280 likes | 440 Views
Information search patterns in risk judgment and in risky choices. Agata Michalaszek Warsaw School of Social Psychology. Expectation Models. rational choice is based on max EV logarithmic function of utility ( Bernoulli, 1738, 1954 ) objective value was replaced with subjectvie utility
E N D
Information search patterns in risk judgment and in risky choices Agata Michalaszek Warsaw School of Social Psychology
Expectation Models • rational choice is based on max EV • logarithmic function of utility (Bernoulli, 1738, 1954) • objective value was replaced with subjectvie utility • people violate EU theory (Allais, 1953) and common ratio rule
Expectation Models – nonlinear functions of value and p • Prospect Theory – value of each outcome is weighted by a decision weight ╥(p) – nonlinear function of probability (Kahneman and Tversky, 1979) • CPT - the separable decision weights was replaced with cumulative (rank-dependent) decision weights(Kahneman and Tversky, 1992)
Expectation Models – the same rule • all those models (i.e. extensions of EV): EV, EU, SEU, OPT, CPT contains the same rule – people choose ‘the best’ alternative by maximizing the expected value Is this a single way to look for a solution to inconsistencies between the EV rule and actual behavior?
Two approaches Extensions of EV rule i.e. nonlinear v and p functions Investigation of the way in which people think • e.g., how they acquire information? • Information board (Payne, 1976)
Information searching due to EV wl(pl) *l(loss)+wg(pg)*g(gain) (e.g. Coombs and Lehner, 1984; Jia and Dyer, 1996; Jia, Dyer and Buttler, 1999; Luce and E.U. Weber, 1986; Sarin and M. Weber, 1993) • probabilities and payoffs are combined multiplicatively • each alternative is evaluated separately (global evaluation)
Situation 1 payoff1 p1 payoff2 p2 … … … payoffi pi Situation 2 payoff1 p1 payoff2 p2 … … … payoffi pi Pattern of information searching due to EV Each alternative is evaluated separately.
Situation 1 payoff1 p1 payoff2 p2 … … … payoffi pi Situation 2 payoff1 p1 payoff2 p2 … … … payoffi pi Pattern of information searchingdue to DIM Each dimension is evaluated separately. Dmensional Model – Payne, 1976
Situation 1 payoff1 p1 payoff2 p2 … … … payoffi pi Situation 2 payoff1 p1 payoff2 p2 … … … payoffi pi Two patterns of information searching DIM EV
Main research question Do peopleuse: the multiplicative or the dimensional pattern of information acquisition, while making risky choices ?
Risk judgement and choice: the same or not • another important issue: risk judgement and choice • the same or not? • no risk concept in EV models • risk attitudes follow from v and p functions
Theories of risk judgement risk aversion for gains risk seeking for losses
R–V Models – Markowitz: • decisions are based on both expected return and its uncertainty or variability (related to risk)(Markowitz, 1959) • risk is associated with the dispersion of the random variable • risk as indepedent concept WTP(x) = f {V(x), R(x)}
Risk judgement ≠ Choice • developed by Coombs • no clear answer
Research Questions Risk judgment Choice Do peopleuse the multiplicative or the dimensional pattern of information acquisition Relative importance of positive and negative dimensions Relative importance of values and probabilities • Do peopleuse the multiplicative or the dimensional pattern of information acquisition • Relative importance of positive and negative dimensions • Relative importance of values and probabilities
Experiment – Design • Subjects: • 120 respondents • Measure of perceived risk • subjects rated riskiness on an 11-point scale (from 0 ‘not risky at all’ to 10 ‘extremely risky’) • Measure of decision making (choice) • subjects chose one of three options 0 10 a) option A b) option B c) option C
Experiment – Design: scenarios • respondents were presented with 7 differentrisky situations related to financial risk, health hazards, gambling, etc. • everysituation consisted of 3 alternative options (A, B, C) • each option consisted of 4 possible outcomes - 2 losses and 2 gains and propabilities of those outcomes • participants could disclose as much detailed information about the options as necessary to judge their riskiness and to choose one of them
Experiment – Design: MouseLabWEB • the MouseLabWEB idea was to monitor the information acquisition process of decision making • information is hidden behind boxes – to access the information, the decision maker moves the mouse pointer over the box on the screen http://www.mouselabweb.org/
Results number of box average – 12 information after 6th information less systematic patterns checked first 6 steps
Results: information search patterns – Risk judgement 69,9%-due todimensional model 4,2%-due to multiplicative model 26% - without any model
Results: information search patterns - Choice • 67,5%-due todimensional model • 1,8%-due to multiplicative model • 30,8% - without any model
Risk judgement 69,9%-due todimensional model 4,2%-due to multiplicative model 26% - without any model Choice 67,5%-due todimensional model 1,8%-due to multiplicative model 30,8% - without any model Results: information search patterns
Results: positive/negative outcomes • positive/negative on top – biased • 2 display orders: • control: the same amount of information the same ratio pos/neg pos payoff … … … neg payoff neg payoff … … … pos payoff vs
Risk judgement ratio pos/neg M=0,95 amount of positive information M=7,04 amount of negative information M=7,62 Choice ratio pos/neg M=0,96 amount of positive information M=6,87 amount of negative information M=7,50 Results: positive/negative outcomes
Risk judgement ratio value/p M=1,30 Choice ratio value/p M=1,23 Results: value or p value ratio = p value = 1 < 1 > 1 p
Risk judgement 41% amount value=p 28,1% amount value>p 16,6% only value 11,2% amount value<p 3,1% only p Choice 47% amount value=p 24,8% amount value>p 12% only value 12,8% amount value<p 3,5% only p Results: value or p
Results: value or p for different situations • ratio value/p different for different situations • more p is considered for financial risk: investmenst and gambles • more value is considered for health hazards and extreme sports F(1,49)=0.117; p=.734 F(1,53)=5,475; p=.023 F(1,56)=0.612; p=.437
Conclusions: • the majority of information search pattern is due to DIM model (about 70%) • no differences in amount of considered infrmation between positive and negative outcomes • p more frequent for precise information (‘experiments’) values more frequentfor less precise information (‘natural setting’)