PSYC10 Study PDF

Prediction corona virus to be infecting kids at a high rate, but in reality ist the share of infected cases not the number of cases - BE CAREFUL ABOUT WHAT DATA YOU SEE MAKE SURE ITS VALID AND FOLLOWS PARAMETERS AND ACCURATE DATA Correlation does not mean causation ever Decisions are forecasts of...

Prediction corona virus to be infecting kids at a high rate, but in reality ist the share of infected cases not the number of cases - BE CAREFUL ABOUT WHAT DATA YOU SEE MAKE SURE ITS VALID AND FOLLOWS PARAMETERS AND ACCURATE DATA Correlation does not mean causation ever Decisions are forecasts of the future Multiple reasons something could be doing well vs not well - Look for data on consumer demand, maybe they will buy more in march vs december The best way to establish causation is to run experiment - Random assignment is crucial E.x Changing the independent variable in this way changes the dependent variable this much Random Assignment: Randomly assigning participants to condition groups - Allows for causality to be established Random Sampling: Randomly sample units to include in the study - Allows for generalizability to be established \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ Chance is streakier than you think, we are usually wrong when it comes to chance Human-generated data is not streaky enough People can see the outcome of chance and think it will change by how many times a certain outcome is the result. Ex. if you flipped a coin 6 times and all were heads, you might think it is more likely that the 7th flip is tails since it is unlikely that 6 heads would come in a row Gamblers fallacy: People are less likely to bet on numbers that just won When we believe that success/fail rate is unchanging, the probability of failure seems greater after a string of successes - The belief chance will correct itself Chance does **not** even out Two approaches to forecasts: 1. This time is different: you take as much info on the topic and explain why this is a good reason to invest 2\. What Usually happens: Identify all similar funds with same time frame and see how they did Start with What Usually happens then adjust a little for the possibility that time time could be different. Testing whether you are in a high chance environment, correlate two measures of the same thing. If the correlation is greater than.7 then its stable IN HIGH CHANCE ENVIRONMENTS: 1. Do not incentivize outcomes, but instead incentivise other indicators of competence 2\. Dont choose between options based on recent past outcomes 3\. Prioritize accumulating opportunities over trying to choose the best one The sample size is crucial for accurate results: Less sample size less reliable knowledge \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ Good performers tend to decline, bad performers tend to improve When one side has extreme values, usually on average the other side has less extreme values\ \ Regression to the mean: If something is doing abnormally well or abnormally bad then it\'s bound to even out For a performance to be extremely good/bad luck has to be unusually good/bad Sports illustrated jinx: Athletes would think that once they are sponsored they do worse Simpsons Paradox: A relationship at one lvl of aggregation disappears or reverses at a different lvl of aggregation When looking at correlations beware of: Regression to the mean Selection (small range, selection distortion) Aggregation (Simpsons Paradox) Heuristics are mental shortcuts or rules of thumb - good enough Heuristics are prone to systematic error AVAILABILITY - When prompted with an option it biases your original guess / opinion - Info that came to mind recently or frequently - or are the focus of our attention ANCHORING - The estimates of unknown quantities are easily biased by what values they consider REPRESENTITIVENESS - Given the base rate of wanted group + how much they seem to be in that wanted group Egocentric Bias: We fail to realise what life is like from others who do not share our knowledge or perspective Hindsight Bias: Once an outcome has occurred, we overestimate the likelihood that we would have predicted that outcome in advance - - - Curse of Knowledge: When we have private info we expect the uninformed to behave as if they know what we know Overconfidence: Unwarranted confidence - - - Overplacement is the better than AVG effect, Overplacment happens on easy tasks and underplacement happens on hard tasks OverPrecision we often dont account for how much we dont know Knowledge in a domain doesn not easily or often translate into being able to make accurate predictions of uncertain events... But it does translate to unwarranted confidence in the ability We need good feedback to learn Feedback must be 1. 2. 3. In the real world precise data is difficult since the real world provides noisy feedback usually The Planning Fallacy: People underestimate how long it will take them to complete tasks - Even if you know you usually run late We Usually overpredict when planning Inside View: - - - Outside View: - - - Youre more likely to have optimism in predictions if you favour the outcome or happen to like one variable ; Although optimistic reasoning obeys, reasonable constraints If a claim is made and we have a BIAS: - - Seen particularly well when participants are given fake disease test and one group has it 50%:31% and other group doesnt have it 50%:50% meaning that some participants that "Have" the disease have trouble believing it while the other group has no problem saying they have no disease Seen with sports gamblers: Talk about wins much more than losses - - Self-serving Bias: unconsciously we do the thing that benefits us/client most - - - - -