While for many years economic judgement and decision-making research has assumed that deviations from rationality are purely accidental (see Rational Choice Theory) or occur due to lack of information (see Herbert Simon’s Bounded Rationality Theory), it was one of the two central contributions of dual-processing literature to show that a large part of human errors and mistakes are by no means accidental, but that these deviations from rationality are systematic. And secondly, that these systems can be explained by the duality of human decision-making (duality of system 1 and system 2).
In the next step, I examine the interaction of the different systems and how this interaction can result in biased decision-making, compared to situations where people behave rational. In particular, I discuss what happens if these two types of systems result in different outcomes, and how such a conflict can be solved - and on the other hand, what happens if it is not solved. At present, one may assume that the system conflicts and the resulting biases are relatively ro- bust, and that they are mostly independent of the cognitive abilities (Stanovich & West, 2000).
For instance, Daniel Kahneman reports that he was able to reproduce his experiments on representativeness heuristics (Tversky & Kahneman, 1971) at a conference of the Mathemati- cal Psychology Society and the American Psychological Association (Kahneman & Frederick, 2002). Both are conferences where one might intuit that many humans with good mathemati- cal abilities are present, and that they should not fall for these simple decision traps. Although calculations were so simple that they could have been calculated without problems on the questionnaires, numerous participants relied on intuitive, but false guesses to estimate proba- bilities (Kahneman & Frederick, 2002). This initially unintuitive finding led Daniel Kahneman and Amon Tversky to a series of follow-up studies, which inspired further researchers to get to the study these biases (see section 2.1.2).
Since the first studies of heuristics and biases, research in the 1970’s, numerous biases have been identified, and described. For instance, a review of cognitive biases in decision sup- port systems identified 37 potential relevant biases for information system design (Arnott, 2006). Another more general review identified up to 76 cognitive biases in the judgement and decision-making literature (Carter, Kaufmann, & Michel, 2007). Although some of the iden- tified biases may overlap conceptually, it still shows the susceptibility of the human brain to rely on systematic decision errors.
A recent, crucial contribution to this research stream was made by Richard Thaler and Cass Sunstein. They gathered and reviewed these findings of heuristics and biases research, and they developed a theoretical framework to use these anomalies of human decision-making to improve our health, wealth, and even happiness (2008). This theoretical framework termed Choice Architecture, is presented in the last part of the following chapter, regarding the fact that biases can be reduced or debiased by decision support system design based on choice architecture (see e.g. Arnott (2006)).
Considering the findings of Section 2.1, it is generally accepted that the interaction of auto- matic system 1 processes and controlled system 2 processes constitute our intentions. These two kinds of processes can work together or against each other, and hence can result in differ- ent responses (for an overview, see Gawronski and Creighton (2013), B. K. Payne and Bishara (2009), B. K. Payne (2008)). Consequently, the processes of system 1 and system 2 can be con- vergent, pushing the decision-making towards the same behavioural response. However, they can also be divergent, meaning that they produce different responses. This kind of relation- ship of divergent and convergent processes, has also been called compatible and incompatible (Gawronski, Deutsch, LeBel, & Peters, 2008), or conflicting and aligned (Alós-Ferrer et al., 2016) in judgement and decision-making research.
It is assumed that errors or suboptimal decision anomalies occur, because one of the two sys- tems produces a dominant behavioural response that is erroneous. In most cases of suboptimal decision-making, people rely on the response of our automatic System 1, even when our con- trolled system 2 indicates the opposite (Thaler & Sunstein, 2008, p.21). In other words, it might be that system 2 produces the correct response, but the response of the other system is too strong or too fast, and people rely on the wrong behavioural response. A typical situation
would be that decision-maker intuitively decides to eat delicious cake as dessert, but at the checkout realizes that the decision was wrong (e.g., because system 2 recalling diet plans). A possible method to model the interaction of the two types of cognitive processes and to un- derstand erroneous decision-making is the control-dominating process dissociation approach (Lindsay & Jacoby, 1994; Jacoby, 1991). This approach relies on existing dual-processing theo- ries, assuming that the interaction of the two systems can be modelled and visualized explicitly. In the first version (Jacoby, 1991), system 2 processes are assumed to be dominant and to easily override the outcome of system 1. If there is no response of system 2 or no strong response (e.g., if system 2 failed to compute a correct response), System 1 gives a single response. To investigate this interaction of the two systems, the control-dominating process dissociation model recommends the usage of processing trees (Gawronski & Creighton, 2013). Processing trees (see Figure 11) are a theoretical expansion of dual-processing theory. They are "tools (a) for measuring the cognitive processes that underlie human behaviour in various tasks and (b) for testing the psychological assumptions on which these models are based" (Erdfelder et al., 2009). Consequently, they give a visual representation of the latent dynamic interaction of the two processes, and illustrate how the different responses result in a behavioural response (see Erdfelder et al. (2009), or Batchelder and Riefer (1999) for a review).
