Neuroeconomics Lecture Notes PDF

Attention Principles of Neuroeconomics Prof. Dr. Todd Hare Premise: Values are constructed at the time of choice – The brain has finite resources – Resource allocations during this construction process will influence valuation and comparison processes (i.e., decisions)....

Attention Principles of Neuroeconomics Prof. Dr. Todd Hare Premise: Values are constructed at the time of choice – The brain has finite resources – Resource allocations during this construction process will influence valuation and comparison processes (i.e., decisions). Page 2 The brain uses efficient coding schemes – The brain uses information about regularities in the environment to represent stimuli and states efficiently. – Although efficient codes are the optimal solution under resource constraints, they can cause biases and imprecision in decision making. – e.g., Woodford, M. (2020). Modeling Imprecision in Perception, Valuation, and Choice. Annual Review of Economics, 12(1), 579–601. https://doi.org/10.1146/annurev- economics-102819-040518 Page 3 Efficient codes and risky choice – There are multiple forms of efficient coding used in the brain. – Codes that determine how we perceive numerical quantities depend on the distribution of quantities in the environment. – Perhaps surprisingly, this includes numbers written as words or shown as digits (e.g., one, nineteen, 3, 42, 12) – I.e., it is independent of the symbols themselves Page 4 Log10 Frequency Dehaene & Mehler 1992; https://doi.org/10.1016/0010-0277(92)90030-L Page 5  Log Frequency Dehaene & Mehler 1992; https://doi.org/10.1016/0010-0277(92)90030-L Page 6 Stimulus distributions influence the precision of efficient codes – The optimal representation with a fixed output range is proportional to the CDF of the prior stimulus distribution. – The exact relationship between the CDF and the efficient code depends on quantity being maximized (e.g., accuracy, information, reward) – The Frydman and Jin study discussed next use a uniform distribution so that the efficient code is identical for all 3 goals. Bhui et al 2021 Page 7 Testing resource-limited representation of numbers – Task = decide if the number on the screen is above or below 65. – The low volatility context has a uniform distribution from 56-74 – The high volatility context has a uniform distribution from 31-99 Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 8 Testing resource-limited representation of numbers – Assume we can distinguish only 19 magnitudes*. – This gives a precision of – 1 for low volatility trials – ~3.6 for high volatility trials *this is an example, we don’t know the true resource limit Page 9 Testing resource-limited representation of numbers – Performance is compared in the common range, 56-74. – i.e., the same comparisons, but different contexts – Payoffs are calculated as 15*accuracy – 10*average seconds – Incentivize accuracy and speed to increase the effect of resource constraints Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 10  Number comparison task Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 11 Accuracy is lower in the high volatility condition Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 12 Response times – RTs increase near 65 for both conditions – Behavior shows the ubiquitous discriminability or difficulty effect – RTs are longer with high volatility – suggests less precise representations Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 13 Risky lottery task Notation in subsequent slides Lottery payoff = X Sure payoff = C One random trial is paid out Page 14  Predictions Range of lottery payoffs Model predictions (qualitative) (p is 0.5) Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 15 Results Data Model predictions (qualitative) Frydman and Jin 2021; DOI: 10.2139/ssrn.3270773 Page 16 Selective attention – Attention is the process of flexibly controlling limited computational resources. – Selective attention refers to attending to some things more than others Page 17  Attention allocation changes decisions under risk Participants must click each box with the mouse to see the information Page 18 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  Attention correlates with Cumulative Prospect Theory parameters Page 19 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  CPT value function Page 20 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  CPT probability function Page 21 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  Attention indices in the mouse-tracking study – Attention_O = the median (across all gamble problems) of the time spent inspecting all outcome (O) information – Attenion_P = the median of the time spent on all probabilities (P) – Attention_LA = the median(across gamble problems with mixed gambles) ratio of time the participant spent inspecting loss outcomes (O-) relative to gain outcomes (O+) – i.e., O- / O+ Page 22 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  Attention is correlated with loss aversion, outcome sensitivity, and probability sensitivity Page 23 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  CPT parameters related to attention Page 24 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  Experiment 2, Does attention play a causal role? – Three attention manipulation conditions: – Loss attention (n = 40) – Loss outcomes open for 900 ms – All others 300 ms – Gain attention (n = 41) – Gain outcomes open for 900 ms – All others 300 ms – Control (n = 39) – All boxes open for 300 ms Page 25 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406  The duration manipulation changed looking times and loss aversion – Attention_LA – Loss-attention group = 2.82 – Control group = 1.00 – Gain-attention group = 0.34 Difference in lambdas between loss- and gain- attention = 0.140 (95% HDI [.008,.287] Page 26 Pachur et al 2018; http://dx.doi.org/10.1037/xge0000406 Summary – Loss aversion is expressed differently under different attention conditions. – Outcome and probability sensitivity are correlated with the amount of attention allocated to outcomes and probabilities, respectively Page 27 Gaze bias both reflects and influences basic preferences Which face is more attractive? Shimojo et al 2003; https://doi.org/10.1038/nn1150 Page 28 Indicate the disliked face instead of more attractive – Dots + solid line = indicate the disliked face – Dotted line = original curve for face attractiveness Shimojo et al 2003; https://doi.org/10.1038/nn1150 Page 29 Gaze also reflects perceptual decisions – Dots + solid line = face roundedness instead of attractiveness – Dotted line = curve for face attractiveness Shimojo et al 2003; https://doi.org/10.1038/nn1150 Page 30 Attractiveness of abstract shapes – Dots + solid line = shape attractiveness – Dotted line = curve for face attractiveness Shimojo et al 2003; https://doi.org/10.1038/nn1150 Page 31  Attention patterns change when choices change Day 1 Day 2 Day 1 vs 2 reversals Day 1 Day 2 Choices changed in 23.3% of trials Shimojo et al 2003; https://doi.org/10.1038/nn1150 Page 32 Causal manipulations of gaze Longer = 900 ms per repetition Shorter = 300 ms per repetition Page 33 Causal manipulations of gaze - The no gaze shift control manipulations are inconsistent with a “mere exposure” effect - An active shift of gaze (proxy for attention) is required. - Longer = 900 ms per repetition - Shorter = 300 ms per repetition - All controls have 6 repetitions Page 34  Visual attention and purchasing Krajbich et al 2012; https://doi.org/10.3389/fpsyg.2012.00193 Page 35 Krajbich et al 2012; https://doi.org/10.3389/fpsyg.2012.00193 Page 36 Does attention amplify value or have a constant effect on choices? – Many studies have shown that people are more likely to choose the item they fixated longer. – There is debate about whether the influence of attention is additive or multiplicative. – Multiplicative: θ * attended item value – (1 – θ) * unattended item value – Additive: θ + attended item value – unattended item value Page 37  Multiplicative vs Additive influence Page 38 Smith & Krajbich 2019; https://doi.org/10.1177/0956797618810521  Predictions for decision time from each model – Decision times are faster when options are more distinct – For perceptual judgements this means more perceptually different – Clearly bigger or smaller for example – For value-based choices this means more different in value – A multiplicative attention effect would amplify differences between sets of high valued options. – Multiplicative: θ * attended item value – (1 – θ) * unattended item value – Additive: θ + attended item value – unattended item value Page 39 Smith & Krajbich 2019; https://doi.org/10.1177/0956797618810521  Multiplicative vs Additive influence Multiplicative prediction Additive prediction Page 40 Smith & Krajbich 2019; https://doi.org/10.1177/0956797618810521  6 datasets all show a negative correlation consistent with a multiplicative effect Page 41 Smith & Krajbich 2019; https://doi.org/10.1177/0956797618810521 Attention effects may vary as a decision unfolds – Westbrook et al examined the effects of a common dopamine agonist drugs on the willingness to engage in cognitive effort. – selective dopamine D2 receptor antagonist (methylphenidate) – Found that baseline dopamine levels in the brain were