People spend a lot of time talking about how to make good decisions. As a student of judgment and decision-making, I personally spend a lot of time thinking about what rational decision-making really entails. My personal stance is that human decision-making should attempt to optimize and maximize value with as little subjective influence as possible, in contrast to a more common belief that it is incorrect to dismiss “human” elements in decision-making as irrational or suboptimal.

Debates about constructs are pointless without defining one’s terms. However, the more I delve into the literature, the more I realize that there is a lack of consensus on what rationality really is. Part of this problem stems from the fact that decision-making now exists as a subfield of both psychology and economics. To simplify the distinction between these two fields, which both deal with human behavior, one can think of economics as describing normative behavior and being littered with unrealistic assumptions, and psychology as describing actual emergent human behavior. In both fields, rationality refers to the maximization of one’s utility curve, which mathematically refers to the differentiation of one’s utility function: the point where the slope of one’s curve equals zero.

So far so good. We have some agreement. The difference, however, lies in the assumptions economics makes about humans. Those who have taken any introductory microeconomics course should be familiar with the fundamental assumptions of economics, the most important of which include perfect information, stable preferences, and utility maximization. However, those who have put down their textbooks and have actually had interactions with human beings know that these assumptions are, at best, highly optimistic of the human condition. These assumptions are the economic version of “this rocketship will successfully land on Mars… given that the rocketship is a massless sphere in a vacuum with no friction.”

Now, where do we go from here? Again, my belief is that “proper” rational decision-making should approximate decisions given the presence of the various economic assumptions, which I will henceforth term “economically rational” in a very academically unrigorous way. An explanatory analogy can be taken from poker: the “fundamental theorem” of poker, as articulated by David Sklansky, states that:

“every time you play a hand differently from the way you would have played it if you could see all your opponent’s cards, they gain; and every time you play your hand the same way you would have played it if you could see all their cards, they lose.”

From a decision standpoint, this is saying that if your actions deviate from what your actions would have been given perfect information, you’re being suboptimal. Thus, to me, increase in decision quality should cause one’s decisions to approach economic rationality.

Logically, then, decisions can be placed on a range, with varying levels of “rationality.” I will conveniently borrow nomenclature from financial economics to label the different levels of rationality in a way that makes sense to myself, again in a very academically unrigorous way. To me, rationality can be described as ‘strong rationality,’ ‘semi-strong rationality,’ and ‘weak rationality,’ with each category describing different levels of optimization.

Importantly, this is a blog post, not a research piece, because at this point I would honestly rather commit seppuku than restrict myself with more academic rules and regulations. The purpose of this post is to explore different assumptions of rationality, and to examine how to optimize one’s decision-making under each assumption.

STRONG RATIONALITY

Herbert A. Simon, in one of the seminal papers on behavioral economics, describes the idea of “omniscient rationality,” which is what we previously described as “economic rationality.” Given perfect information on available options and the absence of any bias or subjective weighing function, calculating utility under the assumptions of strong rationality is simply Payout * P(Success). Additionally, this “strong” rationality is perfectly willing to examine things in the long term, and is largely indifferent to risk and variance. If given the choice between a guaranteed $100 and a 1% chance to get $20,000, under strong rational decision-making, one would always pick the latter — there is no risk aversion in strong rationality.

Strong rationality makes economists happy, as it is very easy to calculate. However, those of us who can overcome their social awkwardness enough to actually interact with other humans (e.g. not me) will immediately realize that strong rationality provides a very poor descriptive model. For starters, people don’t have access to perfect information. Information search is costly — if you’re shopping for the perfect peanut butter in a grocery store, every different jar you examine is both temporally and mentally costly. Additionally, people utilize heuristics and biases in their decision-making, a concept popularized by Daniel Kahneman and Amos Tversky. For example, people are generally risk-averse, and the element of variance inherent in a 1% success gamble is enough to discount the $20,000 to almost nothing for most people.

Let’s look at what happens when people do NOT have access to perfect information.

SEMI-STRONG RATIONALITY

Imagine that I am presenting a choice of lotteries to you, where option 1 has a 100% chance of giving $10, while the second has a 100% chance of giving $20. What would you pick? Clearly, lottery 2 entirely dominates lottery 1, so you pick lottery 2. But guess what, I lied! Ha ha! Option 2 actually has a 0% chance of returning $20! You get nothing, now SIT THE FUCK DOWN!

