The effect of automatic processes versus controlled processes on prosocial behavior

(1)

The Effect of Automatic Processes versus

Controlled Processes

on

Prosocial Behavior

Student: Catherine Endtz Student Number: 10469346

Thesis Supervisor: Prof. dr. Frans van Winden Program: Msc. Economics

(2)

0. INTRODUCTION

Prosocial behavior is an essential part of human life and human society. Because we care, share and cooperate we have been able to develop human civilization to the extent that we have. Prosocial behavior is widespread. The U.S.A alone

donate hundreds of billions U.S. Dollars to charity on a yearly basis (List, 2011). Prosociality is also nearly universal, the majority of cultures practice

pro-sociality in certain contexts (Henrich et al., 2005). Nonetheless, it remains somewhat of a mystery: no other species is know to cooperate with each other on such a large scale as we do, with people completely unrelated to them (Warneken & Tomasello, 2009).

Where does that urge to act prosocial come from? Is it really typically human? Is it essentially human? Is it something we are born with? Is it something we acquire trough cultural transmission? Is it the product of our human capacity to hypothesize and deliberate? Kant (1998) once stated that true morality, and related prosociality, was the result of pure rational thought. On the other hand, Frans de Waal (Waal, 2009) claims that animals have basic prosocial instincts and that our delicate spectrum of prosocial behavior has evolved out of those basic instincts, and is still heavily instinctive. Opinions differ as to if prosocial behavior is instinctive, automatic and innate, or if it is the product of controlled rational thought and requires the inhibition of urges.

Recently, this topic has attracted the attention of several behavioral economists. The last couple of years multiple experiments have been carried out, using different manipulations and different games, to shed light on the

(brain)processes underlying prosocial behavior.

This thesis will create an overview of this collection of experiments, and use the aggregation of results to try to describe the effect of automatic processes versus controlled processes on prosocial behavior. By answering this question we might gain insight into the social essence of mankind: is prosociality an inevitable and defining part of human existence, or is it simply an accidental phenomenon.

(4)

The first chapter will describe mechanisms of prosociality within behavioral economics. The second chapter will clarify the concepts of automaticity and control. The third chapter will provide an overview of behavioral economic experiments on the topic of prosociality and automaticity. Finally, the last chapter will contain a general discussion of the experimental results.

(5)

1. PROSOCIAL BEHAVIOR

The last couple of decades, guided by results from economic experiments, behavioral economists have identified different mechanisms for prosocial behavior. This chapter will start with a short overview of the different

experimental games used to research pro-social behavior and continue with a discussion of prosocial mechanisms. Important to note is that this chapter will focus on the proximate, psychological mechanisms of prosociality, and not on mechanisms that explain the ultimate mechanisms of the evolution of

cooperative behavior.

GAMES

The dictator game (Forsythe, Horowitz, Savin, & Sefton, 1994; Kahneman, Knetsch, & Thaler, 1986) is most probably the simplest of all economic games, if it can be called a game at all. In the dictator game one player, the dictator, can decide on the allocation of a certain amount of money between himself and his opponent, the recipient. The recipient in turn has no choice but to accept the offer. Rational choice theory dictates that the dictator would allocate the complete amount of money to himself. In reality the average dictator transfers 28.3% of the total amount to the recipient and only 36% of dictators do not share at all (Engel, 2011). Because the dictator is the only actor in the game, the proposed allocations cannot be influenced by expectations of the co-players behavior and the dictator game could therefore be said to measure altruism in it's purest form.

The ultimatum game (Güth, Schmittberger, & Schwarze, 1982) is similar to the dictator game. The only difference is that in an ultimatum game the recipient can decide to reject the proposer's offer. In a typical ultimatum game, proposers offer 30 to 40 percent of the total amount, and offers of less than 20 percent are

rejected about half of the time (Oosterbeek, Sloof, & van de Kuilen, 2004). Rejection of any positive amount of money by the recipient is always irrational. Non-zero offers by proposers on the other hand could be explained by rational egoism when considering expectations of non-rational recipient behavior. The

(6)

ultimatum game is thought to show the prevalence of negative reciprocity or revenge.

In the trust game (Berg, Dickhaut, & McCabe, 1995) an investor receives a certain amount, which he can divide between himself and a trustee. The money allocated to the trustee gets tripled and should then be divided by the trustee between herself and the investor. A metastudy showed that investors on average transfer 50% of the money and trustees on average return 37%. The trust game provides proof for the existence of positive reciprocity. There is no reason for the trustee to return any money in a one-shot situation but to reward the trust of the investor.

The prisoners' dilemma is a classic game that requires simultaneous action by two players. Payoffs of the dilemma are depicted in Table 1.

Table 1.1 Prisoners’ dilemma

Assumption T>H>L>S

Cooperate Defect

Cooperate H,H S,T

Defect T,S L,L

The prisoners’ dilemma is thought to mimic many real-life situations in which individual best responses do not lead to socially efficient outcomes. While the (rational) Nash-equilibrium of the game would be attained if both players defect, the most efficient outcome would be attained if both players cooperate. A

prisoners’ dilemma is especially interesting when carried out repeatedly, as that shows us the individual benefits of cooperating as a tactical mechanism to elicit direct reciprocity. However, even in its one-shot form roughly half of players typically cooperate (Camerer & Fehr, 2002)

(7)

The very popular multi-player version of the prisoners’ dilemma is the public

goods game. In a public goods game, all N players are endowed with a number of

tokens. Players can choose to contribute a certain amount of their endowment to the public account. The total amount of tokens in the public account then gets multiplied by a certain amount smaller than N, and will then be divided between all players. A rational selfish player will choose to keep all his money to himself, while the social optimum would be attained if everyone would contribute their complete endowment. Similar to the prisoners’ dilemma, in a one-shot game on average 50% of the endowment is contributed (Ledyard, 1995). However, in repeated games contributions unravel over time, showing the vulnerability of group cooperation to freeriders.

MECHANISMS FOR PRO-SOCIAL BEHAVIOR

ENLIGHTENED EGOISM

It's hard to explain why a dictator would give away half of its endowment to an anonymous opponent in a one-shot laboratory experiment based on nothing but selfish preferences. In real life, however, interactions are never completely isolated. They often take place between people that have a particular relation with each other, they're often observed by bystanders and chances are you might meet your interaction partner again. This creates opportunities for prosocial behavior that is profitable for the agents performing it.

