What I don't know won't hurt me? Decisionmaking under ambiguity compared to decisions under risk

Table of content

1 Introduction
1.1 Basics of decision-making
1.1.1 From the normative view to the descriptive approach
1.1.2 Prospect Theory
1.1.3 The weighting function
1.1.4 The value function
1.1.5 Ambiguity
1.2 Stress and decision-making
1.2.1 Stress
1.2.2 Cortisol
1.2.3 Stress and decision-making
1.2.4 Issue and Hypotheses

2 Methods
2.1 Sample and exclusion criteria
2.2 Experimental design
2.3 Independent variables
2.4 Dependent variables
2.5 Control variables
2.6 Between-subject variables
2.7 Materials
2.7.1 Decision-making Game
2.7.2 Stress-induction
2.7.3 Mood-Scales
2.7.4 Blood-pressure
2.7.5 Intelligence Tests
2.8 Test execution
2.9 Statistical analysis
2.9.1 Control-variables
2.9.2 Test of the hypotheses
2.9.3 Between-subject-factors
2.9.4 Correlations

3 Results
3.1 Control variables
3.2 Expanses
3.2.1 Results for hypothesis 1
3.2.2 Results for hypothesis 2
3.2.3 Results for hypothesis 3
3.2.4 Results for hypothesis 4
3.3 Correlations

4 Discussion
4.1 Control variables
4.2 Trend of expanses
4.3 Comparison of games
4.4 The influence of stress
4.5 Stress and ambiguity
4.6 The process of decision-making-A theory
4.7 Limitations
4.8 Further research and conclusion!

5 Bibliography

1 Introduction

1.1 Basics of decision-making

1.1.1 From the normative view to the descriptive approach

Decision-making describes a choice between competing alternatives of action based on their relative value of consequences (Starcke, 2012). There are two different views on decision-making: A normative one that postulates how people should behave, if their aim was to maximize the outcome and a descriptive one, which describes how people actually behave (Tversky, Advances in prospect theory: Cumulative representation of uncertainty, 1992). A common normative view is resembled in the expected value algorithm (1): A decision is made as a result of a rational consideration, in which the probability (p) of occurrence of each consequence (i) is multiplied with the value (x) of this consequence (i). Then all possible consequences are summed up for each alternative (j). The alternative (j) with the highest expected value ( ) is then chosen (Stigler, 1950).

illustration not visible in this excerpt

As early as 1783 Bernoulli claimed a different view on decision-making (Bernoulli, 1954). His theory posits that expected value and expected utility as subjective evaluation of the value differ in important points. His approach can thus be considered as a descriptive one. This idea was later expanded by Neumann and Morgenstern to expected utility theory (Neumann, 1953). They proved that if four axioms were fulfilled you can find an utility function for every decision-maker. Nowadays the “prospect-theory” as the dominating theory of decision-making has developed these approaches.

1.1.2 Prospect Theory

Prospect-theory distinguishes between two different phases of decision-making called “framing” and “evaluation” (Tversky, Rational choice and the framing of decisions, 1986). In the first phase, a preliminary estimation of the possible acts and outcomes of the alternatives is done. Framing depends on people´s norms and their current emotional state and therefore it is susceptible to emotional heuristics (e.g. if the options are presented as wins or losses). In the second phase, the different alternatives are evaluated and the one with the highest value is then chosen. The two phases are similar to the two processing-mechanisms postulated by the “dual-process-theory”: A first process, which works in parallel and enables us to frame alternatives quickly, a system that is capable of emotional heuristics and a second type, which works in serial and rational and evaluates the single alternatives, being able to correct mistakes made by system 1 (Kahnemann D. S., 2002).

The process of evaluation resembles the expected-value approach, because the probability of occurrence of each consequence is multiplied with the value of it. But there are two fundamental differences. First of all and contrary to the rational model of choice it is not the absolute value of an alternative, which is examined, but the actual change in value: A reference point is set prior to a decision and then the possible deviations from this reference point are evaluated (Tversky, Rational choice and the framing of decisions, 1986). E.g. the expected utility of an alternative where you have 200 € fixed and have the chance to either win or lose 200 € with a chance of 0.5 differs from one where you have the chance to win 400 € with a chance of 0.5, although expected values are equal for both alternatives. The second difference refers to the subjective evaluation of probability and value (2). In prospect theory, the value (e.g. losses and gains) is transformed by the value function changing value (χ) into subjective value (υ). Same goes for the probability (p): p is multiplied with a decision weight (π) changing p into a subjective estimation of p. Summed up over all consequences (i) of an alternative it amounts to the utility (U) of an alternative. The utility (U) of the prospect theory differs in important points from the expected value approach (1.1) and is susceptible to cognitive biases. In what way will be topic of the following two sections.

illustration not visible in this excerpt

1.1.3 The weighting function

illustration not visible in this excerpt

Figure 1: Weighting function. Author’s own figure.

The function, transforming the probability (p) of an event, is called “weighting function” (π(p)). This function has three important properties: First, the function is normalized for π(0)=0 and π(1)=1 meaning that the endpoints match with the real ratios. Second, low p are overestimated π(low p)>p, but moderate until high p are underestimated π(moderate/high p)<p. Third the effect of moderate and high p is stronger than the one of low p, an effect called “subcertainty” (Tversky, Rational choice and the framing of decisions, 1986).: π(low p)+π(1-low p)<1.

1.1.4 The value function

The “value function” transforms losses and gains deviating from a neutral reference point with the value zero. The value-function is s-shaped, concave for wins and convex for losses (Kahnemann D. T., 1986, see fig. 2). This has two effects: For high wins it leads to a diminishing of subjective value and consequently risk aversion. For losses it leads to a diminishing of subjective value as well, but now with the consequence of risk seeking, because further losses are less unpleasant. Secondly the function is steeper for losses than for gains. That means the displeasure of losing an amount is greater than the pleasure of winning the same amount- an effect called “loss aversion” (Tversky, Rational choice and the framing of decisions, 1986).

illustration not visible in this excerpt

Fig. 2: Value function. Author’s own figure.

1.1.5 Ambiguity

In 1961, Daniel Ellsberg carried out an experiment, the results of which violated the independence axiom, a central axiom of the expected utility theory, on a behavioral level (Ellsberg, 1961). The independence axiom states that people´s preference for a bet A over a bet B should stay the same, when identical conditions are added to both bets (Neumann, 1953). In Ellsberg Experiment people´s choice reversed after adding identical conditions- a phenomenon called Ellsberg-paradox. This finding has been replicated in diverse studies e.g. in the Allais-paradox (Machina, 1981). One of the main implications of these studies is that people avoid uncertainty, even with irrational consequences, an effect called certainty-effect (Kahnemann D. &., 1979). According to Curley et al. this effect is caused by people´s fear of being embarrassed (Curley S. P., 1986). At this point “risk” has to be separated from “ambiguity”. Risk is given in the case of known p.s of an event, but uncertainty concerning the outcome of it. Ambiguity is given, when the uncertainty concerns both, the p. and the outcome: Since not even the p. of the event are known (Payne, 1992). Some authors argue that you cannot separate decisions in a binary way like this, but rather in a gradual way. According to them decisions are arranged on a dimension, which reflects the certainty of the outcome. Ignorance (the outcomes are unknown) was one extreme of the scale, certainty the other one and risk and ambiguity were in between (Starcke, 2012). If the uncertainty is high, a rational choice following mathematical rules is not possible. Consequently these situations are potentially leading to an enhanced use of the intuitive system (ibid.).

