CAPÍTULO 3. DISEÑO DE LA ESTRATEGIA COMERCIAL PARA LA TIENDA EL YAREY
3.3. Resultados de la aplicación del procedimiento seleccionado para el
I will now discuss a few examples of how disagreements about rationality have led to
disagreements within the confirmation bias literature, beyond some of the examples already discussed earlier in the section on bias.
As I summarised in the last section, many disagreements about the ‘rationality’ of con- firmation bias might be understood in terms of different construals of the problem being
solved. On the broadest level, the relevant problems are how to choose new information to test hypotheses, and how to draw inferences from that information to update existing
beliefs. From this standpoint, the relevant normative models, understood as the formal 4Though consistency might naturally be thought of as the most minimal standard rationality should
meet, in practice this might not be as minimal a requirement as it seems. Given the wide range of situations we encounter, keeping our beliefs and decisions consistent across those situations may be an incredibly demanding task. Sometimes it may be that some degree of inconsistency is a price worth paying to save time/energy, or to better achieve some other goal. This helps explain why normative rationality, though in a sense the ‘simplest’ type, is also often the most demanding.
solutions to these problems, are Bayesian inference and optimal data selection. A con- firmation bias then exists and is a sign of irrationality if people’s reasoning processes
systematically deviate from these normative standards, and do so in a way that favour prior beliefs. In an earlier section, we listed a number of different ways judgements
might deviate from Bayes’ theorem that could lead to a confirmation bias, building on discussion in Fischoff and Beyth-Marom (1983). A confirmation bias in this sense also
seems likely to be epistemically irrational - i.e. to come at a cost to accuracy.
However, there are other ways we might understand the rationality of a confirmation
bias. People often have other important goals than accuracy and consistency, and it’s possible that in some of the environments people typically encounter with the cognitive
constraints they face, the strategies they use may be effective at achieving those goals. For example, some have argued that forms of confirmation bias, though irrational from a
purely epistemic perspective, might help people to achieve other goals - such as protect- ing one’s ego (Hart et al., 2009), or mental health (Nickerson, 1998). Though strategies
like falsification might be normatively appropriate in certain abstract rule-discovery ex- periments, something more like a positive-test strategy might be ecologically rational
given features of the kinds of hypotheses we typically have to test. Perfors and Navarro (2009) bring together some different perspectives on the rationality of hypothesis-testing
behaviour by suggesting that a positive-test strategy may be rational on the assumption that hypotheses are ‘sparse’: that they are rare - true for less than half of the logically
possible entities (Oaksford and Chater, 1994) - or in the most extreme form determin- istic rules that only predict a single possibility at each trial (Austerweil and Griffiths,
2008).
Taking into account cognitive constraints, it has also been argued that some forms of
confirmation bias might optimally balance the costs of different kinds of errors. Nis- bett and Ross (1980) point out that, given practical time constraints, the tendency to
persevere in one’s current hypothesis might be a ‘stabilising hedge’ against changing one’s mind too frequently. Friedrich (1993) suggests that human inference processes
are designed to identify potential rewards and minimize costly errors, a very different task from pure truth-detection - given this goal and cognitive limitations, he argues, a
confirmation bias may be viewed as rational. Friedrich gives the example of an employer with the hypothesis that extroverts make good salespeople. If his main goal was test- ing the truth of this hypothesis, then he would want to seek potentially disconfirming
evidence by trialling an introverted salesperson. But in practice, false-positive errors (hiring an introvert who turns out to be a poor salesperson) are much more costly than
false-negative errors (hiring an extrovert who is good, and missing out on an introvert who would also have been good.) So seeking to minimise the probability of committing
the former type of error, by seeking to confirm one’s hypothesis, might be rational from a more pragmatic perspective.
Tooby and Cosmides (1992) have emphasised understanding confirmation bias from an evolutionary perspective - suggesting that selection pressures would favour strate-
gies that solved biologically significant problems rather than those that were perfectly consistent or truth-seeking. So we might similarly expect mechanisms that minimise
false-negative errors (undetected predators) by tolerating false positives (assuming all bears are dangerous) to out-reproduce mechanisms more driven by falsification testing.5
It may also be that the strategies we use are not so effective at achieving the relevant goals, but since calculating probabilities and value of information and Bayes’ rule is
incredibly cognitively demanding, we need to come up with realistic strategies that people could use that would lead to better outcomes (prescriptive rationality.) For
example, some studies have found that simply asking people to consider-the-opposite - to consider hypotheticals such as how they would interpret evidence if they believed
the opposite of what they currently believe, say - can minimise supposed confirmation biases (Lord et al., 1984). Mckenzie (2004) uses Monte Carlo simulations to investigate
the accuracy of several intuitive strategies for inference, and finds that some of them perform almost as well as the normative prescriptions of Bayes’ rule, even though they
are simpler, more intuitive strategies. In particular, the ‘relative likelihood average’ strategy, which involves ignoring base rates and simply estimating the relative likelihood
of new data under alternative hypotheses, and then averaging this with the base rate of the focal hypothesis, correlates almost perfectly with Bayes’ rule (of course, this is still
a fairly complex strategy to actually implement and so perhaps not prescriptive - but it does demonstrate that following strict normative standards is not necessary to get very
close to ’optimal’ performance.)
5Note here though, that acknowledging how evolutionary pressures have influenced reasoning is very
different from saying that this makes behaviourrational - this requires the further step that we define rationality relative to evolutionary goals.
4.4.4 Why does this matter? Rationality and improving human rea-