Julgamento Sublinhado
Julgamento Sublinhado
Julgamento Sublinhado
1
What Is Ecological Rationality?
Peter M. Todd
Gerd Gigerenzer
“
M ore information is always better, full information is best. More
computation is always better, optimization is best.” More-is-better
ideals such as these have long shaped our vision of rationality. The
philosopher Rudolf Carnap (1947), for instance, proposed the
“principle of total evidence,” which is the recommendation to use
all the available evidence when estimating a probability. The statis-
tician I. J. Good (1967) argued, similarly, that it is irrational to make
observations without using them. Going back further in time, the
Old Testament says that God created humans in his image (Genesis
1:26), and it might not be entirely accidental that some form of
omniscience (including knowledge of all relevant probabilities
and utilities) and omnipotence (including the ability to compute
complex functions in a blink) has sneaked into models of human
cognition. Many theories in the cognitive sciences and economics
have recreated humans in this heavenly image—from Bayesian
models to exemplar models to the maximization of expected utility.
Yet as far as we can tell, humans and other animals have always
relied on simple strategies or heuristics to solve adaptive problems,
ignoring most information and eschewing much computation
rather than aiming for as much as possible of both. In this book,
we argue that in an uncertain world, more information and com-
putation is not always better. Most important, we ask why and
when less can be more. The answers to this question constitute the
idea of ecological rationality, how we are able to achieve intelli-
gence in the world by using simple heuristics in appropriate con-
texts. Ecological rationality stems in part from the nature of those
3
4 THE RESEARCH AGENDA
Making Money
In 1990, Harry Markowitz received the Nobel Prize in Economics
for his path-breaking work on optimal asset allocation. He addressed
a vital investment problem that everyone faces in some form or
other, be it saving for retirement or earning money on the stock
market: How to invest your money in N available assets. It would
be risky to put everything in one basket; therefore, it makes sense
to diversify. But how? Markowitz (1952) derived the optimal rule
for allocating wealth across assets, known as the mean–variance
portfolio, because it maximizes the return (mean) and minimizes
the risk (variance). When considering his own retirement invest-
ments, we could be forgiven for imagining that Markowitz used his
award-winning optimization technique. But he did not. He relied
instead on a simple heuristic:
make better decisions. Yet our point is not that simple heuristics
are better than optimization methods, nor the opposite, as is typi-
cally assumed. No heuristic or optimizing strategy is the best in all
worlds. Rather, we must always ask, in what environments does a
given heuristic perform better than a complex strategy, and when is
the opposite true? This is the question of the ecological rationality
of a heuristic. The answer requires analyzing the information-
processing mechanism of the heuristic, the information structures
of the environment, and the match between the two. For the choice
between 1/N and the mean–variance portfolio, the relevant envi-
ronmental features include (a) degree of uncertainty, (b) number
N of alternatives, and (c) size of the learning sample.
It is difficult to predict the future performance of funds because
uncertainty is high. The size of the learning sample is the estima-
tion window, with 5 to 10 years of data typically being used to cali-
brate portfolio models in investment practice. The 1/N rule tends to
outperform the mean–variance portfolio if uncertainty is high, the
number of alternatives is large, and the learning sample is small.
This qualitative insight allows us to ask a quantitative question: If
we have 50 alternatives, how large a learning sample do we need so
that the mean–variance portfolio eventually outperforms the simple
heuristic? The answer is: 500 years of stock data (DeMiguel et al.,
2009). Thus, if you started keeping track of your investments now,
in the 26th century optimization would finally pay off, assuming
that the same funds, and the stock market, are still around.
Catching Balls
Now let us think about sports, where players are also faced with
challenging, often emotionally charged problems. How do players
catch a fly ball? If you ask professional players, they may well stare
at you blankly and respond that they had never thought about it—
they just run to the ball and catch it. But how do players know
where to run? A standard account is that minds solve such complex
problems with complex algorithms. An obvious candidate complex
algorithm is that players unconsciously estimate the ball’s trajec-
tory and run as fast as possible to the spot where the ball will hit
the ground. How else could it work? In The Selfish Gene, biologist
Richard Dawkins (1989, p. 96) discusses exactly this:
When a man throws a ball high in the air and catches it again, he
behaves as if he had solved a set of differential equations in pre-
dicting the trajectory of the ball. He may neither know nor care
what a differential equation is, but this does not affect his skill
with the ball. At some subconscious level, something function-
ally equivalent to the mathematical calculation is going on.