However, the certainty effect seems to depend on the decision domain: In gain domains (when money can be won) people discriminate between different p. and are pessimistic about the outcome of an uncertain event. Pessimistic belief is defined as a decision with 0.5 p. of winning being preferred over an uncertain chance of winning. In contrary in loss domains (when money can be lost) people do not discriminate between different p. Consequently the ambiguous decision is equal to the decisions varying in chances (e.g. 1/8 p.=1/2 p.=ambiguous p.) (Cohen, 1987).

1.2 Stress and decision-making

In real life stress and decision-making often occur at the same time. The purpose of the following passage is consequently to give a short introduction into the principles of stress and an overview of the research on stress and its interaction with decision-making.

1.2.1 Stress

In new stress-conceptions stress is seen as an adaption-mechanism allowing the organism to organize its behavior in a way that makes it possible to pursue superior aims under changing environmental conditions. Due to this view stress appears when the current perceived internal or external state differs from the expected state. This discrepancy triggers compensational-mechanisms (Het, 2012).

There are two physiological axis involved in stress-reaction. A fast-reacting one, involving the sympathetic nervous system (SNS) and a slow-reacting one, involving the hypothalamic-pituitary-adrenal axis (HPAA). The SNS-axis works by activating the sympathetic nervous system along the spinal cord and then directly innervates organs (e.g. heart) and the adrenal medulla, which releases adrenalin and noradrenalin (Het, 2012). The HPA-axis is controlled by a hormone cascade: First the hypothalamus is activated which releases a hormone activating the pituitary gland. The gland in turn releases a hormone activating the adrenal cortex which then releases glucocorticoids (GCs). The GCs can pass the blood-brain barrier. There it can bind on two receptor-types and thereby directly modulate the brain´s activity (Starcke, 2012). The GCs reach their peak about 21-40 minutes after the stressor appeared (ibidem). In contrast, adrenalin and noradrenalin cannot pass the blood-brain barrier and reach their peak shortly after the stress-exposure (Het, 2012). Still, they can affect the brain via activating the sensory vagus nerve, which innervates the brain (Starcke, 2012). Both stress-axes can act in concert e.g by modulating the basolateral amygdale activity (Roozendaal, 2009).

1.2.2 Cortisol

Cortisol is a stereoid hormon, which is released in circadian rhythms (Weitzmann, 1971) and as reaction to stressful events (Kirschbaum C. P.-M., 1993). For circadian release four phases can be identified (Weitzmann, 1971): A 6 hour phase of minimal secretory activity 4 hours before and 2 hours after begin of sleep. A 3 hour sleep period with moderate cortisol release (3.-5.hour of sleep). A 4 hour phase of high cortisol level between the last three hours of sleep and the first hour of awakening. And the remaining day phase with a high variability of cortisol release.

The extend to which it is being released under stress depends on several factors like gender, age and psychopathology (Kirschbaum C. K.-H.; 1995, Heather, 2005; Jacobs, 1984). Cortisol is known to have various effect on psychological variables like decision-making (Buckert, 2014), working-memory (Oei, 2006), declarative memory (Kirschbaum C. W., 1996), attention (Vedhara, 2000) and extinction (Stark, 2006). Cortisol effects can be mediated on two ways: A non-genomic, directly affecting the activity of neurons within the brain, and a genomic one, influencing the expression of genes (Groeneweg, 2011). The non-genomic way occurs shortly after stress-exposure, while the non-genomic one occurs after several hours (ibid.).

1.2.3 Stress and decision-making

In the following part a short review of the research on the interaction of stress and decision-making will be given.

As early as 1987 Keinan (1987) surveyed the effect of stress on the performance in an analogies task. Contrary to the recent studies on this topic, the focus of his study was not the influence of stress on the willingness to run risks in financial decisions, but the causes of changed decision-behavior in a cognitive task under stress. The results indicated a decreased performance. Behavioral causes were “premature closure”, which means stopping to scan all alternatives before deciding, and “non-systematic-scanning”, meaning a chaotic pattern in the search for the right answer.

As far as I know only one study directly administered cortisol and observed the effect on decision-making in a modified CGT (Putman, 2010). Cortisol leads to an increased number of risky games played, if the p. of loosing and the amount that could be lost were both high. This effect became even stronger, when there was a high possible gain at the same time. The authors interpreted this interaction-effect as a reduced sensitivity for punishment and an increased sensitivity to reward after cortisol-induction.

Some studies examined the effects of stress under risk. Delgardo and Porcelli (2009) used a modified version of the Cambridge gambling task (CGT), in which the expected values are predictable. They observed more conservative decisions for gain trials and more risky decisions in loss-trials under the influence of stress. This indicated that the effects described by the prospect-theory are enhanced under stress. It should be noted they conducted the decision-task shortly after the stress-induction, so that an effect of GC is questionable, because it has a delayed peak (s. above). Moreover they did not control cortisol directly, but by indirect (e.g. heart rate). Brand et. al. (2008) used the Game of dice task (GDT) to test whether social stress-induction leads to a shift in choices. The GDT uses explicit rules and has a low level of uncertainty. In this game choices with high risk are disadvantageous for the player, while choices with low risk are advantageous. Each trial offers possible wins and losses at the same time. The results of the study showed that the stress group had a lower score, indicating that they made less advantageous and hence more risky decisions. Furthermore, the individual cortisol-level was negatively correlated with the score, meaning that cortisol occurs in conjunction with risky behavior. Pabst et al. (2013 A) used a similar design, but modified the GDT so that gains and loss domains were separated. The results showed that the stress group does not deviate from the control group in the gain domain, but in the loss domain. Stressed subjects made less risky decisions in the loss domain. In a further study decision making under risk was examined with respect to executive functions (Pabst S., 2013 B). GDT was used as decision game. A 2*2 between-subjects design was used: Subjects were matched to social stress vs. placebo condition and to either a parallel executive task (2-n-back) or to no parallel task. Results showed an interaction-effect of executive task and stress: Stress only lead to riskier behavior, when the executive task was absent, while it did not when subjects performed the task. Regression-analysis suggested that executive functions are moderating stress effects on decisions-making. One study, using a highly explicit financial decision task to examine decisions under risk, did not find a significant difference between the group under acute stress compared to a control group (Lempert, 2012).

Young et al. (2012) compared decision making under time pressure with decision making in the absence of time pressure. The decision game consisted of a binary bet. Data were modeled according to prospect theory. Thus the bets did not contain a fixed p. nor did subjects learn by feed-back, but p. was estimated according to their previous rating the experiment can be considered as a mix of risk and ambiguity. Bets were either framed as gains or losses. Results indicated that time-pressure increased risk-seeking for the gain domain and reduced the ability to discriminate between p. for the loss-domain.

