Decision Scienceshow the game

is played

arly in its existence, NSF started to

support research on game theory—the

study of individuals’ rational behavior in

situations where their actions affect

other individuals. Although the research

had little practical value at the time,

NSF continued to support it during the

decades to follow, with substantial

returns on the investment. Game

theory and related areas of decision

science supported by NSF have helped

to solve practical problems once

thought too complicated to analyze.

E

Game t h eo r y deals with the interactions of small numbers of individuals, such as

buyers and sellers. For almost thirty years after its development during World War II, the theory

remained an academic exercise, its dense mathematical proofs defying practical applications.

Yet NSF stood by leading economists who painstakingly demonstrated how to use game theory

to identify winning strategies in virtually any competitive situation. NSF also supported experi-

mental economists who tested theoretical approaches under controlled laboratory conditions,

and psychologists whose studies of individual decision making extended understanding of how

economically rational individuals behave.

Persistence paid off. In 1994, John F. Nash, Reinhard Selten, and John C. Harsanyi, who first

received NSF support in the 1960s, won the Nobel Prize in Economics for “their pioneering

analysis of equilibria in non-cooperative games.” The following year, game theory gave the Federal

Communications Commission the logical structure for innovative auctions of the airwaves for

new telecommunications services. The auctions raised over $7 billion for the U.S. Treasury, and

marked a coming of age for this important analytic branch of economics.

In supporting this field, NSF’s goal was to build the power of economics to elucidate and

predict events in the real world. The support not only advanced the discipline, but also benefited

all individuals in many aspects of daily life.

Decision Sciences — 79

Decisions, DecisionsWe all find ourselves in situations that call forstrategic thinking. Business executives plan strate-gies to gain market share, to respond to theircompetitors’ actions, to handle relations withemployees, and to make career moves. Managersin government think strategically about the likelyeffects of regulations at home and of diplomaticinitiatives abroad. Generals at war develop strate-gies to deploy troops and weaponry to defeat theenemy while minimizing their own losses. At a moreindividual level, buyers and sellers at flea marketsapply strategies to their bargaining. And parentsuse strategy on their children, who—of course—behave strategically with their parents.

What is strategy? Essentially, it is anticipatingthe actions of another individual and acting inways that advance one’s self-interest. Since theother person also behaves strategically, strategyincludes making assumptions about what thatindividual believes your strategy to be. We usuallyassociate strategy with adversarial situations suchas war, but that is much too narrow. In love, weuse strategy, often unthinkingly, to win our lovedone’s heart without sacrificing our self-esteem orour bank account. Some strategists have objec-tives, such as racial harmony, that they feel serveeveryone’s interest. Probably the most commonuse of strategy occurs in basic economic trans-actions, such as buying and selling. However itis applied, the ultimate point of strategy is toachieve objectives.

That is precisely what players of games try todo. Similarities between games and strategicbehavior in the economy formed the frameworkfor Theory of Games and Economic Behavior, abook published in 1944 by mathematicians Johnvon Neumann and Oskar Morgenstern. Theirlandmark work begins where classical economicsleaves off.

The starting point for traditional economics isthe equilibrium price, the point at which a seller’sasking price equals the buyer’s bid price. Classicaleconomic theory goes on to analyze the price in terms of outside influences. Von Neumannand Morgenstern, however, went in anotherdirection: They looked at the relationship betweenthe participants.

Exactly how, they asked, do buyers and sellersget to the equilibrium price? In a world of perfectcompetition, containing so many buyers and sell-ers that any one individual’s acts are insignificant,marketplace dynamics suffice. But what abouteconomic transactions that involve only a fewbuyers and sellers? What happens when, forexample, MCI offers potential customers a dealon long-distance service, and AT&T responds byoffering the public its own new deal? The strate-gic moves in such economic decision makingstruck Morgenstern and von Neumann as mathe-matically indistinguishable from moves in chess,poker, and other games in which some strategiesconsistently win over others. Their book, a com-pendium of mathematical theorems embodyingmany different strategies for winning, was the firstrigorously scientific approach to decision making.

