Applied Visual Analytics for Economic Decision-Making
Anya Savikhin * Ross Maciejewski +
David S. Ebert
* Purdue University Department of Economics
+ Purdue University Regional Visualization and Analytics Center (PURVAC)
A BSTRACT
This paper introduces the application ofvisual analytics techniques
as a novel approach for improving economic decision making. Par-
ticularly, we focus on two known problems where subjects’ be-
havior consistently deviates from the optimal, the Winner’s and
Loser’s Curse. According to economists, subjects fail to recog-
nize the prof i t-maximizing decision strategy in both the Winner’s
and Loser’s curse because they are unable to properly consider all
the available information. As such, we have created a visual ana-
lytics tool to aid subjects in decision making under the Acquiring
a Company framework common in many economic experiments.
We demonstrate the added value of visual analytics in the decision
making process through a series ofuser studies comparing standard
visualization methods with interactive visual analytics techniques.
Our work presents not only a basis for development and evaluation
of economic visual analytic research, but also empirical evidence
demonstrating the added value of applying visual analytics to gen-
eral decision making tasks.
1 T HE W INNER ’ S AND L OSER ’ S C URSE
In order to best demonstrate the applicability of visual analytics to
decision making problems, we have chosen to analyze a familiar
situation, bidding at an auction. When a person bids at an auction,
economists assume that the item under bid is of equal value to all
participants. This type ofauction is known as aCommon Value Auc-
tion. Here, each participant has some estimate of the item’s worth
with a degree ofuncertainty, and each places a bid accordingly. The
winner of the auction is the individual who bids the highest. Un-
fortunately, ifwe assume that the average bid ofall the participants
represents the actual worth of the item, then the winner of the auc-
tion has now clearly overpaid. This phenomenon is known as the
Winner’s Curse [2, 10, 14]. As the number of bidders increases, so
does the severity of the amount overpaid and the more likely it is
that some of them have overestimated the auctioned item’s value.
In technical terms, the winner’s expected estimate is the value of
the f i rst order statistic, which increases as the number of bidders
increases. Since most auctions involve a degree of common value
and uncertainty, this deviation from the optimal bid is an important
phenomenon to study. In this case, companies must assess the value
ofthe market in that area and make a bid accordingly.
The sister problem to this is known as the Loser’s Curse [7],
which appears when the parameters of the experiment are changed
in such a way that the naive bid is now below the prof i t-maximizing
bid, and the prof i t-maximizing bid results in winning the company.
This is because the lower bound for the range of possible company
values is increased, reducing the bid range in which high nega-
tive prof i ts would result (as in the Winner’s Curse setup). Subjects
* e-mail: anya@purdue.edu
+ e-mail:rmacieje@purdue.edu
e-mail:ebertd@ecn.purdue.edu
who do not win the company have lower prof i ts than those who bid
higher and win the company, but subjects consistently underbid in
this setup and do not f i nd the prof i t-maximizing solution.
In order to better understand subject motivations, the Winner’s
and Loser’s Curse have also been studied in a simplif i ed framework
which is referred to as the Acquiring a Company problem, f i rst for-
mulated by Samuelson and Bazerman [14]. The Acquiring a Com-
pany problem is simplif i ed because subjects no longer interact with
or bid against one another; instead, the subject bids on a company,
the value ofwhich is randomly determined by the computer but un-
known to the subject. The subject is also told that the company is
worth more to him than it is to the seller (in this case, the computer
is the seller). Even though subjects no longer compete against one
another to win the company, both the Winner’s and Loser’s Curse
are still present.
In order to compare our results with previous work, we have
chosen to employ the Acquiring a Company framework in our vi-
sual analytics application and evaluation work. The motivation for
this work is to help subjects overcome this failure by helping them
better consider the relevant information, thereby improving their
decision making abilities. We present an extensible visual analyt-
ics framework for economic decision making, and demonstrate the
added value of such techniques through the results of a user study.
Our work shows the benef i t of visual analytics in economic deci-
sion making, improving subject performance in the Acquiring a
Company framework, and aiding in overcoming the Winner’s and
Loser’s Curse phenomena.
2 A PPLYING V ISUAL A NALYTICS TO E CONOMIC D ECISION -
M AKING
We present this study at a crucial time when many researchers are
calling for new techniques and systems to help non-expert users of
varying abilities in complex decision-making and analysis tasks [9].
According to the NIH/NSF Visualization Research Challenges re-
port, visualization is essential to the solution of complex problems
in every f i eld and systems must allow ordinary people to experi-
ment with “what if” situations [9]. Our research addresses these
points through the application ofa visual analytics tool that can aid
in complex decision making tasks in real world applications with
uncertainty.
2.1 Related Literature in Economics
Many studies have attempted to address the Winner’s and Loser’s
Curse problems through the Acquiring a Company framework over
the past 20 years. The goal of these studies was to f i nd a way for
subjects to avoid the Winner’s or Loser’s Curse by improving their
decision-making abilities. Some of these studies allowed for role
reversal or modif i ed the game, while others attempted to make the
game “easier” for subjects, or to prepare subjects better. In one
study, a training package was designed to help subjects, with train-
ing on various conditional probability problems, but this was only
marginally effective in improving learning [8]. In another study,
training was employed where subjects were given a lesson in prob-
ability and then performed four similar tasks where the Acquiring
107
IEEE Symposium on Visual Analytics Science and Technology
October 21 - 23, Columbus, Ohio, USA
978-1-4244-2935-6/08/$25.00 (c)2008 IEEE