A few studies investigated the effects of stress on rather implicit tasks, where subjects learn the rules by feed-back provided within the game (Preston et al. (2007), van den Bos et al. (2009)). This kind of task has thus a higher degree of uncertainty and is closer to the construct of ambiguity. Preston et. al. (2007) used the Iowa gambling task, in which expected-values are implicit. They did not find a change in overall-performance, but a change in the learning-curve. The stressed participants tend to learn slower in the first trials and faster in the last trials. They did not control the GC-level as well, but measured indirect effects like heart-rate. Therefore their results can be attributed to stress, but cannot be ascribed to the GC-level as a hormonal mediator of that effect

One recently published study examined the influence of stress on decisions under risk and ambiguity (Buckert, 2014). A financial lottery task was used as decision game. They directly compared ambiguity independent of feed-back learning with risk tasks. Domains (gain, loss, mixed), expected values and degrees of certainty (ambiguity vs. risk) were varied within the experiment. Stress was induced using the Trier Social Stress Test and the effects were compared to a control group. Cortisol response was controlled and stressed subjects were separated into responders having a high cortisol response and non-responders with low response. The results showed an elevated percentage of risky choices of cortisol responders in the gain domain under risk. Unlike for ambiguity: Here no effects of stress could be observed. Moreover, the study pointed towards the mediator-effect of cortisol: Not stressed subjects in general showed the effects, but only subjects showing a high cortisol response.

1.2.4 Issue and Hypotheses

All in all, studies reveal an effect of stress on decision making, but the direction might depend on the decision task (e.g. the degree of uncertainty), the involved stress-axis and co-variables like gender and age (Starcke, 2012). This study has hence the aim to answer some of the remaining questions. Particularly, the question how stress, ambiguity and decision-making interact has barley been addressed by prior studies and is thus the main issue of this study. In this study, participants played four games under two treatments (stress vs. no stress), respectively. The games had identical rules, but varied in their probability to lose: One game has a high risk of losing, one a low risk, one a moderate risk and one an unknown risk (the condition with ambiguity). Aim of all games was to retain as much of a fixed start capital as possible. The game was a virtual board game in which each step cost credits. To retain the start capital subjects had to find a balance between two contradictory behaviors: On one hand, they had to make more steps than a virtual opponent, because losing against the opponent meant losing all the credits. At the same time they had to make a minimum number of steps, because each step cost credits. According to prior studies four hypotheses have been constructed:

1. a) The decision-game, which is conducted, is a loss-game. That means subjects can not win credits in course of the game, but can only avoid losing it. To them the part of the value-function described above is valid, that concerns with losing money. If a person loses money the value of the lost gradually decreases and she becomes more risk-seeking by that. Transferred to the game this indicates that the subject´s expenses increase linear in course of the game under absence of stress.

1 b) Thus stress influences the decision-behavior it indicates a change in the expanse pattern. So under stress participants might not show a linear increase in course of the game.

2. a) In a status of ambiguity people tend to assume an adverse ratio of likelihood (Ellsberg, 1961). In loss domains however people tend not to discriminate between p. (Cohen, 1987). In my study an “uncertainty effect” is still expected due to structural difference within decision process: In Cohen`s study people directly had to choose between alternatives. On the contrary they directly evaluated the games in my study by their expanses. Thus the “uncertainty effect” is still expected: Expanses for the ambiguous game should be similar to one of the game with low p. of winning, but differ from the game with equal chances and the game with high p. of winning. This should be valid for the games under normal conditions.

2 b) Thus stress influences decision-making the pattern between the games might change after stress-induction. The game under ambiguity might now differ from the game with high p. of losing money.

3. As said before stress has an influence on decision-making. However the direction of changes is not clear up to now. Stress might make people riskier or less risky. According to the experimental observed changes under stress, the mean expense over all games might differ between stress and non-stress conditions.

4. According to the results of Buckert et al. (2014) stress has no effect on ambiguous tasks while it has one on risky tasks. Consequently there might be no difference between treatments for the ambiguous game. In contrast game A, B and C might show difference between treatments.

2 Methods

2.1 Sample and exclusion criteria

A convenient sample was used, since the recruitment took place at the university. Participation was voluntary and rewarded by experimental hours (between 2-3 hours depending on their individual time).

12 Persons participated in the experiment. One man was excluded from analysis due to a time delay after the stress-procedure. Thus the remaining sample size was 11. The average age was 22.27 years (SD.:3.00 years). All subjects studied psychology. 9 Participants were women, two were men.

Exclusion criteria were

Not speaking German fluently

Past participation on the SECPT

Drug intake or alcohol consumption within the last 24 hours

A diagnosed psychological/psychiatric/neurologic disease in the present

Pregnancy

Cardiologic diseases or high blood-pressure

All participants were exposed to both stress conditions (stress/non-stress), but the sequence of exposure was counterbalanced. One group, the AB group, was exposed to stress first, the other, the BA group, was exposed to warm water on the first day. Allocation to the groups was randomized by the subjects birth-days. Subjects having birthday in the first half of the year were group AB, the other BA. 7 persons were in the AB group, 4 persons in the BA group. The slightly imbalance is due to the excluded person and the randomization algorithm.

2.2 Experimental design

A two-group counterbalanced repeated-measure design was used in this study. All subjects were exposed to the same conditions. This design has been chosen to ensure an adequate power despite a relatively small group size. To minimize transfer effects by reason of learning effects the two stress/non-stress conditions were counterbalanced between the two groups. The sequence of stress exposure was thus interchanged, meaning half of the subjects were first in the stress and then in the non-stress condition (AB-group) and the other half vice versa (BA-group). This method is called ABBA as well, the letters indicating the sequence of conditions. Subjects were randomly allocated to the groups depending on their birthday: The ones having birthday in the first half of the year were assigned to the AB-group, the other half to the BA-group.

2.3 Independent variables

In this experiment two independent variables were used. On the one hand there is the within-subject factor “game” that is given in four levels and varies in the p. of losing against the opponent and in the degree of certainty concerning the opponent`s behavior. On the other hand there is the within-subject factor “stress” which is given in the two levels ‘stress’ and ‘no stress’.

2.4 Dependent variables

The dependent variable, which is critical for the hypotheses, is the subject´s expense per game, meaning the amount of credits that has been spent in each game.

2.5 Control variables

A few control variables were applied to check if stress-induction had worked: The blood-pressure (systolic and diastolic) and the current emotions on different scales (see 2.7.3 and 2.7.4).

2.6 Between-subject variables

Some variables, varying between the subjects, are raised to check for possible influences on the dependent variables. These variables were gender, age, cognitive skills and the sequence of conditions (AB/BA-group).

2.7 Materials

2.7.1 Decision-making Game

My tool to measure the decision-performance under ambiguity and risk consists of a virtual game, programmed in MATLAB®, in which each subject has to succeed over an unknown and unseen competitor. This is achieved by making steps on a virtual path. The possible steps range from 0 to 50 steps each round with 50 rounds each game (20 in the tutorial). Also, the steps are associated with a certain amount of money (1 credit/step), which the participant can afford due to his start capital (2500 in each of the four games, 1000 in the tutorial). Doing the maximum amount of steps continuously results in spending the whole start capital. The game´s aim is to retain the start capital received prior to each game. To do that the subjects should spend little capital as possible, which means making as few steps as possible. But there is one rule which complicates this task: They have to make more steps in total within each game than the opponent. If they make fewer (or equal), they lose all their credits they received earlier on the game, if they are made more, they can keep the credit they did not spend. So, there are two somehow contradictory strategies to save the capital: Making as few moves as possible to avoid high expenses on the one hand and making as many moves as necessary to overcome the opponent. The optimal strategy would hence be making as few steps as possible, but as many as necessary.