Prosocial behavior can be carried out with the aim of receiving something in return that outweighs the initial investment. This return could consist of

material payoff, but also of a social reward such as praise, status, reputation or a social tie. Additionally, this return could come both directly from the receiver as from third party bystanders or informed community members (Kolm, 2006). Prosocial behavior is therefore rational from a selfish perspective if one expects other's to have a truly reciprocal nature, even if that belief might be false

The latter point was proven by Kreps Milgrom, Roberts & Wilson (1982) through the modeling of prisoners’ dilemma games. In a repeated prisoners’ dilemma it

(8)

might seem rational for a selfish players to cooperate, hoping that his opponent will reciprocate his cooperation. When both players use this tactic a pattern of repeated cooperation would develop, were it not for the rules of backward induction. In the last round, neither player has a reason to uphold cooperation, as the opponent cannot retaliate. Knowing that a rational interaction partner will defect in the last round however, there is no reason to cooperate in the before-last round and so on. Backward induction leads to the unraveling of cooperation. Modeling prisoners’ dilemmas Kreps et al. (1982) showed, however, that this does not always need be the case. When a rational player believes that there is a chance that his opponent might be a reciprocator it can be rational and payoff maximizing to cooperate up till a certain round (depending on the weight assigned to the belief). This mechanism holds even if the belief is completely unfounded and both players are completely selfish in reality.

Alternatively, repeated mutual cooperation is rational when a prisoners’

dilemma is infinitely repeated or when during each round there is a large enough chance that there will be another encounter. The latter matches some real life situations. Often when we meet people that are part of the same social circle that we are part of, there's always a chance that we'll meet them again, although it's uncertain when and how often. Real life therefore appears to create many situations in which prosocial behavior is in the long-term interest of selfish people.

In short, prosocial behavior can be rational in finite repeated interactions when some part of the population consists of reciprocators, when rational players believe that a part of the population consists of reciprocators, or alternatively, when exerted towards people of whom you think there will always be a chance you'll meet them again.

OUTCOME-BASED OTHER-REGARDING PREFERENCES

Selfish preferences might sometimes explain prosocial behavior in repeated interactions. However, to explain prosocial acts in one-shot laboratory games other mechanisms are needed. More specifically, to explain pro-social behavior

(9)

in one-shot laboratory games, we need to loosen the assumption of rational egoism.

The last decennia, a number of economists have formulated a group of

explanations for prosocial behavior that kept the assumption of rationality while challenging pure selfishness. These are the theories of prosocial preferences. Utility theory states that decision-makers have preferences over the allocation of goods and that those preferences are rational (von Neumann & Morgenstern, 2007) Very often utility theory is interpreted as claiming that utility can only be derived from allocations of goods to oneself. Theories of prosocial preferences, however, state that derived utility can also depend on allocations to other people.

The most straightforward pro-social preference is pure altruism. Pure altruism theories state that other people's well-being has a positive effect on an

individual's own utility (Fehr & Schmidt, 2006). Altruism could provide an explanation for dictator game giving (Eckel & Grossman, 1996). It could also explain real world behavior such as donation to charities, volunteer work and workplace behavior (Rotemberg, 1994). The challenge with pure altruism is that the utility derived from others' wellbeing is independent of the source of

wellbeing. In other words, the more other people give, the less the altruist would be inclined to give. Real world examples, such as donations to very large and popular charities, seem to challenge this view.

As an alternative, Andreoni (1990) has formulated the theory of warm-glow

altruism, suggesting that people derive utility from their own pro-social behavior

in itself. Therefore, personal good-doing would not be crowded-out by external good-doing.

Altruism can by no means explain all other-regarding behavior. Sometimes, people are even willing to incur costs to worsen someone else’s situation.

Inequity aversion provides an alternative. Players that are inequity averse receive

(10)

players. They behave altruistic when another person has less than themselves but act spiteful when another person has more. Fehr & Schmidt's (1999) version of the model distinguishes between advantageous and disadvantageous

inequality. Fehr & Schmidt allow for a heterogeneous population with differing parameters for advantageous and disadvantageous inequality. However, they do state that for everyone the parameter for advantageous inequality is lower that that for disadvantageous. In other words, people always prefer having more than someone else over having less than someone else. An alternative version by Bolton & Ockenfels (2000) assumes that people dislike deviations from the mean. For two-player games predictions are very similar to Fehr & Schmidt, but for multiplayer games they differ, as individual payoff distributions are not taken into account with Bolton & Ockenfels. Inequity aversion can explain giving in dictator games as well as the rejection of offers in ultimatum games and the punishment of freeriders in public goods games.

Inequity aversion implies that people will act to increase equality, even if that entails reducing total output (efficiency). Experiments have shown, however, that people are willing increase the relative income of others and thereby

increase inequality in situations where helping others also increases efficiency (J. Andreoni & Vesterlund, 2001; Charness & Rabin, 2002). Andreoni and

Vesterlund carried out an adapted dictator game in which the dictator under some conditions had to forgo more or less than one Dollar to add a Dollar to their opponents payoff (as opposed to the usual 1:1 ratio) They found that 44% of the players acted completely selfish, 35% showed equitable preferences and 21% acted to maximize total efficiency.

STRONG RECIPROCITY

Unlike what social preferences theories seem to suggest, final pay-off divisions are not the only things that matter when making social decisions. One of the other things that can influence our decisions are the intentions of others people. Theories of strong reciprocity do take intentions into account. Strong direct reciprocity takes place when we act kindly towards those that have been kind to us, and unkind to those that have been unkind to us. With strong reciprocity,

(11)

rewarding or punishing intentions is a goal in itself. In this it separates itself from earlier discussed forms of reciprocity that have eventual self-enrichment as a goal.

One of the best known approaches to modeling reciprocity is the intention-based model of Matthew Rabin (1993). In Rabin's model decisions in a two player game depend on the expected kindness of the opponent. Kindness is defined as the difference between the planned payoff and the average possible payoff, and expected kindness is met by kindness. Rabin's model makes use of psychological game theory. An alternative is the type-based model of Levine (1998). According to Levine altruistic players act more altruistic towards opponents that are also of an altruistisch type. As the model of Levine makes use of expected types, it can be modeled by conventional game theory. In both the model of Rabin, as that of Levine, reciprocity is not limited to reactions to prosocial acts but can be triggered by expectations alone.

Strong reciprocity can be positive, such as seen in a one-shot trust game for example. Much more intensely studied however, are instances of negative reciprocity or punishment. Rejecting positive offers in an ultimatum game, for example, could be explained by referring to strong reciprocity: people want to punish a proposer that offers them too little. Widely studied too is punishment behavior in a public goods game. If a standard public goods game is extended with the option to punish co-players against some personal cost, a large number of players decide to punish freeriders, a mechanism that can prevent the collapse of cooperation on the long run. Punishment behavior is even observed in one-shot games, be it to a lesser extent, thereby excluding weak reciprocity and strategic prosociality as an explanation (Fehr & Gächter, 2000).

An interesting addition to strong reciprocity is the insight that we do not only reciprocate observed behavior, but also expected behavior. Fischbacher and Gächter (2010) found that actual contributions in a public goods game correlated positively with expected average contributions. While initially, it seems to be a possibility that people use their own planned contributions as an anchor to

(12)

estimate other's contributions, Fischbacher and Gächter show that people do in fact base their own contributions on the expected offers of others.

