decision models: some lessons from the field g. klein, r. calderwood presented by: hilary ince

26
Decision Models: Some Lessons From the Field G. Klein, R. Calderwood Presented by: Hilary Ince

Upload: cecil-wood

Post on 27-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Decision Models: Some Lessons From

the FieldG. Klein, R. Calderwood

Presented by: Hilary Ince

IntroductionPrevailing paradigms in decision research do not

apply well to the real world

Although diverse, decision theories center around the decision tree

Decision maker presented as an analystGenerates alternatives, evaluates outcomes

Decision event is the focus of concern

IntroductionMain view is that humans are suboptimal, biased

decision makersTraining focuses on reducing bias

These programs are of limited benefitNovice decision makers are not good decision

analysts, and experts can make faster decisions

Primary effort in naturalistic decision making is situation assessment, not the moment of choice

IntroductionClassical decision theory is useful for prescribing

how to make better decisionsNot describing how people actually make them!

Classical decision models aren’t all bad!Generic, can be widely appliedTechniques ensure everyone speaks the same

language, uses the same metricsGood for framing and guiding decision making

When all underlying assumptions are met, the techniques ensure optimal selection

IntroductionAssumptions can be problematic

Goals can be isolatedUtilities can be assessed independent of contextProbabilities can be accurately estimatedChoices, goals, and evidence are carefully definedUtilities of an outcome are independent of other

outcomes

All difficult to meet in the real world

Can Goals Be Isolated?Slow down an enemy by denying access to key

roads

Goal is linked to many others, like using the roads for a counterattack in a few days

It can be dangerous to oversimplify the interactions of goals

Can Utilities Be Assessed Independent of Context?

Chess is a relatively restricted environment

Chess masters do not rate moves using a standard set of dimensions

Moves are evaluated in context

Can Probabilities Be Accurately Estimated?

Even without time pressure and stress, this is difficult

Will Choices, Goals, and Evidence Be Clearly Defined?The decision maker may not know the desired

end state

Fireman rushing to a fire Is goal to put out fire, or stop it from spreading?Can call for reinforcements, but cannot leave

district vulnerable to other firesActual goal is to do the best with what he has…

not well defined

Are the Utilities of Outcomes Independent of Other Outcomes?Possible for static and limited tasks

In complex operations, outcomes are interrelated

Recognitional DecisionsResearch has been done on urban fireground

commanders (FGCs)Observations and retrospective account of

emergency eventsExamples: whether to initiate search and rescue,

whether to initiate an offensive attack or focus on defense, and where to allocate resources

Difficult to fit their behavior into decision treeThey resisted describing “making choices”,

“considering alternatives”, or “assessing probabilities”

Recognitional DecisionsSaw themselves as acting and reacting based on

prior experience, planning, monitoring, and modifying plans to meet constraints

No evidence of exhaustive option generation

FCGs insisted that they did not deliberate about advantages and disadvantages of options

They “watched” the choice play out mentally to see if anything went wrong, then did the same with another option if the first didn’t work NOT concurrent in the majority of decisions (only 18.6%

concurrent)

Recognitional Decisions26 interviews conducted, 32 incidents studied,

156 decisions probed in detail

Decision point defined as the instances when the FGC could agree, after the fact, that alternative actions were possible

18.6% concurrent, 81.4% recongitional decisions

These strategies can be described by the recognition primed decision (RPD) model

Recognitional DecisionsRPD Model:

A person understands a situation based on familiarity with previous experience

Helps recognize feasible goals, relevant cues, expectancies to monitor, and plausible actions

Decision maker is more likely to generate a viable option as the first one to consider

Evaluation of the option is through mental simulation

Recognitional DecisionsThe model has distinct features:

Assesses that experienced decision makers can generate plausible options rather than randomly generate a large option set

Decision maker can use serial process rather the comparing options to each other

Decision makers can use satisficing rather then taking time to pick the best possible option

Evaluation can involve mental simulationEmphasizes situation understanding

Field Research: Tasks and Paradigms

Many other studies have shown limitations of the decision tree modelCommand and control, business, jury deliberation

Author suggests a continuum of RPDs and analytical decisionsDifferent situations call for different decision

making

Decision Support Systems and Training for Recognitional Decision Making

Decision support systems based on recognitional decisions would improve speed and accuracy of situational understandingAnalogue displays rather than alphanumeric data

Databases of prior cases could be helpful

Emphasis of training should be sensitivity to critical factors, effective use of expectancies, ability to anticipate important contingencies, and situational understanding

Naturalistic Decision Making

G. KleinPresented by: Hilary Ince

OriginsMajor contribution of naturalistic decision

making (NDM) research is to describe how people make decisions in the real world

When NDM emerged, it was only clear how people DIDN’T make decisions

Training methods and decision support tools developed with traditional decision making theory in mind weren’t useful

OriginsNDN began with field research on decision

making strategies

Looked at how people made actual decisions under stress, not how people are suboptimal

Army and Navy were initial financial backersNavy interested because of USS Vincennes

incident

OriginsAt least nine NDM theories were developed in

parallel before the first NDM conference in 1989Hammond’s cognitive continuum theory states

that decisions vary based on intuitive and analytical processes used

Rasmussen’s model of cognitive control distinguished skill-based, rule-based, and knowledge-based behavior

Klein’s recognition primed decision model

OriginsEveryone reached the same conclusion:

People don’t generate and compare options!Prior experience is important

Making the decision means committing to an alternative even if other alternatives haven’t been examined

Recognition Primed Decision Model

RPD describes how people use past experiences to make decisions (recognizable patterns)Patterns highlight most relevant cues, provide

expectancies, identify plausible goals, and suggest typical reactions

RPD model describes how good decisions are made without comparing options

Contributions of NDMSpurred development of cognitive field research

and cognitive task analysis methods

NDM has affected Army doctrineArmy Field Manual on Command and Control has a

section on intuitive decision making

Swedish armed forces use the RPD model adapted for military planning guidance

Decision training has been influencedTactical Decision Making Under Stress (TADMUS)

after USS Vincennes incident

Future of NDMBeing used to study situational awareness,

sensemaking, planning and replanning, and their interactions

Future may focus on team decision making