The decision-program consists of four games (and one tutorial game) being identical except for one condition, namely the p. of losing against a virtual opponent. Three games (and the tutorial) have a fixed p., which stays constant within the games. One game has unknown p. concerning the opponents behavior and is therefore the game in which the condition of ambiguity is incorporated. The exact p. are as following: In game A there is a 50% chance of the opponent making either few steps (a random number from 0-5 steps) and a 50 % chance of him making many steps (a random number from 45-50 steps). The same applies to the tutorial game. In game B the chances were 95% for 0-5 steps and 5% for 45-50 steps. Game C was the other way around: 5% for 0-5 steps, 95% for 45-50 steps. Game D was the one with ambiguity. The opponent did still either 0-5 or 45-50 steps, but now with unknown probabilities. Note that the sequence of games is invariable to realize identical conditions for all subjects. The sequence starts with the tutorial, game A, game B, game C and finally game D.

At the beginning of each game, the start capital and the rules valid for this game are displayed. It is pointed out that the rules do not change during the game. Before the beginning of each round the updated credits and the rules are shown to the player, whereupon he has to decide how many steps he wants to make. The steps are shown to him on a virtual play-board for reasons of spatial visualization. At the end of each game, the player gets the feed-back if he won and how many credits are left.

2.7.2 Stress-induction

Stress is induced using the socially evaluated cold-pressor test (SECPT), a modified version of the cold-pressor test, leading to a stronger stress response of the HPA-axis (Schwabe, 2008). The SECPT had been slighly adapted to suit restrictions like the condition of the room. The SECPT uses social evaluation induced by social aversive stimuli and physical pain induced by cold water as two different kinds of stressors. First, subjects were informed about the procedure and about a video record pretended to be used for analyzing their facial expressions (The record was only simulated). It was clarified, that they could at any time take their hand out of the cold-water. After given their written consent the examiner used an excuse to leave the room. Then a male student-assistant, who was unknown to the subjects, entered the room. He asked the subjects to put their left hand up to the wrist into ice-water (0-4°C). They were instructed to keep their hand in the water as long as possible, whereby they were kept in suspense concerning the duration. During that time they are instructed to look into the camera posited in front of them. The assistant, standing at the left side of the camera, showed a reserved behavior: He did not greet the subjects, did not show any signs of kindness and observed the subjects all the time. All he said were a few standardized sentences to confront the subjects with mistakes they made. He said “put the hand deeper into the water”, if the hand was to close to the surface, “look into the camera”, if they looked elsewhere, “spread your fingers”, if the subjects made a fist. If the subjects still kept their hands in the ice-water after 3 minutes, they were asked to remove it. The assistant left the room and the examiner entered the room again.

For the control condition the warm-water test was used (ibid.). In this test warm-water (33-37°C) instead of cold water was used. Also, there were no camera, no announcement of behavioral analysis and no direct observation. The experimenter stayed in the room and no assistant entered the room. The examiner showed a normal behavior and did not use standardized sentences to call attention to the subject`s mistakes. This warm-water test leads to distinct lower cortisol response than the SECPT and is therefore an appropriate placebo-condition (ibid.).

2.7.3 Mood-Scales

A German and modified version of the Self-Assessment-Manikins (SAM) was used (Bradley, 1994). The SAM is classified as a quick and valid estimation of the current emotional state. It consists of three scales referring to the subject`s current emotional state: Pleasure, Arousal and Fear. Each scale has 9 levels. For pleasure level 1 indicates “very unpleasant” and 9 “very pleasant”. For arousal 1 indicates “very calm and 9 “very aroused”. For fear 1 indicates “not anxious at all and 9 “highly anxious”. Each of the levels of the scale for pleasure and arousal is illustrated by comic-figures (s. appendix for more).

Additional four German questions were used to assess the emotional state during the stress-procedure itself. The four questions (FQ) consist of 10 levels and refer to the perceived pleasure, stress, pain and difficulty during the cold and warm water test reaching from “not at all” to “extreme” (s. appendix).

2.7.4 Blood-pressure

Systolic and diastolic Blood-pressure was assessed by a mechanical blood-pressure meter.

2.7.5 Intelligence Tests

The Mehrfachwahl-Wortschatz-Intelligenztest (MWT-B) (Lehrl, 2005) is conducted to get an economic estimation of the participant´s verbal skills. It consists of 37 items, each item consisting of one real German word and four fictive words with the items arranged according to difficulty. The task is to find the existing word. The whole execution takes approx. 5 minutes.

The first subtest of the first test-half of the CFT-R 20 (Weiß, 2006) was conducted afterwards to get an economic estimation of the participant´s non-verbal skills. The CFT-R 20 has a high factor-loading on the fluid ability after Cattell (ibid.). The subtest used consists of 15 visual-spatial tasks. In each task a sequence of three patterns is given and a fourth pattern has to be chosen out of four other patterns which logical complements the sequence. The time is limited to three minutes.

2.8 Test execution

The experiment was conducted in a room of Ruhr-Universität Bochum. Every participant took part in two sessions at an interval of one day up to 9 days being identical in setup, only varying in the stress-condition. The start of the experiment lay between 8.30 a.m and 14.30 p.m to roughly control circadian cortisol release.

illustration not visible in this excerpt

Fig. 3 Test execution for both treatments (Author’s own figure)

After welcoming the participants, giving them an information sheet, which explains the procedure to them, getting their written consent and answering remaining questions the experiment starts. First they fill in a questionnaire asking for demographic data and the exclusion criteria (s. appendix). Then the participants fill in the SAM (Pre-measuring). After that the blood-pressure measurement is conducted (on the right hand). It was measured three times (with intervals of 45 seconds) to equal out measuring errors. Then they are either exposed to stress by the SECPT or to the placebo warm-water condition. During the test blood-pressure is measured again (three times). Subsequent to the test they fill in the SAM again (Post-measuring) and answer the four questions concerning their emotions during the water-test itself. After that a pause of 15 minutes was inserted. During that time the subjects could read different magazines offered to them. 18 minutes after begin of the SECPT/Placebo-test the decision-game started. Initially the rules were explained and they played the tutorial. This took approximately 3 minutes, so the real decision game started ca. 21 minutes after begin of SECPT/Placebo-test. In case of stress-condition the CFT-R 20 and the MWT-B was conducted afterwards. After the second session the real purpose of the experiment was explained. At the end the subjects were thanked, rewarded and it was appealed for secrecy concerning the purpose of the experiment (see. Fig. 3 for illustration of the procedure).

2.9 Statistical analysis

Statistical analysis was done by SPSS®.

2.9.1 Control-variables

To check for the effectiveness of stress-induction the means of the control-variables blood-pressure, SAM and the four questions were compared across and within conditions. Kolmogorov-Smirnov test (KST) was conducted-both to check for the assumption of normal-distribution (rejected at p<.05). KST was performed on variables, which contained values of all conditions (e.g. treatments, games, measure-points). This procedure was chosen to equal out measure-points and random deviations likely to occur in small sample sizes. The comparisons were all done with a repeated-measure-ANOVA (RMA) with p<.05.

To check assumption of normal-distribution one blood-pressure mean was computed consisting of systolic and diastolic values overall conditions and measure-points.

To check the effect of stress-induction on blood pressure eight blood-pressure means were computed: Four systolic ones, one for each time-point (pre/post) for both conditions (warm/cold), each consisting of the three measuring-points. Same was done with the diastolic variables. Afterwards four RMAs were conducted. The systolic variables were compared within conditions (pre vs. post) both for cold and warm water condition and the diastolic variables were compared between conditions (pre vs. post) for both conditions as well.

To check for normal-distribution of the four questions one mean was computed: The mean of the four questions for both conditions were accounted to one variable (again to equal out measuring errors). This variable was tested with KST.