Finally, strong reciprocity seems to extent to third-party situations. Not only do we reciprocate kind behavior towards us with kindness and unkind behavior towards us with unkindness, we also seem to be willing to reward those people that are kind to others and punish those that are unkind. Fehr and Fischbacher (2004), for example, carried out an experiment based on a dictator game, but added a third party that could observe the decision of the dictator and punish the dictator against some cost. A significant percentage of third-party onlookers did indeed decide to punish the dictator if he chose an allocation that was perceived as unkind (zero or almost zero).

HETEROGENEITY

Economists like to make predictions on the aggregate level, and therefore economic models are usually based on a type of universal behavior. However, in reality people are not all the same, do not all strive for the same and do certainly not all act the same. Pretending that we do might keep us from gaining essential insights into human behavior that could lead to more accurate predictions.

To capture differences in social preferences between people the concept of social value orientation (SVO) has been developed (Liebrand & McClintock, 1988; Van Lange, De Bruin, Otten, & Joireman, 1997). The different measures of social value orientation measure people's preferences for patterns of outcome distributions between oneself and another person. The SVO measures consist of

questionnaires with a collection of hypothetical mini-dictator games, without a constant-sum restriction. As the SVO measures have their roots in the

psychological sciences they are typically not incentivized, however, some economists (e.a. Offerman, Sonnemans, & Schram, 1996; E. van Dijk, De Cremer, & Handgraaf, 2004). adapted the SVO measures to function in incentivized experiments. Based on the SVO measure, van Lange distinguishes three different types: pro-socials, that act to maximize efficiency or maximize equality;

(13)

individualists, that do not care about the payoff of others; and competitives, that strive to maximize their payoff relative to the others.

As described earlier, people do not only base their choices on preferences over expected outcomes. People care about other things, such as intentions. It makes sense to expect that there is also a certain amount of heterogeneity in the way people reciprocate others' intentions. Fischbacher, Gächter and Fehr (2001) found that, in a public goods game, three different types of reciprocators could be distinguished: conditional cooperators, hump-shaped cooperators and free riders. Conditional cooperators contributed if they expected others to contribute, and typically contributed slightly less than they expected others to contribute; hump-shaped contributors contributed the same as others for low average contributions but turned into free riders for high contributions. Fischbacher et al. based this conclusion on a variant of what is called the 'strategy method'. After playing a normal PGG (what they labelled the unconditional contribution) they also asked subjects how much they would contribute for each possible given average contribution of the other players (the conditional contribution). At the end of the experiment, payoffs for a four-person group were decided using the unconditional contributions of three players and the conditional contribution for one randomly chosen group member. In their experiment they found that 50% of the subjects acted as conditional cooperators, 14% showed a hump-shaped pattern, and 30% were free riders. The fact that the majority of cooperators are imperfect cooperators that reciprocate slightly less than they get means that, even in the absence of free riders, a multi-round PGG would collapse in the long run in the absence of additional enforcement mechanisms or incentives.

Finally, van Dijk & van Winden (1997) showed the importance of heterogeneity of specific as opposed to general prosociality. Besides a general tendency to be more or less altruistic, people have a tendency to be more or less altruistic towards specific people. A history of positive (negative) interaction can result in the formation of a positive (negative) social tie between two people. This

positive social tie will in its turn increase the likelihood of positive future interactions and people will now acquire personal utility from helping the

(14)

person they're socially tied to. A short inspection of social phenomena such as friendship and family bonds shows the self-evidence of this theory.

People are heterogeneous. It is highly likely that the different social preference models described in this chapter fit the behavior of some people better than others. It is highly likely that the world population will contain some purely selfish people, and some people with purely altruistic, reciprocal, or efficiency-maximizing preferences. The social value orientation measure captures some of these differences, but not necessarily all of them. Finally, social preferences might not be general. It is highly probably that our preferences differ depending on the person we're interacting with. This strong heterogeneity might provide a challenge for economists' quest to formulate predicting theories on the

aggregate level. Nonetheless, awareness of this heterogeneity will provide useful when interpreting experimental result later in this theses.

CONCLUSION

Over the last couple of decennia, behavioral economists have reintroduced the idea that man can sometimes act to favor others into the current economic discourse. Economic games such as the dictator game, the trust game and the public goods game have all shown that at least part of the population does not always act selfishly. While strategic behavior can explain some of these results, prosocial behavior in one-shot games can only be explained by referring to other-regarding preferences or strong reciprocity. Experimental results

furthermore suggest that the tendency to behave prosocial is not universal, and that people can be assigned different prosociality-profiles. Further research should clarify in what kind of situations which kind of prosocial behavior is dominant for which kinds of people. This can be done by enriching the

experimental environment as well as by unraveling the (neuro-)psychological processes that lie at the basis of prosociality.

(15)

2. AUTOMATICITY and CONTROL

Standard economic theory assumes that man is a calculating animal. It assumes that when making choices, we carefully weigh the different utilities of future outcomes of our current behavior, and actively choose the option that maximizes final utility. Constantly computing and comparing possible outcomes would require massive brain capacity and therefore doesn't seem to be very realistic. In reality, we do not constantly consciously deliberate about every detail of life. Much of our actions are carried out automatically, outside of our conscious control. This provides a challenge for economists. While it is theoretically possible that our automatic responses are unconsciously guided by extensive calculations suggested by preference functions, it is far more likely they are guided by instincts, emotions and habits. This distinction between automatic and controlled behavior has fascinated thinkers of all times. Well-known is Freud with his theory of the conscious ego that stands opposite the unconscious id (Lear, 2005). Amongst current psychologists, the automatic-controlled

distinction has given rise to numerous dual-process theories of the brain stating that there are two categories of brain processes: one automatic, fast and

unconscious and another controlled, slow and conscious. Camerer, Loewenstein and Prelec (2005) take dual-processes to another level by incorporating two dualities in their framework: both the automatic-controlled distinction as well as a distinction between affective and cognitive processes. The coming chapter will explore the concepts of automatism and control in the brain, taking Camerer, Loewenstein and Prelec's two-dimensional framework as a starting point.

AUTOMATIC-CONTROLLED

Given the purpose of this paper, the automatic-controlled distinction is the most interesting of the two dimensions of the brain proposed by Camerer et al.

Schneider and Shiffrin (1997) were the first to introduce the distinction between automatic and controlled brain processes. Since then, many have proposed their own so-called dual-process theories, be it under different names and each with a slightly different focus. Examples are intuitive and analytic (Peters, Hammond, & Summers, 1974) experiential and rational (Epstein, 1994) reflexive and

(16)

reflective (Lieberman, 2007) and System I and System II (Stanovich & West, 2000). While the names often indicate what is regarded as the leading

characteristic by the authors, typically many different characteristics were used to describe the differences between the processes.