Significant as it was, game theory took a longtime to catch on. A small group of academicsrecognized its significance as a research tool.And some military applications appeared in the1950s, when the Rand Corporation used gametheory to anticipate responses of potential ene-mies to weaponry of various kinds. The world ofbusiness, however, regarded game theory as anarcane specialty with little practical potential.

80 — National Science Foundation

NSF Lends Support NSF took a different view, and began to supportgame theory mathematics in the 1950s. “Theappeal of game theory always was the beauty ofthe mathematics and the elegance of the theorems,”explains Daniel H. Newlon, senior program direc-tor for economics at NSF. “That was part of theappeal to a science agency.”

A few years later, when NSF began to fundresearch in the social sciences, leading scholarsurged the agency to continue its support for gametheory. John Harsanyi of the University of Californiaat Berkeley began receiving NSF grants in the1960s, as did other major game theorists. “Whatkept NSF interested in game theory,” says Newlon,“was the drive of people working in this field tounderstand how people interact, bargain, andmake decisions, and to do it in a more rigorous,systematic fashion. For years, the problems wereso difficult, given the state of computers and themathematical tools at people’s disposal, that youdidn’t see significant results. Yet NSF hung in there.”

NSF went beyond supporting individual gametheorists. It also sponsored conferences that gavegame theorists the opportunity to gain visibility fortheir work. One such event was the annual StanfordInstitute Conference in Theoretical Economics—run by game theorist Mordecai Kurtz—which NSFbegan to fund in the mid-1960s. Another NSF-funded meeting at the State University of New Yorkhad two goals: to use game theory to advancethe frontiers of economic research and to improvethe skills of graduate students and junior facultyin economics departments. From time to time, NSFinvited proposals for workshops and awardedgrants for computers and other needed equipment.

Into the LaboratoryOriginal work in game theory consisted entirely ofmodels—simplified representations of the under-lying logic in economic decision-making situations—which may have contributed to the business world’sreluctance to accept its usefulness. A theory inphysics or biochemistry can be tested in a con-trolled laboratory situation. In real-world decisionmaking, however, conditions are constantly alteredas a result of changes in technology, governmentinterventions, organizational restructuring, andother factors. The business world and most econ-omists found it hard to see how reading Theoryof Games and Economic Behavior could actuallyhelp them win games or make money.

In the early 1960s, Charles R. Plott and hiscolleagues at the California Institute of Technologystarted to make game theory into an experimentalpursuit. Supported by NSF, his group conducteda series of experiments that helped to answerquestions about one facet of game theory: theideal number of stages in an auction and theiroverall length. Experimentation, which Plott referredto as “debugging,” became increasingly popularin economics as a complement to field researchand theory.

The general idea was to study the operation ofrules, such as auction rules, by creating a simpleprototype of a process to be employed in a com-plex environment. To obtain reliable informationabout how test subjects would choose amongvarious economic alternatives, researchers madethe monetary rewards large enough to induceserious, purposeful behavior. Experiments withprototypes alerted planners to behavior that couldcause a system to go awry. Having advance warn-ing made it possible to change the rules, or thesystem for implementing the rules, while it wasrelatively inexpensive to do so.

Other economists refined and expanded gametheory over the years to encompass more of thecomplex situations that exist in the real world.Finally, in the early 1980s, business schools and

Starting in the 1960s, withsupport from NSF, Charles R.Plott of the California Instituteof Technology made advancesin game theory that paved theway for practical applicationsthree decades later. Here, heoutlines the practical relevanceof NSF-supported economicsresearch:

“The fruits of economicresearch are everywhere.Because NSF is the only dedi-cated source of funding in theUnited States for basic researchin the economic sciences, itsimpact has been large. We seeit in the successful applicationof game theory to the designof the FCC auctions of licens-es for new telecommunica-tions services.