To check the effectiveness of stress induction four RMAs were conducted on all scales: The variables for the scales pleasure, stress, pain and difficulty were compared between treatments.

The SAM scales were not accounted, because of the different content between them. Hence 3 variables (pleasure, arousal and fear over all treatments and measure-points) were checked for normal-distribution.

To check for effectiveness of stress induction on SAM scales six RMAs were conducted to compare measure-points (pre vs. post) for both treatments (stress/no-stress) and all three scales.

2.9.2 Test of the hypotheses

The subject`s expenses were compared in the same way as the control-variables. First the assumption of normal-distribution was checked by KST (p<.05). Then different RMAs were conducted (p<.05) between and within conditions. Post-Hoc-Test for pairwise comparisons was included (p<.05). P-Values were adjusted with Bonferroni-correction. Test of with-subject contrasts was included to fit different trend-models to the data (linear, quadratic, cubic, p<.05).

First one variable was computed out of the subject´s average expense for all games and treatments. Again, this merged variable is assumed to equal out measuring errors and random deviations. Then normal-distribution was examined.

To check the first hypothesis (1A/B) four variables were computed for both conditions (warm/cold). The average expanse for the first, the second, the third and the fourth quarter was computed across all games. Then two RMAs with four levels (1.,2.,3.,4.quarter) were conducted for both conditions (cold/warm).

To check the second hypothesis (2A/B) the same procedure as before was used, but now the means of the four games were compared. Thus two RMAs with four levels (Game A, B, C, D) were conducted (for cold and warm). To check for equality between games the alpha-error level had to exceed p>0.2, because a lower level would lead to an inflation of beta-errors and by that the hypothesis would be inacceptable favored.

To check the third hypothesis the average expense across all games was computed for both conditions (cold/warm water). Then a RMA with two levels was conducted to compare the mean expense between stress and no-stress condition. To further exclude any influences by order of treatment (AB/BA-condition) the sequence of condition was introduced as between-subject factor.

To check the fourth hypothesis the average expense for each game was compared separately between the cold and warm water condition. Hence four RMAs with two levels were conducted (e.g the average expense for Game A under stress vs. Game A under no stress). Again the p-level for comparison of game D is elevated to p.>0.2 to check for equality.

2.9.3 Between-subject-factors

A mixed between-within subject analysis of variance was conducted to assess the impact of different between-subject variables (age, gender, sequence of treatment, CFT-R, MWT-B) and to check for interactions with treatment on expenses (Hyp.3&4). To do that a median-split was computed for age, CFT and MWT to adapt their scale level. The new variables had two levels (age: young vs. old, CFT: high score vs. low score, MWT: high score vs. low score). If not otherwise reported assumption of equality of covariance matrices (Box´s test p<.001) and assumption of homogeneity of variances (Levene’s Test of Equality of Error Variances, p<.01) was not violated for significant results.

2.9.4 Correlations

Pearson`s product moment correlation will be conducted (p<.05) between systolic and diastolic BP for the post measurement under stress, MWT-B, CFT-R, age and the average expanses across all games.

3 Results

3.1 Control variables

Kolmogorov-Smirnov Test (KST) did not reject normal-assumption BP (M=98.80 (SE=2.40) D(11)=0.14; p>.20). For stress-condition RMA revealed a significant time main-effect (pre vs. post) both for systolic BP (pre: M=121.5 (SD=14.70); post: M=134.85 (SD=13.69); Wilks’Lambda=0.26, F(1,10)=28.89, p>.001, multivariate partial eta squared =.74) and diastolic BP (pre: M=73.33 (SD=6.81), post M=83.09 (SD=9.44); Wilks’Lambda=0.26, F(1,10)=28.89, p<.001, multivariate partial eta squared =.74). On the contrary no time main-effect for BP became significant for the non-stress condition (Systolic pre: M=118.61 (SD=17.51); systolic pre M=113.88 (SD=8.44); Wilks’Lambda=0.93, F(1,10)=0.80, p=.39, multivariate partial eta squared =.07; Diastolic pre M=72.09 (SD=7.61); diastolic post: M=73.06 (SD=8.07); Wilks’Lambda=0.93, F(1,10)=0.55, p=.42, multivariate partial eta squared =.07).

The compound variables for the FQ can be considered as normal distributed (M=6.27 (SE=0.52); D(11)=0.16; p>.20). All questions differ significantly between stress and non-stress-condition (Pleasure stress treatment M=6.55 (SD=1.81), Pleasure no-stress treatment M=1.09 (SD=2.66), Wilks’Lambda=0.27, F(1,10)=26.67, p<.001 multivariate partial eta squared =.73; Stress stress treatment M=5.27 (SD=2.61), Stress no-stress treatmentM=0.36 (SD=0.50); Wilks’Lambda=0.22, F(1,10)=35.39, p>.001, multivariate partial eta squared =.78; Pain stress treatment M=7.27 (SD=2.00); Pain non-stress treatment M=0.91 (SD=0.30); Wilks’Lambda=0.8, F(1,10)=119.10, p>.001, multivariate partial eta squared =.92;difficulty stress treatment M=6.00 (SD=2.68); difficulty non-stress treatment M=0.00 (SD=0.00 Wilks’Lambda=0.15, F(1,10)=55.00, p>.001, multivariate partial eta squared =.85).

For all three SAM-variables the normal-distribution could not be rejected (SAM fear: M=2.30(SE=0.30), D(11)=0.15, p>0.20; SAM arousal: M=3.59 (SE=0.32), D(11)=0.15, p>0.20; SAM pleasure: M=6.84 (SE=0.31), D(11)=0.15, p>0.20).

RMA revealed no main time-effect (pre/post) for the SAM scales, neither for stress nor for no-stress (all p>.17, s. appendix for all values), except for the scale arousal for stress treatment (M pre=3.55 (SD=2.30), M post=5.09 (SD=1.76);Wilks’Lambda=0.63, F(1,10)=5.87, p>.05, multivariate partial eta squared =.37).

3.2 Expanses

Normal-distribution-assumption could not be rejected for the average expanses (M=27.59 (SE=1.38); D=0.25; p=.06). Thus normal-distribution was inferred for all games. Abbildung in dieser Leseprobe nicht enthalten

Figure 4 Trend for average expenses within the game for both treatments (Mean expanse/quarter(treatment)). Time within the game is represented on x-axis, treatment on y-axis. Bars indicate standard error. Expanse shows a constant decrease of expanses. Unlike for the Stress-condition: Expanses are increasing again in the 4.quarter. (Author’s own figure)

3.2.1 Results for hypothesis 1

RMA revealed a significant main-effect between the four time-points for the no-stress treatment (Wilks’Lambda=0.46, F(3,8)=3.11, p=.09, multivariate partial eta squared =.54, see fig. 3). The test for within-subject contrasts revealed a significant linear trend between time-points (F(1,10)=10.80; p<.01). RMA did not show differences between the time-points for stress treatment (Wilks’Lambda=0.46, F(3,8)=3.13, p=.09, multivariate partial eta squared =.54, see fig. 4). Test for within-subject contrasts rejected a linear-slope (F(1,10)=0.20; p=0.67), but supported a quadratic (F(1,10)=5.72; p<.05) as well as a cubic (F(1,10)=5.17; p<.05) trend.

