Audience Decision-Making Expertise 27
As team members interact and acquire greater expertise in making specifi c types of decisions,
their shared mental models change and become more refi ned,^234 with the more highly skilled
members of the team using them much more successfully than their less skilled colleagues.^235
Expert Audiences vs. Linear Models and Normative Rules
Linear Models: The Gold Standard of Rational Decision Making
As we have seen, expert audiences typically consider multiple decision criteria when making deci-
sions and their decisions typically produce superior outcomes to those of novices. But do experts
consistently make the best decisions possible given their decision criteria? And do experts con-
sistently follow the rules of normative, or proper, decision making? If experts followed normative
rules, such as those prescribed by multi-attribute utility theory (MAUT), they would consistently
choose those alternatives that maximized their values.^236
To answer the fi rst question—“Do expert audiences consistently make the best decisions pos-
sible given their decision criteria?”—decision researchers use a statistical technique called multiple
regression analysis to identify how much weight (if any) the experts should ideally give each of their
decision criteria in order to choose the best alternative. The result is called a linear model. Linear
models are usually represented as equations. For example, a linear model for deciding on the best
MBA applicants might to represented as:
Applicant value = 0.2 (GMAT score) + 0.3 (GPA) + 0.4 (Communication Skills)
+ 0.1 (Work Experience)
But linear models can also be expressed in the form of value trees, or, like decision schemata, in the
form of decision matrices.
The decisions experts make are usually inferior to those computed by linear models. Lin-
ear models are more accurate than clinical psychologists in diagnosing psychiatric patients.^237
Linear models are more accurate than medical doctors in diagnosing medical patients.^238 Lin-
ear models are better than faculty members on admissions committees at choosing the best
students for graduate school.^239 Linear models are more accurate than commercial bankers in
deciding which fi rms are most likely to go bankrupt.^240 Linear models are more accurate than
expert investors in predicting stock prices.^241 In fact, hundreds of studies have consistently
shown that in virtually every domain, statistical predictions produced by linear models outper-
form those of experts.^242
Even bootstrapped models , improper linear models that use the same weights experts give their
decision criteria, produce better decisions than the experts they model.^243 A bootstrapped model
can be expressed in the same three ways as a linear model. Bootstrapped models are called improper
because, unlike linear models, they cannot produce an optimal decision. Surprisingly, another type
of improper linear model, the equal-weight additive model, which weights all of the experts’ deci-
sion criteria equally, also produces better decisions than experts.^244
The Causes of Individual Experts’ Normally Inferior Performance
One reason linear and bootstrapped models outperform experts is that experts do not consis-
tently follow the rules of normative decision making when choosing among alternatives.^245