“More broadly, we see theimpact of NSF-supported workin some of the most importanteconomic trends of our life-times, such as deregulation ofairlines and other industries,nongovernmental approachesto environmental protection,and the liberalization of world-wide trade. The recent reex-amination of the ConsumerPrice Index and how it shouldbe measured relies heavily onNSF-sponsored basic researchon price indices.

“In economics it is easy tofind problems that are notsolved, and perhaps are notsolvable in any scientificsense. Yet measured in a cost-benefit sense, the achievementsof economic research standagainst those of any science.”—Charles R. Plott

The Fruits of Economic ResearchAre Everywhere

82 — National Science Foundation

Ph.D. programs in economics began to appreciatethe power of game theory. By the 1990s, it hadall but revolutionized the training of economistsand was a standard analytical tool in businessschools. In 1994, game theory received the ulti-mate recognition with the award of Nobel prizesto Nash, Selten, and Harsanyi—three pioneeringresearchers in the field.

Practical PayoffsFrom a financial standpoint, the big payoff forNSF’s long-standing support came in 1995. TheFederal Communications Commission (FCC) estab-lished a system for using auctions to allocate bandsof the electromagnetic spectrum for a new gener-ation of wireless devices that included cellularphones, pagers, and hand-held computers with email capabilities.

Instead of the standard sealed-bid auction,Stanford University game theorist Paul Milgrom,an NSF grantee, recommended open bidding, whichallows each bidder to see what the others areoffering. Participants could also bid simultane-ously on licenses in the fifty-one zones establishedby the FCC. Game theory’s models of move andcountermove predicted that open bidding wouldreassure bidders who, in trying to avoid the so-called winner’s curse of overpaying, might beexcessively cautious. Open bidding would alsoenable bidders to carry out economically advan-tageous strategies to consolidate holdings inadjacent territories, although FCC rules guaranteedthat no one could obtain a monopoly in any zone.The intended outcome was an optimal solution forall parties. The bidders would get as many licens-es as they were willing to pay for, while the U.S.Treasury would earn the maximum possible.

In the final accounting, the FCC’s 1995 simul-taneous multiple-round auctions raised over $7 billion, setting a new record for the sale ofpublic property. Not only was the decision a

landmark in the recovery of private compensationfor use of a public resource, it also represented avictory for the field of game theory, whose lead-ing scholars had applied what they knew aboutstrategic decision making in recommending anauction design to the FCC.

Game theory has proved its worth in many otherpractical areas, among them management planning.Alvin E. Roth of the University of Pittsburgh appliedgame theory to analyze and recommend match-ing mechanisms for allocating thousands of med-ical interns among hundreds of hospitals in sucha way as to give both the hospitals and the internsthe matches they favor the most. He had thebroader research aim of understanding how marketinstitutions evolved to determine the distributionof doctors and lawyers.

Polls, Markets, and AllocationsGame theory represents one important facet ofdecision science. In fact, decision science dealswith the entire subject of markets—for goods, ser-vices, and ideas, as well as labor. NSF-fundedresearcher Bob Forsythe, at the University of Iowa,provides one example of efforts in this area: Hisinnovative Iowa Electronic Market, started in 1988,offered speculators a “real-money futures market.”This type of market deals with abstract, but mea-surable, items. Participants bet on what the priceof an abstract item, such as pork bellies, will bedays, weeks, or months into the future. Betweenthe time the bet is placed and the point at whichit is paid off, the price of pork bellies will be influ-enced by a succession of economic and politicalevents, including elections and the stock prices offirms in the pork industry.

Forsythe’s “market” actually represented aneffort to elicit more accurate information from votersthan opinion polls provided. Instead of pork bellies,it focused on the electoral prospects of politicalcandidates. “Market prices” summed up what

Decision Sciences — 83

players knew, or thought they knew, about a candi-date’s true chances of success in an election.Participants would win if the candidates on whomthey bet were elected, and would lose if the can-didates lost. Plainly, market participants couldinfluence the result by their own votes; they wouldslightly improve the chances of the candidatesthey bet on by voting for those candidates, andslightly diminish those chances by voting for theopponents. Nevertheless, in its first ten years,the Iowa Electronic Market predicted election outcomes more accurately than did pollsters.