6 THE RESEARCH AGENDA
Gaze heuristic: Fixate your gaze on the ball, start running, and
adjust your running speed so that the angle of gaze remains
constant.
The angle of gaze is the angle between the eye and the ball, rela-
tive to the ground. Players who use this rule do not need to measure
wind, air resistance, spin, or the other causal variables. They can
get away with ignoring all these pieces of causal information. All
the relevant facts are contained in only one variable: the angle of
gaze. Note that players using the gaze heuristic are not able to com-
pute the point at which the ball will land, just as demonstrated by
the experimental results. But the heuristic nevertheless leads them
to the landing point in time to make the catch.
Like the 1/N rule, the gaze heuristic is successful in a particular
class of situations, not in all cases, and the study of its ecological
rationality aims at identifying that class. As many ball players say,
the hardest ball to catch is the one that heads straight at you, a situ-
ation in which the gaze heuristic is of no use. As mentioned before,
the gaze heuristic works in situations where the ball is already high
WHAT IS ECOLOGICAL RATIONALITY? 7
in the air, but it fails if applied right when the ball is at the begin-
ning of its flight. However, in this different environmental condi-
tion, players do not need a completely new heuristic—just a slightly
modified one, with a different final step (McBeath, Shaffer, & Kaiser,
1995; Shaffer, Krauchunas, Eddy, & McBeath, 2004):
Modified gaze heuristic: Fixate your gaze on the ball, start run-
ning, and adjust your running speed so that the image of the
ball rises at a constant rate.
What Is a Heuristic?
(Continued )
Table 1-1: Twelve Well-Studied Heuristics With Evidence of Use in the Adaptive Toolbox of Humans
Heuristic Definition Ecologically rational if: Surprising findings (examples)
Gaze heuristic To catch a ball, fix your gaze on it, The ball is coming down Balls will be caught while
(Gigerenzer, 2007; start running, and adjust your from overhead running, possibly on a curved
McBeath, Shaffer, & running speed so that the angle path
Kaiser, 1995) of gaze remains constant.
1/N rule (DeMiguel, Allocate resources equally to each High unpredictability, Can outperform optimal asset
Garlappi, & Uppal, of N alternatives. small learning sample, allocation portfolios
2009) large N
Default heuristic If there is a default, follow it. Values of those who Explains why advertising has
(Johnson & Goldstein, set defaults match little effect on organ donor
2003; chapter 16) those of the decision registration; predicts behavior
maker; consequences when trait and preference
of a choice are hard to theories fail
foresee
Tit-for-tat (Axelrod, Cooperate first and then imitate The other players also Can lead to a higher payoff than
1984) your partner’s last behavior. play tit-for-tat “rational” strategies (e.g. by
backward induction)
Imitate the majority Determine the behavior followed Environment is stable or A driving force in bonding,
(Boyd & Richerson, by the majority of people in only changes slowly; group identification, and moral
2005) your group and imitate it. info search is costly or behavior
time consuming
Imitate the successful Determine the most successful Individual learning is A driving force in cultural
(Boyd & Richerson, person and imitate his or her slow; info search is evolution
2005) behavior. costly or time consuming
Note. For formal definitions and conditions concerning ecological rationality and surprising findings, see references indicated and related chapters
in this book.
WHAT IS ECOLOGICAL RATIONALITY? 11
Evolved Capacities
Building blocks of heuristics are generally based on evolved cap-
acities. For instance, in the gaze heuristic, to keep the gaze angle
constant an organism needs the capacity to track an object visually
against a noisy background—something that no modern robot or
computer vision system can do as well as organisms (e.g., humans)
that have evolved to follow targets. When we use the term evolved
capacity, we refer to a product of nature and nurture—a capacity
that is prepared by the genes of a species but usually needs experi-
ence to be fully expressed. For instance, 3-month-old babies spon-
taneously practice holding their gaze on moving targets, such as
mobiles hanging over their crib. Evolved capacities are one reason
why simple heuristics can perform so well: They enable solutions
to complex problems that are fundamentally different from the
mathematically inspired ideal of humans and animals somehow
optimizing their choices. Other capacities underlying heuristic
building blocks include recognition memory, which the recogni-
tion heuristic and fluency heuristics exploit, and counting and
recall, which take-the-best and similar heuristics can use to esti-
mate cue orders.