Automatic processes are generally thought to be evolutionary old and shared with animals. They are fast, parallel, effortless and not available to

consciousness. Only their final products might be consciously experienced. Automatic processes contain habits, instincts but also basic bodily functions such as the beating of our heart. They are the default mode of the brain and constitute most of the electro-chemical activity in the brain. (Camerer et al., 2005)

Controlled processes on the other hand, are believed to have evolved quite recently. They are conscious, slow and sequential and create the possibility of abstract hypothetical thinking. Because of their sequential character, controlled processes are very fragile. One mistake or blockage and all controlled processes are misdirected or slowed down. Because controlled processes are conscious, people can usually provide a good account of controlled processes through introspection. (Camerer et al., 2005) Because of this, the controlled processes are closely related to our sense of self. We like to think of ourselves as being actively and consciously in control of our lives.

The boundary between controlled and automatic processes is less clear-cut than it might seem at first sight. As Bargh (1989) explained: by definition, controlled processes are processes of which we are aware, that cost effort, that are started intentionally and that can be and have to be controlled until completion.

Automatic processes however, are processes that lack any of those qualities, but not necessarily all of them. Automatic processes could be uncontrollable but conscious, such as some intense emotional experiences. They could be started involuntarily, but open to controlled interference, such as a hand movement in the direction of the cookie jar caused by an urge to munch. Or they could be initiated voluntarily but continue autonomously, such as driving a car. In other words, what we usually refer to as automatic processes consists of a myriad of

(17)

different processes that have in common that they are not fully controlled. In reality many processes that are described as automatic actually involve interplay between automatic and controlled systems in the brain.

They most widely held view about interactions between the two systems is the default-interventionist model (Camerer et al., 2005; J. Evans, 2007; Kahneman, 2003). Default-interventionist models assume that the automatic system always generates fast default responses and that the controlled system may or may not interfere. They state that controlled processes only interrupt the automatic processes at special moments, for example in case of unexpected events, novel decisions or experienced strong visceral states (Camerer et al., 2005).

Nonetheless, interaction between automatic and controlled processes is not solely one-directional. Interference of automatic processes by the controlled system might in turn cause other unconscious automatic processes. To make this distinction clear, Bargh and colleagues (Bargh, Schwader, Hailey, Dyer, &

Boothby, 2012) distinguish preconscious and postconscious automaticity.

Lieberman (2007) has attempted to locate the two different categories of processes in the brain. He distinguishes the reflexive system X (automatic) and the reflective system C (controlled). Globally, it could be said that the automatic processes are concentrated in the back (occipital lobe), top (parietal lobe) and side (temporal lobe) parts of the brain, while the controlled processes are centered around the front (frontal lobe). More specifically, important X-system area's are the amygdala, known for its role in negative affective reactions; the basal ganglia, known for its role in implicit learning of abstract sequential patterns and linked to positive affective responses to stimuli; the ventromedial prefrontal cortex (vmPFC), related to implicit learning and framing effects; the lateral temporal cortex (LTC) associated with implicit semantic processing and recognition of facial expressions; and the dorsal anterior cingulate cortex (dACC), involved in automatic conflict-detection. The C-system consists of the rostral anterior cingulate cortex (rACC), involved in symbolic expectancy violations; the lateral posterior parietal cortex (lPPC), supporting working memory, logic, and self-focused attention; the hippocampus and the surrounding

(18)

medial temporal lobe (MTL) region, involved in episodic memory; and the lateral prefrontal cortex (lPFC), involved in many higher cognitive processes such as working memory, inhibition and implementation of top-down goals. The

prefrontal cortex (PFC), sometimes called 'the executive region', could be said to form the core of the controlled system. It collects input from a multitude of other regions and integrates them to plan actions.

AFFECTIVE-COGNITIVE

The second dimension proposed by Camerer et al. is the affective-cognitive dimension. Many dual-process theorists group affect and its cousin emotion with the automatic processes and cognition with controlled processes, either

indirectly or explicitly. An explicit example is the affective-deliberate dual-process model by Loewenstein & O'Donoghue (2004). Camerer et al. claim that while this simplified merging of dimensions might be useful for some purposes, it is important to realize that in reality the two dimensions do not perfectly correlate. Affect is not always automatic. Although rare, it can be controlled, for example by a method actor. Additionally a vast number of automatic processes have nothing to do with affect or emotion at all, but are cognitive in kind, such as moving your hand to catch a ball.

Essential to the distinction between affect and cognition is the concept of valence. Affect involves valence; affect can be positive or negative. Affect motivates us to (not) do something, and often carries action-tendencies.

Cognition is value-free and only involves truth and falsehood. As such, controlled processes, often thought to be purely cognitive, need input from affect to

determine the final goals of their planning activities. (Camerer et al.,2005)

Affect is often equated with emotions or feeling states, but, though related, they are certainly not the same. Most affect operates unconsciously and unnoticed. Emotion constitutes those affects that do reach our conscious awareness and that are therefore available through introspection. Affect embodies not only what we call emotions such as fear, anger, joy but also drive states such as

(19)

hunger, thirst and sexual desire and motivational states such as pain (Camerer et al., 2005)

The brain area that is involved with affect has historically been called the limbic system. The limbic system was generally thought to consist of parts of the brain that were evolutionary older. It takes up large parts of the mammelian brain: the area between the brainstem and the neocortex. The limbic system shows partial overlap with the X-system as defined by Lieberman (2007), which is an

indication of the partial overlap of affective and automatic processes. Brain area's typically associated with the limbic system are the hippocampus, the hypothalamus, the amygdala, the cingulate gyrus and the nucleus accumbens. However, recently multiple researchers have expressed their doubts about the existence of a unified limbic system. Many classic limbic system areas have been shown to be strongly involved with cognitive processes, such as the

hippocampus that plays a large role in memory. (Baars & Gage, 2010)

LeDoux (2000) suggested that instead of focusing on the imprecise concept of the limbic system in it's totality, it might be better to look at separate affects and their neural pathways. LeDoux, following his own advice, carried out an

intensive study of fear mechanisms in the brain that might provide insight for the functioning of emotions in general. He discovered that the amygdala, aided by the hypothalamus, is essential in regulating fear reactions, causing an increase in blood pressure and hart rate, and freezing or flight-preparation. There are two pathways towards the amygdala: one, the low road, directly from the thalamus, where sensory inputs enter the brain; and one, the high road, from the thalamus through the sensory cortex to the amygdala. The low road provides a rapid and crude representation, while the high road is more precise but also slower. This allows for people to, for example, freeze immediately when they see something snakelike (low road) but to continue what they were doing once they realize the alleged snake is only a curled stick (high road). At first glance, this two-road distinction shows similarities with the distinction between the fast automatic system and the slow controlled system. However, there is one essential difference: both the high and the low road are most likely processed

(20)

unconsciously and therefore outside of our control. For consciousness, and control, involvement of the prefrontal cortex would be necessary. (LeDoux & Phelps, 2008)

SOURCES OF AUTOMATIC BEHAVIOR

To understand automatic behavior, and especially to understand automatic social behavior, it is essential to understand its origins. Is all automatic behavior genetically programmed, inescapably linked to what makes us human? Or can automatic behavior be learned, changed and manipulated? As often, the answer lies in the middle. Three broad categories of automatic behavior can be

distinguished: processes developed through conscious skill-acquisition, implicitly learned processes, and innate processes

The most commonly accepted form of automatic behavior is behavior learned through conscious skill acquisition (Bargh et al., 2012). When we, consciously and deliberately, repeatedly respond to the same external cues in the same manner, over time the behavior is internalized and automated. Take driving a car. During the first couple of driving lessons the driving student consciously learns how en when to change gears. After years of car driving, however, the use of gears is completely automated and many drivers don't consciously register the right moment to change gears, they simply do it.