Recently, Forsythe and his colleagues receivedNSF funding to develop instructional materials thatuse the electronic market as a laboratory exerciseto help undergraduates studying economics betterunderstand market concepts

In another innovative area, called “smart markets,”computers have become partners in the processof making allocation decisions, such as assignmentsof airport landing rights and management of gaspipelines and electric distribution systems.Computers process information, coordinate activi-

ties, and monitor allocation situations historicallythought to be impossible to manage with anythingother than a heavily bureaucratic administrativeprocess. For example, the results of NSF-sponsoredresearch have been used to shape a particulartype of market that sets pollution limits and allowsfacilities that generate pollution to trade ‘pollutionpermits’ among themselves, so long as the overalllimit is not exceeded.

Real-World Decision MakingThese examples illustrate the common threadamong the diverse projects of economists who build and test models: All of the projectsare designed to explain more of what occurs inthe real world.

Economic models work well when applied tomarkets and other institutions, in part becausepeople gathered together in large numbers seemto behave as “rational” decision makers. Butindividual behaviors, and the behavior of smallgroups of individuals like the bidders in the FCC

Decision science has broad implications

for all sectors of our society. It plays

a role in understanding outcomes in

financial markets and assessing the

role of a centralized matching system

in ensuring a stable supply of medical

school graduates to hospital residency

programs. Among the many questions

that NSF-funded decision science

researchers are attempting to answer:

how people behave in economic envi-

ronments, how information is distrib-

uted within economic institutions, and

the influence of expectations and beliefs

on decision making.

84 — National Science Foundation

on these tendencies, Paul Slovic, Baruch Fischhoff,and Sarah Lichtenstein of Decision Research inEugene, Oregon, wrote:

“People greatly overestimate the frequency ofdeaths from such dramatic, sensational causesas accidents, homicides, cancer, botulism, andtornadoes, and underestimate the frequency ofdeath from unspectacular causes that claim onevictim at a time and are common in nonfatalform—diabetes, stroke, tuberculosis, asthma,and emphysema . . . . The errors of estimationwe found seemed to reflect the working of amental shortcut or ‘heuristic’ that people commonlyuse when they judge the likelihood of risky events.”

The authors explain that people judge an eventas likely or frequent if instances of it are easy toimagine or recall. On the other hand, individualsoften don’t bother to consider information that isunavailable or incomplete. Every time we makedecisions that involve probabilities, we confirmthe reality of the phrase “out of sight, out of mind.”

Another area where humans are systematicallyerror-prone involves what economists call utility.We frequently face choices between doing the safething and taking a risk. One example is the choicebetween driving to work on secondary roads ortaking the interstate, which usually saves severalminutes but can occasionally take an extra half-houror more because of back-ups. Another is thedecision between investing in a safe money mar-ket account and taking a flier on a volatile stock.

No two people feel exactly the same aboutwhich risks are worth taking. The concept of utilitycombines several factors in decision making: therange of possible outcomes for a particular choice,the probability associated with each outcome,and an individual’s subjective method of rankingthe choices.

Imagine choosing between two tempting oppor-tunities. One is a coin toss—heads you win$1,000, tails you win nothing. The other is a surething—an envelope with $500 inside. Do youchoose the safe and sure $500 or the 50/50

“A bird in hand is worth two in the

bush” describes one action a person

might take to minimize risk and maxi-

mize utility—the real or perceived

ability of a product or service to satisfy

a need or desire. Utility theory attempts

to define the many factors that influence

how people make decisions and to pre-

dict how an individual will behave when

faced with difficult choices.

auctions, often seem inconsistent with rationality.NSF’s Decision, Risk, and Management ScienceProgram, within the Directorate for Social,Behavioral, and Economic Sciences, supportsleading scholars in the decision sciences wholook at these inconsistencies from another pointof view. They try to determine the nature and origins of systematic errors in individual decisionmaking, and use game theory to provide sets ofstrategies for anticipating and dealing with them.