April 8, 1779
If you doubt, set down all the Reasons, pro and con, in
opposite Columns on a Sheet of Paper, and when you have
considered them two or three Days, perform an Operation
similar to that in some questions of Algebra; observe what
Reasons or Motives in each Column are equal in weight, one
to one, one to two, two to three, or the like, and when you
have struck out from both Sides all the Equalities, you will see
in which column remains the Balance.… This kind of Moral
Algebra I have often practiced in important and dubious
Concerns, and tho’ it cannot be mathematically exact, I have
WHAT IS ECOLOGICAL RATIONALITY? 13
to ignore much of the available information and use fast and frugal
heuristics. And yet this approach is often resisted: When a forecast-
ing model does not predict a criterion, such as the performance of
funds, as well as hoped, the gut reaction of many people, experts
and novices alike, is to do the opposite and call for more informa-
tion and more computation. The possibility that the solution may
lie in eliminating information and fancy computation is still
unimaginable for many and hard to digest even after it has been
demonstrated again and again (see chapter 3).
Sample Size In general, the smaller the sample size of available data
in the environment, the larger the advantage for simple heuristics.
One of the reasons is that complex statistical models have to esti-
mate their parameter values from past data, and if the sample size is
small, then the resulting error due to “variance” can exceed the error
due to “bias” in competing heuristics (see chapter 2). What consti-
tutes a small sample size depends on the degree of uncertainty, as
can be seen in the investment problem, where uncertainty is high:
In this case, a sample size of hundreds of years of stock data is
needed for the mean–variance portfolio to surpass the accuracy of
the 1/N rule.
There are many other important types of environment struc-
ture relevant for understanding ecological rationality. Two of
the major ones also considered in this book are redundancy and
variability.
(e.g., a disastrous hurricane) and other people (e.g., the public reac-
tion to a disaster). Each of the heuristics in Table 1-1 can be applied
to social objects (e.g., whom to hire, to trust, to marry) as well
as to physical objects (e.g., what goods to buy). As an example, the
recognition heuristic (see chapters 5 and 6) exploits environment
structures in which lack of recognition is valuable information and
aids inferences about, say, what microbrew to order and where to
invest, but also whom to talk to and whom to trust (“don’t ride with
a stranger”). Similarly, a satisficing heuristic can be used to select a
pair of jeans but also choose a mate (Todd & Miller, 1999), and the
1/N rule can help investors to diversify but also guide parents in
allocating their time and resources equally to their children.
Environment structures are also deliberately created by institu-
tions to influence behavior. Sometimes this is felicitous, as when
governments figure out how to get citizens to donate organs by
default, or design traffic laws for intersection right-of-way in a hier-
archical manner that matches people’s one-reason decision mecha-
nisms (chapter 16). In other cases, institutions create environments
that do not fit well with people’s cognitive processes and instead
cloud minds, accidentally or deliberately. For instance, informa-
tion about medical treatments is often represented in ways that
make benefits appear huge and harms inconsequential (chapter 17),
casinos set up gambling environments with cues that make gam-
blers believe the chance of winning is greater than it really is
(chapter 16), and store displays and shopping websites are crowded
with long lists of features of numerous products that can confuse cus-
tomers with information overload (Fasolo, McClelland, & Todd, 2007).
But there are ways to fix such problematic designs and make new
ones that people can readily find their way through, as we will see.
Finally, environment structure can emerge without design
through the social interactions of multiple decision makers. For
instance, people choosing a city to move to are often attracted by
large, vibrant metropolises, so that the “big get bigger,” which can
result in a J-shaped (or power-law) distribution of city populations
(a few teeming burgs, a number of medium-sized ones, and numer-
ous smaller towns). Such an emergent distribution, which is
seen in many domains ranging from book sales to website visits,
can in turn be exploited by heuristics for choice or estimation
(chapter 15). Similarly, drivers seeking a parking space using a par-
ticular heuristic create a pattern of available spots that serves as the
environment for future drivers to search through with their own
strategies, which may or may not fit that environment structure
(chapter 18). In these cases, individuals are, through the effects
of their own choices, shaping the environment in which they
and others must make further choices, creating the possibility of a
co-adapting loop between mind and world.
