcher gathers additional information from additional searches, the
searcher's expected benefit of seeking new information decreases (i.e., their uncer-
tainty is reduced), and this increased confidence results in a lower probability of the
searcher soliciting new information (i.e., doing an additional search).
Now, one can model this consumer search process by developing a model of the
search process as the probability that individual
i
searches an
x
th time (i.e., submits
an addition query
x
th) as a decrement of the probability of visiting the (
x
i
-1
)st site
[based on the work of 17, 36]:
Xx
x
==
−
(
1
)
θ
i
i
Pr[
]
Pr[
Xx x
=− =
1
],
2 3
,
,
...
,
i
i
i
i
i
x
i
Equation 3.1. Probability model of consumer search.
The model presented in
Equation 3.1
is simply a rephrasing of the search behav-
ior described earlier, except it is now presented in mathematical symbols rather than
words.
The model is recursive. Any series of states in the searching process is just com-
posed of a set of individual search states. Therefore, we can present this recursive
model as a logarithmic distribution.
The revised model is presented in
Equation 3.2
, where
a
i
= -[ln(1-θi)]-1
i
)]
-1
and
0<θ
i
<1
.
x
Xx
a
θ
i
ii
Pr[
==
]
,
x
=
12...
,
,
,
i
i
i
x
i
Equation 3.2. Logarithmic probability model on consumer search.
To illustrate the model of consumer search (and how this model is valuable at the
aggregate level),
Figure 3.7
plots the shape of the consumer search model for a variety
of probabilities. To explain
Figure 3.7
, with a
θ
of 0.2, there is a 90 percent probabil-
ity that the searcher will click on only one result. With a
θ
of 0.5, there is a 71 per-
cent probability that the searcher will click on only one result. At three site visits, we
address only 10 percent of searchers with a
θ
of 0.08. We could do a similar graph (and
model) for query length, session length, and landing pages visited. They all would plot
similarly.
We could also plot the logarithmic distribution of our model. In these cases, the
lines would be straight and of different slopes, but we could derive the same percent-
ages and site numbers. More on this later.
What does this mean?
