Thinking fast and slow: A book summary (Part 2/5)
The various biases that our brain and thinking are prone to
In his book ‘Thinking fast and slow’, Daniel Kahneman throws light on human behaviour and rationality that helps us remove our biases and make fewer mistakes in our decisions.
Part 1/5 of the summary of this book (NOT a pre requisite)
In the first part, Kahneman talks about the two systems in our head that are responsible for most of our decision making — System 1 (One that is responsible for the mental events that occur automatically and mostly involuntarily — our instincts, biases and all actions that we do on an autopilot mode) and System 2 (One that analyses, doesn’t jump to conclusions and requires attention and effort)
The second part is about the biases we are prone to in our thinking and the mathematical basis behind seemingly counter intuitive observations
Heuristics and Biases
“A heuristic, is any approach to problem solving or self-discovery that employs a practical method, not guaranteed to be optimal, perfect, logical, or rational, but instead sufficient for reaching an immediate goal” — Exactly what our brain does!
Small Samples: Fooling us into drawing inferences that do not exist
“A study of the incidence of kidney cancer in the 3141 districts of USA reveals a remarkable pattern. The districts in which kidney cancer is lowest are mostly rural, sparsely populated and located in Republican states”. While you’d have rejected that Republican politics have any influence, you probably have reasoned out that low cancer rates are due to the rural lifestyle — no air pollution, no water pollution and fresh food.
Interestingly, if you look at districts with the highest incidence of kidney cancer, they tend to be rural too. If presented with this fact first, it’d be tempting to infer that high cancer rates are due to rural lifestyle — no access to health care, high consumption of tobacco.
But wait, the rural lifestyle cannot explain both! While System 1 always tries to find causation in the world around it, the key factor at play above is — rural districts have small populations, and extreme outcomes are more likely to occur in small samples. To understand why, let us get back to some quick probability calculations. Consider the example where there is a bag with half green marbles and half red marbles. Jack draws 2 marbles in a turn and Jill draws 3 marbles — and both of them record whenever they get a homogeneous sample (all reds or all greens). Jack will observe such extreme outcomes (all marbles of the same colour) more often than Jill — 50% vs 12.5% — just because his sample size is small. For someone drawing samples of 7 marbles, the probability drops further down to 1.56%
The same applies to the kidney cancer study, where rural districts are usually the ones with small populations (and small samples) and hence the higher likelihood of being in both ends of the spectrum of cancer incidence. This is all there is to the story, but System 1 needs a coherent story and tries to see causation even when it doesn’t exist. Similarly, consider the gender of 6 babies born in a sequence in a hospital –
Which of them is more likely to occur? The intuition keeps telling us that third sequence is most likely and the second is least likely, even when they are equally likely — 0.5 to the power of 6 or 1.56% This is because of the System 1’s obsession with causality and its need for coherence in a story. For it, regularities happen only with someone’s intention and it cannot accept that random processes can generate regularities once in a while. This is why one needs to exercise caution when judging the success of a new CEO or investment advisor too soon, before having a large enough sample.
Anchoring
In an experiment, there is a wheel of fortune that stops at either 10 or 65. After spinning, participants were asked to estimate the percentage of African nations among UN members. The mean of the estimate among participants who got a 10 was 25% vs 45% for those who got a 65. While the wheel of fortune has nothing to do with the question asked, the responses were ‘anchored’ by the number the participants got. This phenomenon when people consider a particular value for an unknown quantity before estimating the quantity is called ‘anchoring’.
Anchoring is regularly used in retail — when a high list price followed by a lower sale price, the consumer anchors the value of the product at the list price and believes that he is getting a great deal. Similarly when you enter a shoe store (with no initial idea on the price ranges), the price tag of the first shoe you see becomes the anchor. But the interesting (or rather sad) aspect is — anchors that are obviously random (like the wheel of fortune) can be as effective as informative anchors (price of the product)
Availability Bias
The availability heuristic is a flavour of other heuristics (substituting one question for another), where — in order to estimate the actual frequency of occurrence of an event, you substitute it with the ease with which the instances of the event come to your mind. When spouses are independently asked “How large was your percentage in keeping the place tidy/ taking out garbage”, the self-assessed contribution added up to more than 100%. With availability bias at play, both spouses remember their own individual contributions more clearly than those of the other, and this difference in availability leads to the difference in judgement. Similarly, when there are a couple of plane crashes in a month, people prefer to take the train. While the risk of flying hasn’t changed, there is an increase in the availability of flight crash instances, that increased the perceived risk of flying.
Surveys show that people judge death by accidents to be 300 times likely than death by diabetes, when in reality it is only 0.25 times likely. In this case along with availability bias (media always covers death by accidents and never covers a death by diabetes) there is also ‘affect heuristics’ at play — the emotional reaction to a risk (how you are affected by it) becomes a substitute to the harder question of probability of that risk. As accidents invoke stronger emotional responses in people compared to diabetes, people over estimate its likelihood.
This combination of availability bias and affect heuristics is extremely well used by terrorists. The number of casualties from terror attacks is relatively small compared to other causes (road accidents, diabetes, heart strokes), but they disproportionately impact public policy and budget spend.
Less is more
“Linda is 31 years old, single, outspoken, and very bright. She majored in philosophy. As a student she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations”
Which of these is more probable:
(a) Linda is a bank teller
(b) Linda is a bank teller and is active in the feminist movement
Contrary to logic, studies showed a large proportion of people chose (b) instead of (a). Termed as conjunction fallacy (judging a conjunction of two events to be more probable than one of the events), the reason is again System 1 that looks for more coherent stories, which need not be most probable, but are plausible (the story gets richer with the additional event)
Regression to Mean
An Israeli Air Force instructor commented “On many occasions I have praised flight cadets for clean execution of some aerobatic manoeuvre . The next time they try the same manoeuvre they usually do worse. On the other hand I have often yelled at the cadet for bad execution and in general he does better on his next try. So please don’t tell us that reward works and punishment doesn’t, because the opposite is true”
So, does that mean rewarding is bad and punishment is good? Not really. This phenomenon observed is called regression to mean which is caused by random fluctuations in the quality of performance. Look at it this way — the commander praises only the cadet whose performance is better than average (probably a lucky day), and it is likely to deteriorate in the next attempt irrespective of praise or punishment. Similarly, the cadet who gets shouted at performed below average (unlucky day) and it is likely to improve in the next attempt irrespective of praise or punishment. The commander here made a causal interpretation for the fluctuations in a random process on which he doesn’t really have a lot of control.
This is also seen in ‘Sports Illustrated Jinx’ — if a sportsman appears on the sports illustrated cover, he is doomed to perform poorly in the next season. Various causes like overconfidence, pressure of high expectations are offered. Similarly, a player in poor form is predicted to perform better as there are no expectations and the player is relaxed. While the causal account is good to hear, what is happening is a simple regression to mean, as seen below —
Evaluate the truth in the statement — “Depressed children treated with energy drink improve significantly over three month period” Here, depressed children are an extreme group (more depressed compared to most children) and would regress to mean over time, i.e. get better with time. The only way to establish causality in such cases is to have a control group which is given a placebo, i.e. splitting depressed children into randomly selected groups where Group 1 is given energy drink and Group 2 is just given flavoured water labelled as energy drink. If children Group 1 show significant improvement than those in Group 2, only then can we say that the above statement is true.
So the next time you read a catchy headline (New research shows chocolate helps in losing weight), ask what is the sample size of the study. Question if there really is causation. Similarly, whenever you seem to have a logical explanation for any event, ask yourself if there actually is an explanation or if it is just a random occurrence that you are reading too much into.