3.2.2 Results for hypothesis 2

RMAs showed a significant main-effect between games for the no-stress treatment (Wilks’Lambda=0.95, F(3,8)=25.39, p<.001, multivariate partial eta squared =.91, see tab. 1 for means). Post-hoc test showed a significant difference between game D and game B (T(10)=-4.57; p<.01) and between game D and game C (T(10)=4.15; p<.05) (see fig. 5). Indeed, post-hoc test did not show a difference between game A and game D (T(10)=3.01; p=.08, but the p-value did not exceed the defined level of p>.2. A significant main-effect was found between the games for the stress-condition (Wilks’Lambda=.09, F(3,8)=26.33, p<.001, multivariate partial eta squared =.91). The post-hoc test revealed significant differences between game D and game B (T(10)=-8.77; p<.001) and between game D and game C (T(10)=4.45; p<.01) (see fig. 5). Between game D and game A existed no significant difference (T(10)=1.86; p=.56). Thus the value exceeds p>.2 equality can be inferred.

Abbildung in dieser Leseprobe nicht enthaltenAbbildung in dieser Leseprobe nicht enthalten

Figure 5 Mean expenses per game for no-stress condition (left) and stress-condition (right). The brackets and stars indicate significant differences in post-hoc tests (p<.05) between two games. Error bars represent standard error. One can notice the similarity of patterns for stress/no-stress conditions. (Author’s own figures)

3.2.3 Results for hypothesis 3

Abbildung in dieser Leseprobe nicht enthaltenAbbildung in dieser Leseprobe nicht enthalten

Fig. 6 Interaction between treatment*MWT-B (left) and treatment*age (right). Mean expense over all games is plotted on y-axis, treatment on x-axis. Color indicates level of between-subject factor. Bars indicate standard error. One can see that the means are similar for stress and differ for no-stress condition. (Author’s own figures)

The average expanses between stress and no-stress games did not differ significantly (Stress: M=27.57 (SD=5.00); No-stress: M=27.61 (SD=4.6); Wilks’ Lambda = 1.00, F (1,10) =0.001, p=.97, multivariate partial eta squared >.001). When using the sequence of treatment (AB vs. BA) as between-factor no significant interaction-effect between treatment and sequence could be assessed, Wilks’ Lambda = .82, F (1,9) = 1.87, p =.19, partial eta squared =.18. MBW-ANOVA revealed a significant interaction for age*treatment (Wilks’ Lambda = .52, F (1,9) = 8.48, p<.05, partial eta squared =.49., fig. 3) and MWT*treatment (Wilks’ Lambda =.43, F (1,9) = 12.09, p<.01, partial eta squared =.57., fig. 6). All other interactions are not significant (p>.05, see appendix for statistic).

3.2.4 Results for hypothesis 4

Tab. 1 Mean expanses for games depending on treatment (M/game(treatment)). Games are represented on columns, treatment on rows. Numbers in brackets indicate standard error. One can see that the means for each game are similar between treatment. Indeed, RMA did not reveal any significant difference between treatment. (Author’s own table)

illustration not visible in this excerpt

Fig. 7 Interaction effect of MWT-B with treatment for the ambiguous game Mean expense over all games is plotted on y-axis, treatment on x-axis. Color indicates level of between-subject factor (MWT-B). Bars indicate standard error. (Author’s own figure)

RMA did not show any significant differences between treatments for the single games (see tab. 1): Game A did not differ between treatments (Wilks’Lambda=.99, F(1,10)=0.10, p=.76, multivariate partial eta squared =.01), nor did game B (Wilks’Lambda=.95, F(1,10)=0.55, p=.48, multivariate partial eta squared =.05), nor game C (Wilks’Lambda=1.00, F(1,10)=0.03, p=.87, multivariate partial eta squared =.003), nor game D (Wilks’Lambda=.99, F(1,10)=0.10, p=.76, multivariate partial eta squared =.01). MWB-ANOVA revealed an interaction-effect between gender*treatment for game A (Wilks’ Lambda = .48, F (1,1) = 9.66, p <.05, multivariate partial eta squared =.51). Game C showed an interaction-effect for treatment*MWT (Wilks’ Lambda = .60, F (1,1) = 5.97, p <.05, partial eta squared =.40). MWB-ANOVA showed an interaction treatment*MWT for game D (Wilks’ Lambda = .61, F (1,1) = 5.85, p <.05, partial eta squared =.39, see fig. 7). All other interactions are not significant (p>.05, see appendix for statistics).

3.3 Correlations

Correlation between MTW-B with age was significant (r(11)=.85, p=.001). All other correlations did not become significant (p>.05, see appendix).

4 Discussion

In this section the results are summarized and discussed in the light of existing research. Following, a short explorative theory is drawn out of the results and prior studies. Following limitations of the study are discussed and ideas for future research are presented.

4.1 Control variables

Stress induction was successful: Blood pressure differed between time-points within stress treatment, but not within placebo treatment. The FQ differed on all scales between treatments. Arousal scale of the SAM differed between time-points within stress treatment, but not within placebo treatment. Only the two remaining SAM scales (pleasure, anxiety) did not differ significant. It is still uncertain how subjective stress and cortisol are connected. E.g. Kirschbaum et al. (1995) did not find a connection between the level of cortisol response and subjective ratings. As a result it is no falsification for the stress induction, if stress was not affecting all scales of the subjective rating. There might be other factors moderating the stress-experience and the cortisol level not controlled in this study. As evidenced by Bollini et. al. (2004) perceived control and the locus of control (internal or external control beliefs) moderate the cortisol response to a social stressor while they do not for subjective ratings. Cognitive and context factors might have different effects on subjective vs. physiological stress in this study as well. However, stress procedure was successful on most levels. Moreover is had strong effects on the physiological level, which is more relevant for this study, because it indicates cortisol response.

4.2 Trend of expanses

Hypothesis 1a can not be validated. There was a linear trend of expanses within the game, but instead of increasing the expanses were decreasing. This raises the question how the results can be explained in the light of diminishing of value and by that of an expected enhanced risk behavior during the game (Tversky, Rational choice and the framing of decisions, 1986). Weighting function as one possible explanation can not be taken into account, because p. stayed constant during the game. So effects should be in line with the value-function. The trend is very clear, so coincidental effects are not very likely (p<.01). Solution to this contradiction might be easy: As Kahnemann and Tyersky (1986) stated decision making depends on framing of the context. If people frame something as loss, the loss part of the value-function is being used. People can value the same decision problem as a win situation and thus use the gain part of the value function. Unlik at first sight the logic of the game used in this study might be one of a win game. Indeed, people were continously lossing money within the game, but central aim of the game was to gain all credits, which have not been spent. By that framing of the game might be that of a gain domain.

Hypothesis 1b can be validated. The trend of expanses within the game did not show a linear trend under acute stress. Data rather support a quadratic or cubic trend. According to Preston et. al. (2007) the learning curve changed under influence of stress. Stressed participants tended to learn slower in the first trials and faster in the last trials. Indeed the game in my study did not work with feed-back. Therefore no classical learning effect can be assumed during the game. But subjects might as well learn the rules and gain better insight into the mechanisms of the game as well. That is why the results by Preston et. al. (2007) might be valid for my study as well. An alternative explanation is delivered by Keinan (1987) Changes in the attentional system under stress (“Premature closure” and “non-systematic-scanning”) could lead to a more chaotic pattern of decision making, which differs from the value function.