correlated with willingness to engage in cognitive effort. – Methylphenidate increased willingness to complete a harder version of a memory task. Westbrook et al 2020; DOI: 10.1126/science.aaz5891 Page 42 Westbrook et al 2020; DOI: 10.1126/science.aaz5891 Page 43 Differences in dopamine correspond to fixation patterns Westbrook et al 2020; DOI: 10.1126/science.aaz5891 Page 44 Westbrook et al 2020; DOI: 10.1126/science.aaz5891 Page 45  – Attention has a multiplicative effect until the bifurcation point – Decision made, but not executed yet – Attention has an additive effect after the bifurcation point Westbrook et al 2020; DOI: 10.1126/science.aaz589 Page 46 Westbrook et al 2020; DOI: 10.1126/science.aaz589 Page 47 Multiplicative vs additive influence of attention – The debate is not fully resolved – There do seem to be multiplicative effects – They may be limited to early in the decision phase – Additive effects may kick in later once an option is (nearly) selected Page 48 Attention and goal-relevance vs value – Often, the goal is to select/obtain the most valuable option – What is the role of visual attention when you must identify the worst option? Page 49  Select best vs worst Sepulveda et al 2020; https://doi.org/10.7554/eLife.60705 Page 50  Select most vs least Sepulveda et al 2020; https://doi.org/10.7554/eLife.60705 Page 51  Attention is goal-relevant in value-based choices *the regression estimates the prob. of choosing the right option Sepulveda et al 2020; https://doi.org/10.7554/eLife.60705 Page 52  Attention is goal-relevant in perceptual choices *the regression estimates the prob. of choosing the right option Sepulveda et al 2020; https://doi.org/10.7554/eLife.60705 Page 53 Final fixations to chosen options C & D) Pearson correlation between gaze position and difference in value or dots at each time point in a trial Page 54 Summary – Attention is associated with goal-relevance – Goal-relevance does not necessarily equal value Page 55 It’s not just about the quantity of attention an option receives – We’ve seen examples of how looking longer influences choices – The order of information acquisition also influences choices Page 56  Information acquisition changes temporal discounting Participants must click each box with the mouse to see the information Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 57  Search strategies Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 58  Comparative searchers are more patient Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 59  Experiment 2: Manipulating search to test causality – Introduce a 1 second delay for discouraged searches – Easy comparative condition – Comparative transitions are immediate – Integrative transitions incur a 1 sec delay – Vice versa for easy integrative condition Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 60 Results for all trials – Easy Comparative show more patience than Easy Integrative Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 61 Results for trials in which search matched the intended manipulation – Recover the framing effect in comparative searchers Reeck et al 2018; https://doi.org/10.1073/pnas.1707040114 Page 62 The association between search and patience is task dependent – Choose between pairs of outcomes – One sooner – One later – Text is only visible if you look in a dotted rectangle – analogous to the mouse-tracking studies Khaw et al., 2018; https://doi.org/10.3389/fpsyg.2018.02102 Page 63 Payne index – PI = (Alternative – Attribute) / (Alternative + Attribute) – Higher PI = more integrative search Khaw et al., 2018; https://doi.org/10.3389/fpsyg.2018.02102 Page 64 – In contrast to the design in Reeck et al., now participants need to integrate sooner and later payoffs to compute the total amount of money. – Does this change the relationship between search patterns and patience? Page 65 In this context, integrative search is associated with more patience Khaw et al., 2018 Page 66 Integrators’ choices are more sensitive to total reward Page 67 Summarizing both studies – Reeck et al., show that information search patterns change temporal discounting – Khaw et al., show that the way information search patterns relate to temporal discounting depends on the choice context. Page 68 Example exam questions – The brain uses efficient codes A. Because they are easier B. Because they are faster C. Only when it is under time pressure D. Because they are optimal given finite resources Page 69 Example exam questions – If an option is fixated longer during a value-based choice, then A. It is more likely to be chosen B. It is less likely to be chosen C. It is probably bigger Page 70