Systematic errors abound in decisions thatinvolve probability. Most people are not good atestimating the statistical likelihood of events, andtheir mistakes fall into distinct patterns. Reporting

All in a Day’s WorkImagine you are a cab driver. Whatyou earn in a given day varies accord-ing to the weather, time of year, con-ventions in town, and other factors.As a rational person, you want tomaximize both your income and yourleisure time. To achieve that, youshould work more hours when wagesare high and fewer hours whenwages are low.

What that means is that cab driversshould work more hours on busy daysand fewer hours on slow days. Dothey? Not at all; they do the opposite.

Colin Camerer of the CaliforniaInstitute of Technology made this dis-covery when he and his colleaguesinterviewed a large sample of cabdrivers. They found that the cabbiesdecide how many hours to work bysetting a target amount of moneythey want to make each day. Whenthey reach their target, they stopworking. So on busy days, they workfewer hours than on slow days.

Why? Camerer suggested thatworking an extra hour simply may

not be worth an hour of leisure time;in the language of economics, themarginal utility is too low. On theother hand, it may not be the moneyas much as cab drivers’ feelingsabout the money—or, more precisely,how they think they may feel if theydepart from their usual workinghabits. Will a cab driver who worksan extra hour or two on a busy dayfeel later that it wasn’t worth theeffort? Will one who knocks off earlyon a slow day feel guilty about it?Setting a target may be a way toavoid regrets.

NSF has supported Camerer andothers in their efforts to explain thisand other paradoxes that character-ize human economic behavior. Fromthe beginning, decision scienceresearch has had the goal of a betterfix on people’s feelings about wages,leisure, and tradeoffs between them,with implications for labor relations,productivity, and competitivenessacross a wide spectrum of industries.

A B

A B

33%33%CHANCECHANCE600600LIVELIVE

66%66%CHANCECHANCE

0LIVELIVE

100%100%CHANCECHANCE200200LIVELIVE

33%CHANCE600LIVE

66%CHANCE

0LIVE

100%CHANCE200LIVE

100%100%CHANCECHANCE400400DIEDIE

33%33%CHANCECHANCE

0DIEDIE

66%66%CHANCECHANCE600600DIEDIE

100%CHANCE400DIE

33%CHANCE

0DIE

66%CHANCE600DIE

86 — National Science Foundation

chance to win $1,000? An economically rationalperson makes the choice that reflects the highestpersonal utility. The wealthier that individual is, forinstance, and the more he or she likes to gamble,the higher is the utility of the risky 50/50 choiceversus the sure thing.

Questioning Utility TheoryUtility theory postulates that it should not matterhow alternatives are presented. Once we knowwhat’s at stake, and the risks involved, we shouldhave enough awareness of ourselves to make thechoice that serves us best.

In fact, psychologists Daniel Kahneman ofPrinceton and the late Amos Tversky of Stanforddemonstrated that the way alternatives are framedcan make quite a difference in our choices. In oneof their most famous studies, they presentedpeople with a choice between two programs thataddressed a public health threat to the lives of

600 people. When the outcomes of the programswere described as (a) saving 200 lives for sure,or (b) a one-third chance to save 600 lives and atwo-thirds chance to save no one, most respon-dents preferred the first option. But when theoutcomes were presented as (a) 400 people dyingfor sure, or (b) a two-thirds chance of 600 peopledying and a one-third chance that no one would die,most respondents preferred the second option.Of course, the two versions of the problem arethe same, because the people who will be savedin one version are the same people who will notdie in the other. What happens here is that peopleare generally risk-averse in choices between suregains and favorable gambles, and generally risk-seeking in choices between sure losses andunfavorable gambles. “Some propensities,” pointsout former NSF Program Director Jonathan Leland,“are so ingrained that the trick is to help peopleunderstand why their decisions are bad.” Noone, it seems, is immune to the power of thewell-chosen word.