Figure 11: Process dissociation model of System 1 and System 2 processing with dominant System 2, based on Erdfelder et al. (2009); Conrey et al. (2005); Gawronski and Creighton (2013).
Process trees built on the assumption that the behavioural outcome of decision-making, mea- sured by different frequencies of a nominal outcome variable (for instance yes and no) follow a multinomial distribution (Erdfelder et al., 2009). Following this rationale, different probabil- ities to a correct response can be calculated and the influence of the latent cognitive process can be estimated.
An established alternative tool to processing trees are generalized linear models (Erdfelder et al., 2009), with a dummy variable for the decision conflict. These models are also able to ac- commodate the specific characteristics of dual-processing theory, and provide a more general perspective on modelling and investigating cognitive processes. By modelling cognitive pro- cesses with a generalized linear model, different direct and indirect measures are regressed on
the independent variable.
Response = V ariablei+ DummyConf lict+ V ariablei× DummyConf lict (2.2.1)
In the model the dependent variable is response and as described by an independent vari- able termed variablei, and a dummy variable dummyConf lict, which represents the conflict of
system 1 and system 2. This approach differentiates between the influence of the variablei
regardless of the decision situation, or the response of the two systems. However, by consid- ering the dummyConf lict, the influence of the variable in case of process conflict (e.g. an in-
tuitive reinforcement process and a deliberative Bayesian Updating process) can be measured (C: cognitive processing vs. A: automated processing). Consequently, variableidescribes the
influence of the variable itself, dummyConf lict the influence of the processing conflict, and
variablei × dummyConf lictthe influence of the intuitive processing on the decision outcome
response.
Studies in dual-processing research rely on this approach to model competitive system 1 and system 2 processing. For instance, Alós-Ferrer et al. (2016) and Charness and Levin (2005) rely on it to model inertia in decision-making, and conflicting heuristic-intuitive and deliberative processing. (Achtziger & Alós-Ferrer, 2013) follow this approach to model decision-making under uncertainty.
Following this generalized linear model approach, the latent variable is measured by induc- ing process divergence and convergence by different tasks, and regressing on the decision outcome. Through investigation of the interaction effect of the variable and the divergence dummy the drivers of the latent cognitive process can be investigated. The discussed ap- proaches rely on different measures of system 1 and system 2 processing. This is necessary, to validate the conceptualization and measure the conflict of the two systems.
In judgement and decision-making research, various other direct and indirect measures have been proposed to induce and investigate possible correlates of system 1 and system 2 process- ing. A first approach is to make usage of the relationship of System 1 and system 2. Following, Generalized Dual-Processing Theory, intuitive system 1 processing is effortless compared to deliberative system 2 processing (see section 2.1). As a result, system 1 processing is much faster than system 2 processing. Various studies have measured the active systems by com- paring the response time in miliseconds (see e.g. Achtziger and Alós-Ferrer (2013)).
Another common approach is to work with bio-physiological correlates. Yerkes and Dodson (1908) have reported that deliberative decision-making works best with an intermediate level of arousal, while low arousal or high arousal reduces performance in complex cognitive tasks. This relationship is also known as the Yerkes-Dodson-Law, as illustrated in Figure 12.
Today, it is generally assumed that arousal significantly influences the proper functioning of system 2 (Kahneman, 2011; Strack & Deutsch, 2004). Dual-processing literature assumes (in analogy with the Yerkes-Dodson Law), that System 2, which is responsible for solving
Figure 12: The performance of the controlled system 2 depends on an individual’s arousal level, induced by the stimulus. Extreme arousal level are associated with poor processing, while intermediate arousal is associated with optimal System 2 processing, hence better per- formance in complex decision-making tasks.
complex tasks, works best for an intermediate level of arousal. High arousal is associated with low performance in experimental tasks requiring deliberative processing (Baron, 2000), while low arousal has been also explicitly linked to low performance of System 2 (Baumeister & Heatherton, 1996).