4.3 Comparison of games

Hypotheses 2a/b can not be validated. For the no-stress treatment the ambiguous game differed from the high risk game and the low risk game and did not differ from the game with equal chances. It should be noted that the similarity threshold was not reached. This pattern did not change under stress: The ambiguous game still differed from high risk and low risk game and did not differ from the game with equal chances (now the similarity threshold was reached). The results deviate from the expected aversion of ambiguity postulated by Ellsberg (1961). However, Cohen et al. (1987) reported a limited sensitivity for p. in the loss domain. Therefore the absence of an ambiguity aversion could simply result out of the missing sensitivity for p. Thus the p. under risk are not taken into account the distinction between the games might get fuzzy. Nevertheless my results do not support this view, thus there is an overall significance between the games. On contrary Cohen et al. (1987) reported significant differences between risk and ambiguity for the gain domain. Here people tended to be pessimistic and rated ambigious decisions lower than decision with 0.5 p. of winning. At this place the framing of the game as either a gain or a loss domain gets important again. If the game belongs to the loss domain, then results can hardly be interpreteted as general ignorance of the probability, because the games tend to differ. If the game belongs to the gain domain, the results stand in contradiction to previous studies (e.g. Cohen et al. (1987), Ellsberg (1961), because the uncertaity effect was misssing and the ambigous game was estimated realistic.

4.4 The influence of stress

Hypothesis 3 can not be validated. Expanses did not differ between treatments. Instead, interaction effects (age*treatment and MWT*treatment) were found. Under normal conditions older adults tended to take a higher risk than younger adults. Under stress the age effect disappeared. With a partial eta squared of .49 the interaction effect can be interpreted as large. Mather et al.(2009) reported interaction effects of age with stress on decision making. However the effect was the reverse: Older subjects did not differ from younger ones under normal conditions, but differed from them under stress. Under stress older subjects took lower risks than younger ones. This contradiction can partly be explained by different aged subgroups: In my experiment young adults ranged between 19-21 years while old adults ranged from 22-29. Mather et al. (2009) used a range of 18-33 years for young and 65-89 years for old adults. One can see that the older adults in my study were closer to the younger ones in Mather et al. (2009) than to the older ones. However, the inversion of the stress-effect can not be explained by that. It can only be speculated if stress lead to a diminishing of age effects between young adults while it lead to an enhancing of age effects between old and young adults.

Interaction effects of MWT*treatment can be interpreted as large as well (partial eta squared =.57). Adults with a higher verbal IQ (MWT=35+) tended to take a higher risk under normal conditions, while adults with a lower verbal IQ (MWT<=34) tended to take a lower risk under normal conditions. Under stress this effect disappears. In a review 24 studies comparing the correlation between intelligence and the IGT was examined. Only 10 studies reported a significant correlation and the correlation was with a median value of r=.23 small. One study reported an effect of nonverbal memory on cortisol release under acute stress: Subjects with lower performance in the nonverbal memory task had a lower cortisol response (Slattery, 2013). So there might be a small effect of intelligence on decision-making on one hand and an effect of intelligence on stress on the other hand. However, no study did- as far as I know- report an interaction of intelligence and stress on decision-making. A simple explanation is offered by the high correlation between the MTW-B with age (r(11)=.85, p=.001). Older subjects tended to have higher verbal skills and vice versa, thus the interaction effect of intelligence*treatment might be explained by effect the interaction effect of age*treatment.

4.5 Stress and ambiguity

Hypothesis 4 can not be validated: None of the games differed between treatments. Indeed, the ambiguous game did reach equality level between the treatments. Though none of the other games showed a stress this can not be considered as a systematical pattern. Instead, three interaction effects could be assessed (Treatment*gender for game A, treatment*MWT for game C and D).

For game A an interaction of gender*treatment could be observed: Under normal conditions men were risk seeking while women were risk avoidant, while under stress the effects disappeared. Past studies reported interaction effects of gender and stress on decision making as well. Lighthall et al. (2009) reported more risk taking for men compared to women under stress. Contrary to my study the effect disappeared under normal conditions in their study. However, thus the male group is small in my study results should be interpreted with care. The same as said in 4.4 applies to the interaction effect of MWT-B with treatment for game C and D as well except for the relation with age. Hence age did not have an interaction effect with treatment on decision-making for game C and D it can not account for the effect of verbal intelligence. As consequence the interaction-effect could be considered as a real effect of verbal intelligence. The interaction effect of verbal intelligence did not go solely for the ambiguous game (or only for the games under risk). Therefore no interaction of verbal intelligence with stress and ambiguity can be inferred.

4.6 The process of decision-making-A theory

Many different variables seem to interaction in their influence on decision-making. Decision-making is a complex process, which might be influenced by experience, rational consideration, motivation and hormones. To better estimate the influence of stress on decision-making interaction processes must be examined. A theory combing some variables found to be important in this and previous studies is presented (fig. 8): On a first level a framing process of the domain is done (gain vs. lost domain). If stress occurs its effect depends on different moderator variables like age, gender and IQ. At the end of the process a decision is made.

4.7 Limitations

illustration not visible in this excerpt

Fig. 8 Hypothetical process chart of the decision process. On a first level framing is conducted. Stress on a second level mediated by age, intelligence and gender on a third level finally affects the decision. (Author’s own figure)

It can not be assumed that the sample is representative. Therefore results should not be transferred to general population. E.g. age and education had a limited range within sample. This had the advantage that possible interaction-effects disturbing the main effect of treatment were minimized. In reverse this limitation of range now complicates interpretation of interaction effects, which have been found. Consequently the interaction effects can only be transferred to a population of similar age and educational background. Furthermore the game conducted in this study only examined decision making within loss domain. Results should therefore be restricted to this domain. Moreover most studies addressing stress-effects on decision-making have either directly administered cortisol or measured the cortisol level. In this study indirect effects of cortisol have been measured. Consequently stress-effects can not directly be attributed to cortisol.

4.8 Further research and conclusion!

Results are encouraging: Different effects of stress could be observed and the ambiguity effect, which is long known in cognitive psychology, could be determined to a specific risk level. These are good reasons for replicating the results. Replication studies should correct the limitations described above. But there are further supplements, which could be considered.

Past studies have outlined the influence of hormonal status of women and the intake of oral contraceptives on cortisol level after stress-induction e.g. Kirschbaum, 1999. Hence future studies could examine hormonal effects on decision making. Intelligence had interaction effects with treatment: Prospective studies could include intelligence tests containing many levels of cognitive skills to examine which of these abilities exhibit effects. One possible issue could be if it were broad cognitive skills mediating the effect or rather specialized skills like the ability to operate with numbers. Another interesting issue could be to compare ambiguity in decision making between game with explicit rules and more implicit games like the Iowa Gambling Task. It is possible that stress-effects are not only mediated by the degree of certainty itself, but also by the process of gaining the information (e.g. feed-back learning or explicit declaration). Most studies, which examined ambiguity, worked with implicit feed-back task making it hard to separate between the certainty of the decision and the process of learning. Buckert et al. (2014) and this study offer ways to solve this problem. Explanation of the diverse results concerning stress effects on decision making might succeed if all the possible co-variables and mediators were controlled. Future studies might solve these problems and unify the results in a theoretical frame-work.

5

Bibliography

Bernoulli, D. (1954). "Exposition of a new theory on the measurement of Risk". Econometrica: Journal of the Econometric Society, pp. 23-36.

Bollini, A. M. (2004). The influence of perceived control and locus of control on the cortisol and subjective response to stress. Biological Psychology, 67 (3), pp. 245-260.