Alternatively, automatic behavior can be the result of unconscious development of behavioral patterns. This is especially apparent in early childhood, when many abstract social as well as cognitive concepts and processes that influence adult life are learned without any involvement of consciousness (Bargh et al., 2012). A toddler, for example, doesn't learn grammar rules through the intensive study of grammar books, a toddler automatically recognizes grammatical patterns in heard speech, without realizing. Language acquisition is an example of implicit learning: the unconscious classification and categorization of our experience that is a source of many of our intuitive responses (Norman & Price, 2012)

The final sources of automatic behavior are our genes; part of our automatic behavior can be said to be innate. This category is both the most interesting for

(21)

the purpose of this thesis as the most complicated. Interesting, because only the existence of innate behavior can tell us something about human essence and its true selfish or prosocial nature. Complicated, because a large percentage of innate behavior is not necessarily the focus of decision-making research, such as the beating of our heart, or our knee-reflex. Some innate automatic behaviors that do touch the realm of decision-making are to be found in automatic emotional and motivational responses that guide our actions, such as for example, disgust, or survival and reproductive urges. Many other behavioral patterns, often emotion-based, have both innate and learned components. (Bargh et al., 2012)

MODELS OF SELF-CONTROL

Two models of control functions have been very influential in psychology in general, but also in the development of multiple experimental manipulations of consciousness and control. The following paragraphs will provide a short overview of the theories of executive control as proposed by Alan Baddeley (1996) and Roy Baumeister (Baumeister & Heatherton, 1996).

In the brain, behavioral control functions have been assigned to an entity referred to as the working memory, first described by Alan Baddeley (1996). Baddeley's working memory is a very broadly defined concept that regulates attentional control, impulse inhibition and action planning, and it is additionally the locus of our mental workspace. It is the place where we can hold limited information in a temporarily accessible state (Cowan et al., 2005), to be used for further conscious manipulation. Baddeley's working memory consists of four main components: the central executive, the phonological loop, the visuo-spatial sketchpad and the episodic buffer.

The central executive is the guiding component of the working memory, and could be called the core of our controlled processes as described by Camerer et al. (2005). It controls cognitive resources and plans and monitors information-processing. It is responsible for selective attention and inhibition. It transfers information from the short-term to the long term memory and back, and it controls information flow from and to the other three 'slave systems'. The

(22)

phonological loop and the visuo-spatial sketchpad are responsible for short-term information retention. The phonological loop is responsible for retention of phonological information by acting as an 'inner voice' and repeating series of words in loops to prevent them from decaying; while the visuo-spatial sketchpad retains visual and spatial information, and is involved in spatial planning. Finally, the episodic buffer is said to merge information from the phonological loop and the visuo-spatial sketchpad into integrated visuo-phonologic time sequence units. Central to the theory of Baddeley is the idea that the different components of working memory all have limited capacity, which means that they can only handle a limited amount of tasks at the same time. (Baddeley, 1996)

An alternative theory explaining capacity for control is the strength-model hat has been put forwards by Roy Baumeister and colleagues (Baumeister,

Bratslavsky, Muraven, & Tice, 1998; Baumeister & Heatherton, 1996;

Baumeister, 2002). They define self-control as the capacity to modify, change or override human impulses, desires or habitual responses. According to

Baumeister, self-control acts like a muscle in the sense that it can get tired after intensive usage. When self-control is tired, or 'depleted', impulses and implicit motivation have no chance of being contained, until strength is regained and resources are replenished. Baumeister later broadens his concept of self-control, stating that it's not only involved with impulse inhibition, but also with making difficult choices, manipulating information and developing implications through logical reasoning (Schmeichel, Vohs, & Baumeister, 2003). In its extended

version, Baumeister's self control shows strong overlap with the functions of Baddeley's central executive. However, essential to Baumeisters theory is that central executive functions are not only limited in their capacity to function simultaneously, but are also impacted when carried out consecutively.

CONCLUSION

In conclusion, we could say that the brain is best understood by a two dimensional framework with an automatic-controlled axis and an

affect-cognition axis. This especially allows for a better understanding of emotions, that are not necessarily limited to the automatic sphere, and that are defined by their role in valuation. This thesis, however, is predominantly focused on the

(23)

distinction between automatic and controlled processes. While separating cognitive from affective automaticity when deemed necessary, in the following chapters the two will often be merged out of practical considerations, mimicking the dual-process frameworks used by the authors discussed. When zooming in on automatic processes, it is useful to note the difference between processes that are innate, processes that are acquired unconsciously, often during early

childhood socialization, and processes that are acquired through conscious repetition and internalization. On the other side of the dual-process divide, controlled processes are usually ascribed to an entity called the 'central executive'. A key characteristic of the central executive is its limited capacity, that limits simultaneous activity (Baddeley) and possibly even limits consecutive activity (Baumeister). This limited capacity of the controlled system as well as its limited processing speed have given rise to multiple experimental manipulations that will be discussed in the next chapter.

(24)

3. AUTOMATICITY, CONTROL and PROSOCIAL BEHAVIOR in

ECONOMIC EXPERIMENTS

The main interest of this thesis is the effect that automatic versus controlled processes have on prosocial behavior. A number of researchers have tried to study this effect using economic games. However, results are varied and

sometimes contradictory. Closer inspection of the experiments might shed light on the variation in results. This chapter will provide an overview of those studies as well as a first reflection on the methods used and results obtained. It will limit itself to economic experiments, and will not go into alternative methods that have been used to investigate this behavior, such as purely psychological experiments and neuro-imaging studies. The main focus will be on the dictator game and the public goods game. They are the experiments most commonly used in studies on the topic of interest. Moreover they measure one-direction prosociality (altruism) and multi-direction prosociality (cooperation) in their purest form, without any interference of strategic considerations, at least in their unrepeated form.

One of the reasons for the variations in experimental results might be the multitude of methods used to study the same thing. Correlations between automaticity and prosociality have been measured using reaction times, while popular ways to measure to actual effect of automaticity vs. control on

prosociality have been time pressure studies, cognitive load manipulations, and ego depletion tasks. In the coming chapter, selected experiments will be

discussed according to method used. The fourth chapter will contain a general discussion, comparing the results of all different methods in order to create a general picture.