20 THE RESEARCH AGENDA
This view starts from the dictum that more is always better, as
described at the beginning of this chapter—more information and
computation would result in greater accuracy. But since in the real
world, so the argument goes, information is not free and computa-
tion takes time that could be spent on other things (Todd, 2001),
there is a point where the costs of further search exceed the
benefits. This assumed trade-off underlies optimization-under-
constraints theories of decision making, in which information
search in the external world (e.g., Stigler, 1961) or in memory (e.g.,
Anderson, 1990) is terminated when the expected costs exceed its
benefits. Similarly, the seminal analysis of the adaptive decision
maker (Payne et al., 1993) is built around the assumption that heu-
ristics achieve a beneficial trade-off between accuracy and effort,
where effort is a function of the amount of information and compu-
tation consumed. And indeed, as has been shown by Payne et al.’s
research and much since, heuristics can save effort.
The major discovery, however, is that saving effort does not nec-
essarily lead to a loss in accuracy. The trade-off is unnecessary.
Heuristics can be faster and more accurate than strategies that use
more information and more computation, including optimization
techniques. Our analysis of the ecological rationality of heuristics
goes beyond the incorrect universal assumption of effort–accuracy
trade-offs to ask empirically where less information and computa-
tion leads to more accurate judgments—that is, where less effortful
heuristics are more accurate than more costly methods.
These less-is-more effects have been popping up in a variety of
domains for years, but have been routinely ignored, as documented
in chapter 3. Now, though, a critical mass of instances is being
assembled, as shown throughout this book. For instance, in an age
in which companies maintain databases of their customers, com-
plete with historical purchase data, a key question becomes pre-
dicting which customers are likely to purchase again in a given
timeframe and which will be inactive. Wübben and Wangenheim
WHAT IS ECOLOGICAL RATIONALITY? 27
Methodology
of the gaze heuristic and its variants (Saxberg, 1987; Shaffer &
McBeath, 2005 Todd, 1981). Furthermore, the predicted process of
trajectory computation implies that players will calculate where a
ball will land, whereas the gaze heuristic makes no such predic-
tion. Comparing these process-level predictions can help explain
an apparent fallacy on the part of expert players—that they are
not able to say where a ball will come down (e.g., Saxberg, 1987).
When using the gaze heuristic, players would not have this ability,
because they would not need it to catch the ball. Such an analysis
of heuristics and their ecological rationality can thus help research-
ers to avoid misjudging adaptive behavior as fallacies (Gigerenzer,
2000).
There are a number of useful methodological considerations that
are prompted by the study of ecological rationality. First, research
should proceed by means of testing multiple models of heuristics
(or other strategies) comparatively, determining which perform best
in a particular environment and which best predict behavior
observed in that environment. This enables finding better models
than those that already exist, not just assessing only one model in
isolation and then proclaiming that it fits the data or not. Second,
given the evidence discussed earlier for individual differences in
the use of heuristics, the tests of predictive accuracy should be
done at the level of each individual’s behavior, not in terms of
sample averages that may represent few or none of the individuals.
Finally, because individuals may vary in their own use of heuristics
as they explore a new problem, experiments should leave individu-
als sufficient time to learn about the alternatives and cues, and
researchers should not confuse trial and error exploration at the
beginning of an experiment as evidence for weighting and adding
of all information.
Several studies of heuristics exemplify these methodological
criteria. For instance, Bergert and Nosofsky (2007) formulated a
stochastic version of take-the-best and tested it against an additive-
weighting model at the individual level. They concluded that the
“vast majority of subjects” (p. 107) adopted the take-the-best strat-
egy. Another study by Nosofsky and Bergert (2007) compared take-
the-best with both additive-weighting and exemplar models of
categorization and concluded that “most did not use an exemplar-
based strategy” but followed the response time predictions of take-
the-best. There are also examples where not following some of these
criteria has led to results that are difficult to interpret. For instance,
if a study on how people learn about and use cues does not provide
enough trials for subjects to explore and distinguish those cues,
then lack of learning cannot be used as evidence of inability to learn
or failure to use a particular heuristic (e.g., Gigerenzer, Hertwig, &
30 THE RESEARCH AGENDA