Bradley, M. L. (1994). Measuring emotion: the Self-Assessment Manikin and the Semantic Differential. Journal of behavior theory and experimental psychiatry, 25 (1), pp. 49-59.

Brand, M. M. (2008). Anticipatory stress influences decision making under explicit risk conditions. Behavioral neuroscience, 122 (6), p. 1352.

Buckert, M. S. (2014, May 06). Acute stress affects risk taking but not ambiguity aversion. Frontiers in neuroscience, 8, pp. 1-11.

Cohen, M. J.-Y. (1987). Experimental Comparison of Individual Behavior under Risk and under Uncertainty for Gains and for Losses. Organizational Behavior and Human Decision Processes, 39 (1), pp. 1-22.

Curley S. P., Y. J. (1986). Psychological sources of ambiguity avoidance. Organizational behavior and human decision processes, 38 (2), pp. 230-256.

Delgado, M. R. (2009). Acute Stress modulates risk taking in financial decision making. Psychological Science, 20 (3), pp. 278-283.

Ellsberg, D. (1961). Risk, ambiguity and the Savage axioms. The quarterly journal of economics, pp. 643-669.

Groeneweg, F. K. (2011). Rapid non-genomic effects of corticosteroids and their role in the central stress response. Journal of endocrinology, 209 (2), pp. 153-167.

Heather, M. D. (2005). Depression and cortisol responses to psychological stress: A meta-analysis. Psychoneuroendocrinology, 30 (9), pp. 846-856.

Het, S. S. (2012). Stress-Induced Cortisol Level Elevations are associated with reduced negative affect after stress: Indications for a mood-buffering cortisol effect. Psychosomatic medicine, 74 (1), pp. 23-32.

Jacobs, S. M. (1984). Urinary-Free Cortisol Excretion in Relation to Age in Acutely Stressed Persons with Depressive Symptoms. Psychosomatic medicine, 46 (3), pp. 213-221.

Kahnemann, D. &. (1979). Prospect Theory: An Analysis of decision under risk. Journal of Econometric Society, 47, pp. 263-292.

Kahnemann, D. S. (2002). Representativeness revisited: Attribute substitution in intuitive judgment. In G. D. Gilovich T. (Ed.), Heuristics of intuitve judgment: Extensions and applications. New York: Cambridge University Press.

Keinan, G. (1987). Decision-making under stress: scanning of alternatives under controllable and uncontrollable threats. Journal of personality and social psychology, 52 (3), p. 639.

Kirschbaum, C. K. (1999). Impact of Gender, Menstural Cycle Phase, and Oral Contraceptives on the Activity of the Hypothalamus-Pituitary-Adrenal Axis. Psychosomatic medicine, 61 (2), pp. 154-162.

Kirschbaum, C. K.-H. (1995). Sex-Specific effects of social support on cortisol and subjective response to acute psychological stress. Psychosomatic medicine, 57 (1), pp. 23-31.

Kirschbaum, C. P.-M. (1993). The ‘Trier Social Stress Test’ – A Tool for Investigating Psychobiological Stress Responses in a Laboratory Setting. Neuropsychobiology, 28 (1-2), pp. 76-81.

Kirschbaum, C. W. (1996). Stress- and treatment-induced elevations of cortisol levels associated with impaired declarative memory in healthy adults. Life science, 58 (17), pp. 1475-1483.

Lehrl, S. (2005). Mehrfachwahl-Wortschatz-Intelligenztest MWT-B. Balingen: Spitta Verlag.

Lempert, K. M. (2012). Individual differences in delay discounting under acute stress: the role of trait perceived stress. Frontiers in psychology, 3.

Lighthall, N. R. (2009). Acute Stress Increases Sex Differences in Risk Seeking in the Balloon Analogue Risk Task. pLoS ONE, 4 (7).

Machina, M. J. (1981). "Rational" decision making versus "Rational" decision modelling? (M. A. Hagen, Ed.) Journal of Mathematical Psychology, 24 (2), pp. 163-175.

Mather, M. G. (2009). To Brake or Accelerate When the Light Turns Yellow? Stress Reduces Older Adult´s Risk Taking in a Driving Game. Psychological Science, 20 (2), pp. 174-176.

Neumann, J. M. (1953). Theory of Games and Economic Behavior. Princeton, NJ: Princeton University Press.

Oei, N. E. (2006). Psychosocial stress impairs working memory at high loads:An assoiciation with cortisol and memory retrieval. Stress: The International Journal on the Biology of Stress, 9 (3), pp. 133-141.

Pabst S., S. D. (2013). Paradoxical effects of stress and an executive task on decisions under risk. Behavioral neuroscience, 127 (3), p. 369.

Pabst, S. B. (2013). Stress effects on framed decisions: there are differences for gains and losses. Frontiers in behavioral neuroscience, 7.

Payne, W. B. (1992). Behavioral decision research: A constructive processing perpective. Annual review of psychology, 43 (1), pp. 87-131.

Preston, S. A. (2007). Effects of anticipatory stress on decision making in a Gambling Task. Behavioral neuroscience, 121 (2), p. 257.

Putman, P. A. (2010). Exogenous cortisol actuely influences motivated decison making in healty young men. Psychopharmacology, 208 (2), pp. 257-263.

Roozendaal, B. M. (2009). Stress, memory and the amygdala. Nature Reviews Neuroscience, 10 (6), pp. 423-433.

Schwabe, L. H. (2008). HPA axis activation by a socially evaluated cold-pressor test. Psychoneuroendrocrinology, 33 (6), pp. 890-895.

Slattery, M. J. (2013). Neurocognitive function and state cognitive stress appraisal predict cortisol reactivity to an acute psychosocial stressor in adolescents. Psychoneuroendocrinology, 38 (8), pp. 1318-1327.

Starcke, K. B. (2012). Decision making under stress: A selective review. Neuroscience& Biobehavioral Reviews, 36 (4), pp. 1228-1248.

Stark, R. W. (2006). Influence of the stress hormone cortisol on fear conditioning in humans: Evidence for sex differences in the response of the prefrontal cortex. Neuroimage, 32 (3), pp. 1290-1298.

Stigler, G. J. (1950). The Development of utility theory. 1. Journal of Political Economy, pp. 307-327.

Toplak, M. E. (2010). Decision-making and cognitive abilities: A review of associations between Iowa Gambling Tak performance, executive functions, and intelligence. Clinical psychology review, 30 (5), pp. 562-581.

Tversky, A. &. (1986). Rational choice and the framing of decisions. The journal of business, 59, pp. 251-278.

Tversky, A. &. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and uncertainty, 5 (4), pp. 297-323.

Van den Bos, R. S. (2009). Stress and decision-making in humans: performance is related to cortisol reactivity, albeit differently in men and women. Psychoneuroendocrinology, 34 (10), pp. 1449-1458.

Vedhara, K. H. (2000, August). Acute stress, memory, attention and cortisol. Psychoneuroendocrinology, 25 (6), pp. 535-549.

Weiß, R. H. (2006). Grundintelligenztest Skala 2-Revision (CFT 20-R). Hogrefe.

Weitzmann, E. D. (1971). Twenty-four Hour Pattern of the Episodic Secretion of Cortisol in Normal Subjects. Journal of Clinical Endocrinology& Metabolism, 33 (1), pp. 14-22.

Young, D. L. (2012). Decision making under time pressure, modeled in a prospect theory framework. Organisational behavior and human decision processes, 118 (2), pp. 179-188.