(25)

REACTION TIMES

Method

Reaction time measurements are one of the first measures that were used to try and objectively study our brain processes. Before reaction time measurements the only means psychology had to discover what happened 'inside our head' was introspection. A pioneer in the area of reaction times was F.C.Donders (1969) who researched the processing time of certain speech reactions. In the context of dual-process research, reaction time measurements are based on the idea that automatic responses are faster than controlled responses. Automatic processes are more efficient, require less attention and are therefore faster. Consequently, researchers have used reaction times as a direct measure of the amount of

automaticity involved in a decision (see e.a. Bargh & Chartrand, 2014). The use of reaction time measurements in experimental economics is quite novel. However, the last couple of years saw a fast rise of these numbers, also in the area of

prosocial behavior.

Experiments

Piovesan and Wengström (2009) carried out an adapted Dictator game in which participants could choose one out of four different allocations of money between themselves and an opponent. They found that subjects with lower reaction times were generally more selfish than subjects with higher reaction times, suggesting that egoism is automatic. Several years later though, Wengström cooperated with Cappelen et al. (2014) in a large-scale classic dictator game study and found the opposite effect: a positive correlation between reaction time and selfishness, indicating automatic prosociality.

A collection of articles measured time-pressure in public goods games (PGGs), and many of them found that quicker reactions were more pro-social (A. M. Evans, Dillon, & Rand, 2014.; Lotito, Migheli, & Ortona, 2013; Nielsen, Tyran, & Wengström, 2013; Rand et al., 2012). Rand Green & Nowak found a significant negative relation between reaction time and contributions in a one-shot PGG through the online labor market MTurk. Nielsen et al. found in a similar online

(26)

PGG that freeriders take longer to decide than conditional cooperators. Lotito and & Migheli found similar results in a repeated PGG.

Alternatively, Fiedler et al. (2013) found a positive relation between prosociality and reaction times in a PGG while studying reaction times in relation to SVO. The study found a positive effect between the absolute degrees of the SVO angle (where zero degrees equals a purely individualistic SVO) and reactions times, indicating that both prosocial and competitive people take longer to decide. This in turn could be explained by the fact that both competitive and prosocial people are other-regarding and take other peoples output into account before deciding, a process that might be time-consuming.

Likewise, Lohse, Goeschl & Diederich (2014) found a positive relation between prosociality and reaction times using data from an extra-laboratory PGG on the Internet. Their 3483 subjects faced a choice two times between monetary payment and a verifiable reduction in CO2 emission. The reaction times of those contributing to the public good was significantly larger than that of those who didn't. Additionally, people that decided to switch choices the second time took longer if they decided to contribute and chose quicker when they decided to take the money themselves.

The most insightful amongst the reaction time studies is that by Evans et al. Evans et al. find that reaction times in a PGG and a Prisoners’ dilemma follow a U-pattern. Extreme reactions, be it extremely prosocial or extremely selfish, were always quicker than intermediate reactions. The agregate positive correlation that Evans found between automaticity on prosociality was caused by the fact that there were more extremely prosocial reactions than extremely selfish reactions. This suggests that extreme reactions are always more

automatic, while intermediate reactions are often reached through controlled deliberation.

(27)

Table 3.1: results reaction time studies

*when multiple experiments are described within one article, the subject numbers

of the different experiments are mentioned separately

Analyses

Table 3.1 shows an overview of the different experiments described. At first sight results seem to be very contradictory, with a number of studies stating that short reaction times are accompanied by automatic prosociality and a number of studies claiming the opposite. A closer look, however, raises one set of results above the other.

The experiments by Piovesan and Wengström (2009) and Fiedler et al. (2013) have a very small subject pool compared to the other experiments. They used 72 and 38 subjects respectively, while for example Nielsen et al. (2013) and

Cappelen et al. (2014) both used over 1500 subjects. Rubinstein (2007) claimed that measuring response times of economic decisions is meaningless for small samples because of the noisy approximation of mental processes. This makes the results of Piovesan and Fiedler much less credible than those of the other

experiments.

Researchers Game Results Subjects*

Cappelen et al. (2014) Dictator game automatic prosociality 1508

Evans et al. (2014) Public goods game automatic extremity 324, 156, 28, 6252, 303, 503, 864 Fiedler et al. (2013) Public goods game automatic intermediacy 36, 38 Lohse et al. (2014) Public goods game automatic egoism 3483 Lotito et al. (2013) Public goods game automatic prosociality 128 Nielsen et al. (2013) Public goods game automatic prosociality 2081 Piovesan & Wengström

(2009)

Dictator game automatic egoism 72

Rand et al. (2012) Public goods game automatic prosociality 212, 48, 104, 278, 192, 256

(28)

Lohse et al. (2014) found automatic selfish behavior in an experiment using a convincing number of 3483 participants. However, the study strayed from the original public goods experiment by using a public good that would benefit the whole of society instead of the traditional small number of (anonymous) co-subject in the lab. While indeed the experiment might more closely resemble a large number of real world situations, it does differ categorically from all the other experiments. Providing a public good to the benefit of the whole human species - as a reduction of CO2 emission is - likely taps into a different set of habits, social norms, emotions, and goals than the provision of a shared good for a small-number of co-players does. The identifiable victim effect (Jenni &

Loewenstein, 1997) shows that while we're often inclined to help needy people we can identify, we're generally indifferent to the needs of larger vaguely defined groups. Results of Lohse et al., seen in the light of other studies with opposing results, might provide support for this thesis. We might be less inclined to provide a good to the whole of society compared to providing for three anonymous though identifiable co-players.

Despite apparently conflicting results, the odds seem to be favoring Rand et al. (2012), Evans et al. (2014), Lotito & Migheli (2013), Nielsen et al. (2013) and Cappelen et al. (2014), who claim a correlation between quick automatic response and prosociality. Amongst those, the study by Evans et al. (2014) demands special attention.

When looking into the quadratic relationship between prosociality and reaction times, Evans et al. found that all extreme reactions were quick, not only

extremely prosocial but also extremely selfish ones. This explanation provides a possible means of reconciling the various conflicting experimental results. Considering the results of Evans et al. the aggregate relationship found between reaction time and prosociality would depend on the ratio between extremely prosocial and extremely selfish deciders. The mainstream result of quick prosociality could be explained by a larger portion of extreme prosocials in the subject pools, as with the experiment by Evans et al., while alternative results

(29)

such as those of Piovesan and Wengström could indicate a slightly different population composition with people tending more to extreme selfishness. At first sight, the explanation of the results of Evans et al. seems obvious: automatic reactions are always quite extreme, while controlled deliberation often leads to more nuanced responses; additionally, people are heterogeneous in their automatic reactions but on average tend to have automatic reactions that are more prosocial. However, further experiments by Evans et al., discussed in the next section, challenge this interpretation.