NSF-funded researchers Daniel Kahneman

and the late Amos Tversky were instru-

mental in the development of rational

choice theory. First used to explain and

predict human behavior in the market,

advocates of rational choice theory

believe that it helps integrate and

explain the widest range of human

behavior—including who people vote

for, what they buy at the grocery

store, and how they will react when

faced with a difficult decision about

medical treatment.

with higher expected payoffs. With NSF support,M.H. Bazerman of Northwestern University docu-mented these stubborn tendencies in a variety ofsettings, and proposed corrective measures thatorganizations can take to counteract them.

Just as it supported game theory from the veryearly stages, NSF has funded research on the appli-cation of psychology to economic decision makingfrom the field’s infancy. That support yielded evenfaster dividends: Within a few years, the researchhad given rise to popular books advising managersand others on how to correct for error-prone ten-dencies and make better decisions. “We knowthat people bargain and interact, that informationis imperfect, that there are coordination problems,”NSF’s Daniel Newlon explains. “NSF’s long-termagenda is to understand these things. Even ifthey’re too difficult to understand at a given time,you keep plugging away. That’s science.”

Decision Sciences — 87

Why We Make Foolish DecisionsIndividuals frequently ensure poor decision makingby failing to obtain even the most basic informationnecessary to make intelligent choices. Take, forexample, the NSF-supported research of HowardKunreuther of the University of Pennsylvania’sWharton School. He and his colleagues observedthat most people living in areas subject to suchnatural disasters as floods, earthquakes, andhurricanes take no steps to protect themselves.Not only do they not take precautions proven to becost-effective, such as strapping down their waterheaters or bolting their houses to foundations;they also neglect to buy insurance, even when thefederal government provides substantial subsidies.

What accounts for such apparently foolish deci-sion making? Financial constraints play a role. ButKunreuther found the main reason to be a beliefthat the disaster “will not happen here.” Hisresearch suggested “that people refuse to attendto or worry about events whose probability is belowsome threshold.” The expected utility model, headded, “is an inadequate description of the choiceprocess regarding insurance purchases.”

Kunreuther next applied decision science todevising an alternative hypothesis for the behavior.First, he posited, individuals must perceive that ahazard poses a problem for them. Then they searchfor ways, including the purchase of insurance, tomitigate future losses. Finally, they decide whetherto buy coverage. They usually base that decisionon simple criteria, such as whether they knowanyone with coverage. The research showed that,since people do not base their purchasing decisionson a cost-benefit analysis, premium subsidiesalone did not provide the necessary impetus topersuade individuals to buy flood insurance.

Decision making within organizations is alsoriddled with systematic bias. One example is thefamiliar phenomenon of throwing good money afterbad. Corporations frequently become trapped in asituation where, instead of abandoning a failingproject, they continue to invest money and/oremotion in it, at the expense of alternative projects

NSF Division of Social and Economic SciencesEconomics Programwww.nsf.gov/sbe/ses/econ/start.htm

NSF Decision, Risk, and Management Sciences Programwww.nsf.gov/sbe/ses/drms/start.htm

Consumer Price Indexhttp://stats.bls.gov/cpihome.htm

Decision Researchwww.decisionresearch.org

Iowa Electronic Marketwww.biz.uiowa.edu/iem/index.html

The Nobel FoundationNobel Prize in Economic Scienceswww.nobel.se/economics/laureates/1994/index.html

Stanford Encyclopedia of PhilosophyGame Theoryhttp://plato.stanford.edu/entries/game-theory/

Stanford Institute for Theoretical Economicswww.stanford.edu/group/SITE/siteprog.html

To Learn More