Finally, Cappelen et al (2014) correctly remarked that response time does not solely depend on the amount of automatism or control involved. People could be said to go through three stages when making a decision during an experiment. First they try and understand the decision problem, then they make the actual decision, and finally they execute the decision by pressing a button or filling in a form. Only the middle stage is of interest when we study the automaticity of decision-making. However, even people that rely heavily on controlled processes (and possibly even specifically them) could be smarter or faster readers and therefore quicker to braze through the first stage. Even people that rely heavily on controlled processes could have very swift motor reactions and therefore blink through phase three. Accordingly, even if automatic decision time would always be shorter, it does not necessarily follow that the total response time of an automatic responder is shorter. In other words, when prosocial responses are given faster, this does not necessarily mean that prosocial responses are more automatic. It could be that prosocials are quicker to understand or have better motor control.

In their own experiments, Cappelen et al. did find a robust relationship between prosocial behavior and automatic behavior in, controlling for both swiftness and cognitive ability. Nonetheless, their observation provides an interesting point of discussion relevant not only for reaction time measurements but for all methods used to measure or manipulate the reliance on automatic or controlled

processes. All methods used measure reliance automatic and controlled processes indirectly. Therefore, differences in prosociality could be caused by

(30)

side-effects of the manipulations used. This makes it harder to compare the results of the different experimental manipulations with each other.

Conclusion

In conclusion, in a classic experimental game theory setting, there appears to be an average negative correlation between reaction time length and amount of prosociality. The longer the reaction time, the more selfish the player is. The few studies that found otherwise either use a very small sample or an

unconventional game set up. This average negative correlation might very well be the result of a tendency of all extreme reactions to be quick, combined with the fact that a larger percentage of extreme deciders act extremely prosocial, as shown by Evans et al (2014). Caution is required when equating reaction times with automatic or controlled decision styles, as there are many other variables, such as intelligence and swiftness that influence the speed of decisions. Finally, reaction time studies are limited because they are correlational studies only. They can never prove that automatic or controlled processes actually cause prosocial behavior.

(31)

TIME PRESSURE

Method

Time pressure manipulations take the rationale of reaction time measures but go one step further from correlation to causal relation. Time pressure interventions try to actively disentangle automatic and controlled processes (Baddeley, 1996). By forcing subject to decide quickly, experimenters hope to limit the

performance of slow deliberate processes. The underlying thought is that when time is limited, deliberate processes that are initiated will have no time to be brought to completion and will be cut of in the middle. An advantage over reaction time studies is that because of the active experimental manipulation, a definite causal relation between time pressure and any experimental effects can be concluded.

Experiments

Well known amongst the time pressure studies are those carried out by David Rand (A. M. Evans et al., 2014; Rand et al., 2014, 2012) In 2012, Rand Green & Nowak studied the effects of both time pressure and time delay on PGG contributions in two different experiments and find a positive effect of the former and a negative effect of the latter. In 2014 Rand, Peysakovich and Kraft-Todd found similar effects in a metastudy of 15 economic games they carried out (mostly one-shot PGGs but also a PD, a hypothetical DG and a repeated PGG). The majority of the experiments carried out by Rand were conducted through

Amazon Mechanical Turk, an online labor market through which people from different countries can perform different online tasks, such as participation in an experiment, against a small fee. A small minority of experiments is carried out with Boston area students.

Evans et al. (2014) used the data of Rand et al. (2014) as well as data from additional experiments to look at the effect of time pressure on the amount of intermediate versus extreme responses. They found that the overall positive effect of time pressure on prosociality consisted of a decrease in extreme selfish

(32)

responses and an increase in extreme prosocial responses. The share of intermediate responses remained unchanged under time pressure.

Verkoeijen & Bouwmeester (2014) as Tinghög et al. (2013) challenged the conclusions of Rand and his colleagues. Both studies replicated the PGG experiments carried out by Rand at al., Tinghög et al. recruited Australian

students and Verkoeijen & Bouwmeester recruited subjects using the online tool MTurk. Neither study found a significant effect of time-pressure.

Finally, there are a number of studies of Ultimatum games under time pressure. For example that of Cappeletti (2011) who finds that proposers make higher offers under time pressure. However, the problem with studying proposer-behavior in an ultimatum game is that the proposers proposer-behavior is not only influenced by his prosociality, but also influenced by his strategic behavior and ability to predict reactions of the responder. The responder’s behavior, on the other hand, tells us something about the human desire for reciprocal

punishment, not necessarily about his prosociality. It can therefore be stated that the ultimatum game is of lesser interest for this inquiry.

Table 3.2: results time pressure studies

Analyses

Table 3.2 shows an overview of the time pressure studies. As with the reaction

Researchers Game Results Subjects

Cappeletti (2011) Ultimatum game Automatic prosociality 376 Rand et al. (2012) Public goods game &

Prisoners’ dilemma

Automatic prosociality 680, 207

Rand et al. (2014) Public goods game Automatic prosociality 6252 Tinghög et al. (2013) Prisoners’ dilemma

(1) & Public goods game (4)

No effect 167, 199, 583,

320, 1184

Verkoeijen &

Bouwmeester (2014)

(33)

time experiments a first glance through the experiments shows multiple contradictory results. The first thing to notice is that the majority of results favoring automatic prosociality have been found by David Rand and different constellations of colleagues. The results are collected during more than fifteen experiments, using over 6000 participants from two different subject pools and seem to be tested and retested. However, it would add to the reliability of the research if independent researches would be able to replicate the results.

The two studies that do indeed try to replicate some of the studies of Rand (Tinghög et al., 2013; Verkoeijen & Bouwmeester, 2014), fail to replicate the results. This could possibly be explained by heterogeneity of the population. Rand himself failed to find significant effects in some of the later MTurk studies that were included in the 2014 meta-analyses (Rand et al., 2014) . An alternative explanation for this lack of significant results in the study by Verkoeijen & Bouwmeester is the supposed recent corruption of the MTurk population (Rand & Kraft-todd, 2013). Rand claims that automatic prosociality is not an essential human quality, but a behavioral pattern formed through experience. Therefore, it would possible that automatic behavioral patterns of the MTurk population have changed, for example by their repeated exposure to online economic games. In any case it is important to note that while some experiments fail to find any effect, none of the selected studies found a negative effect of automaticity on prosociality through time pressure experiments.

The results obtained by Evans et al. (2013) support the conclusions of Rand et al. (2014). Even when considering the amounts of extremely prosocial responses, intermediate responses and extremely selfish responses separately, time pressure appears to increase extremely prosocial responses and decrease extremely selfish responses. However, these results disagree with results found during reaction time experiments discussed in the previous section, where shorter reaction times were positively correlated with both extremely social and extremely selfish responses. Time pressure and reaction times appear to relate to social behavior in different ways, and it has to be concluded to either one of the two or both cannot be taken as a direct proxy for automatic behavior.

(34)

Evans et al. suggest that it is reaction time that does not form a direct proxy for automatic behavior. They claim that reaction times are not only an indication of the amount of deliberation used, but that they are also strongly influenced by level of experienced conflict during a decision. They indeed find a correlation between experienced conflict and reaction times. However, this could be perfectly explained by stating that decision-conflict triggers increased

deliberation. If you have two simultaneous conflicting impulses it makes sense to think a bit longer to decide which of the two impulses to act on.

On the other hand, there are multiple mechanisms that could potentially

interfere with the effects time pressure has on automaticity, and therefore limit it's function as a proxy of increased automaticity. Time pressure encompasses more than the simple fact of limited time, it also entails responder awareness of time limitations and that awareness might have an effect in itself.

Research has identified three additional effects time pressure can have on performance. Time pressure can lead to stress, which can distract people from the task at hand and impede their performance. Additionally, it requires

timekeeping, a constant monitoring of the time remaining, which can be an extra burden on mental capacity in the same way that cognitive load is. Finally, time pressure has been found to help people work in a more focused manner, leading to an increase rather than a decrease of performance (Roskes, Elliot, Nijstad, & De Dreu, 2013). Therefore, while stress and timekeeping could work

concurrently with time pressure in decreasing free executive capacity, the tendency of time pressure to increase focus might have the opposite effect. Additionally stress or focus might influence prosociality directly, without mitigation of automaticity. Stress, for example, has been directly linked with an increase in prosocial behavior (Dawans, Fischbacher, Kirschbaum, Fehr & Heinrichs, 2012)

Finally, as time pressure is supposed to limit the functioning of slower brain processes, it could potentially also limit the functioning of slower automatic

(35)

brain processes. As mentioned last chapter, LeDoux mentioned two fear pathways, a 'low' quick thalamus-amygdala pathway and a 'high' slower

thalamus-cortical-amygdala pathway. While both of them are automatic and fast, one is yet faster than the other. Therefore, besides constraining controlled processes, time pressure might therefore also constrain automatic high pathways signals

Conclusion

Results of time pressure studies are contradictory. The vast amount of supporting evidence of automatic prosociality collected by Rand seems convincing but does not succeed in being replicated. If a corrupted MTurk subject pool is indeed the culprit, new non-MTurk replications are needed to support Rand's claims. At the methodological level, time pressure as such can be used as a proxy for intuition but should never be equated with it. There are ways in which time pressure might influence prosociality, without mitigation by (increased) automaticity, such as through an increase in stress levels.

(36)

COGNITIVE LOAD

Method

The second manipulation that will be discussed is the cognitive load

manipulation. As with time pressure experiments, cognitive load experiments try to limit the exertion of controlled thinking. In this case, by distracting the

controlled system with a concurrent secondary task.

Cognitive load interventions have their roots in Baddeley's ideas of working memory and the central executive. Essential in cognitive load manipulations is the idea that the working memory has limited capacity, and can only carry out a limited number of things simultaneously. By burdening the participant with a secondary task, limited working memory capacity is left to use on the primary task. Originally, cognitive load tasks were used by Baddely to show the

independence of the slave systems. He showed that a visual and a phonological task could be carried out simultaneously without altering performance while two visual or two phonological tasks carried out at the same time would interfere with each other. Nowadays, cognitive load tasks are often used to measure the effect of a constrained central executive.

Many traditional cognitive load tasks predominantly appear to make use of the slave systems. The string assignment for example, during which the participant has to remember a string of usually seven digits, would rely heavily on the phonological loop. Toms (1993) however, discovered that both verbal and spatial memory loads would interfere with conditional reasoning tasks,

suggesting that heavy slave system loads could overflow to the central executive.

Most famously, Shiv and Fedorikhin (1999) used this effect when showing that under a cognitive string load, people are more likely than usual to eat a chocolate cake when being forced to choose between the cake and a fruit salad. While chocolate cake is expected to generate the highest amount of instant satisfaction, the fruit salad is thought to be the healthier and therefore more beneficial option in the long run. Future-orientation requires hypothetical

(37)

thinking and impulse inhibition, which are typical controlled central executive processes.

Experiments

The first to examine the effect of automatic processing on prosociality using cognitive load were Benjamin et al. (2006). Benjamin et al. found no significant effect of cognitive load on DG selfishness with a small (37) sample. As load the study used the traditional 7-item string assignment. Hauge et al. (2009), seeking to improve these results, carried out a number of dictator games under cognitive load with a larger samples (62, 28, 122 and 146). Even after increasing the degree of difficulty from seven numbers to a combination of numbers and letters Hauge et al. found no significant effect of cognitive load on allocations.

Schulz, Fischbacher, Thoni and Utikal (2014), suspecting that the cognitive load that had been used by earlier researchers might not be demanding enough, repeated the experiment using an n-back test with n=2 as cognitive load (Jaeggi et al., 2003). The n-back test requires subjects that are exposed to a recording of letters to press a button when they hear a letter that sounded two letters before. The test requires not only memorizing, but also monitoring and updating, and is therefore thought to be more demanding of the working memory than the 7-item-string task. This indeed resulted in the registration of significant results: a positive effect of cognitive load on dictator game offers, or, a positive effect of automaticity on prosociality.

Cornelissen et al. (2011) managed to get significant results despite using the lighter 7-digit load. In line with Benjamin et al. and Hauge et al., Cornelissen et al. did not find any overall significant effect of cognitive load using the string-exercise. However when separating the participants into prosocials and proselves based on their SVO scores in the Ring test, Cornelissen did find a significant increase of prosocial behavior by prosocials under cognitive load. This suggests that automatic behavioral pattern difer from person to person, and only for a certain group of people (prosocials) does an increased reliance on automatic processes lead to an increase in prosociality.

(38)

Table 3.3: results cognitive load studies

Analyses

Table 3.3 shows the results of the experiments using cognitive load tasks. The null results found by Benjamin et al. (2006) and Hauge et al. (2009) using a string assignment combined with the positive results found by Schulz et al. (2014) using an N-back task suggest that the string assignment might not prove enough of a burden on the deliberate system to significantly disrupt its

functioning. Judging from Baddeley's framework, the assignment engages the slave systems but is not demanding enough to cause their overflowing towards the central executive. The results of Cornelissen et al. (2011) however paint another picture. Cornelissen et al. do find an effect of the string assignment on prosociality, but only for prosocial people. This means that the overall null effects found in other studies might also be the results of a combination of a null effect for proselves and a positive effect for prosocials. Moreover, the significant aggregate effect found by Schulz et al., besides being explained by a stronger effect of the n-back test, could also be explained by a possible larger share of prosocials in their subject pool.

The results obtained by Cornelissen et al. relate to earlier described patterns of heterogeneity of social preferences. It has been established that social

preferences differ from person to person. It is not unlikely therefore that the processes causing and influencing those social preferences also differ from person to person. The results of Cornelissen et al. suggest that some people, prosocials, have automatic prosocial reactions and other people, proselves, do

Researchers Game Results Subjects

Benjamin et al. (2006) Dictator game no effect 37 Cornelissen et al. (2011) Dictator game automatic prosociality for

prosocials

160

Hauge et al. (2009) Dictator game no effect 62, 28, 122, 146 Schulz et al. (2014) Dictator game automatic prosociality 136

The effect of automatic processes versus controlled processes on prosocial behavior