Was your decision irrational? I don’t think anyone would reasonably say so. You made the best choice you could have GIVEN THE AVAILABLE INFORMATION… which just so happened to be entirely false. It is possible to make an entirely wrong decision that is still rational. Rationality is dependent on the amount and quality of information one has, rather than the underlying system parameters themselves. Thus, in contrast to strong rationality, there is always a nonzero probability that the fundamental assumptions you have used to construct your utility curve are incorrect.

Before we go further, it would be useful to define what exactly ‘information’ entails. I read the first few paragraphs of Wikipedia’s article on Information Economics and decided I didn’t give a shit, so I’m going to pull some more definitions out of my ass. From how I see it, perfect information includes two components: accuracy of choice parameters, and completeness of choice set.

Accuracy of parameters refers to how correct one’s knowledge of odds of success and payout is. Our earlier example had incorrect parameters — the odds of success were said to be 100%, but were in fact 0%. On the other hand, completeness of choice set refers to knowledge of what options are available in a decision scenario.

Recently, I’ve been car shopping again, a process which I had hoped not to repeat and is making me want to gouge my eyes out with rusty spoons. Let’s say I go to a dealership. A salesman with slicked back hair, dripping with freshly applied pomade, wearing a comically tight suit pulls me over to three cars. He gives me detailed information on each, and the mystical leprechaun residing in my pocket confirms the accuracy of this information. I weigh the attributes of each, and then pick the best car out of the three. Was this the best possible decision? What if, hidden in the back, there is a car superior to each of these three, at half the price? The car dealer, paid by commission, simply neglected to make this option visible to me — in other words, my choice set is incomplete.

Essentially, semi-strong rationality is strong rationality where there is imperfect information, either in the form of inaccurate parameters, or incomplete information. Clearly, making decisions that are informed by incorrect or incomplete information isn’t optimal. Someone seeking to make the best possible decision under semi-strong rationality would seek to increase the accuracy of available information as much as possible, and then maximize utility using this information. While in many cases, one can actively seek information via the Internet or other sources, I believe that most of this information search happens passively and implicitly through a process known as Bayesian updating.

What is Bayesian updating? Using the golden standard of academic references, we get this definition:

In statistics, Bayesian inference is a method of inference in which Bayes’ rule is used to update the probability estimate for a hypothesis as additional evidence is acquired. -Wikipedia

Think back to our earlier inaccurate lottery. As you have no reason to doubt the accuracy of my statements, you do not consider the possibility of me lying. However, what if this lottery was repeated? Logically, you will weigh the stated outcomes vs. your own experiences. Realizing there is a discrepancy between what you think you know and your experiences, you will implicitly modify your beliefs on the probability of certain outcomes. This process is Bayesian updating, where one’s prior beliefs about the parameters of a system update in real time with experience.

Everyone has been implicitly conducting Bayesian updating since youth. Handing money to complete strangers is clearly stupid and has negative expected value. But as kids, we’ve seen our parents hand money to strangers thousands of times, and then five to ten minutes later, food is magically brought out to us! Our mental framework considering monetary transactions is thus built ground up: we don’t instruct children in barter economics, they learn for themselves by observation and Bayesian updating. When we give the cashier at McDonald’s $5 for a Big Mac, do we think that the cashier will disappear with our money? No, because throughout the countless times we’ve seen transactions occur at McDonald’s, no such thing has happened. Every successful transaction observed at a McDonald’s boosts our evaluation of P(Success), and over the course of our lives this information is seen as essentially perfectly reliable. Children learn about the value of money and the guarantee of businesses through witnessing transactions happen in real life.

So does semi-strong rationality provide a good model of human behavior? It’s definitely more accurate than strong rationality… but still doesn’t take into account human errors, heuristics, and biases! Which leads us to the third category: weak rationality.

WEAK RATIONALITY

Weak rationality adds one more element to the mix: human subjective preference. This is most clearly demonstrated through the temporal discounting problem, a commonly used tool to in judgment and decision-making research. Essentially, the procedure involves asking a participant if they prefer some amount of money now (say $100) versus some amount of money in the future (say $250 one year from today). Based on the previous models of rationality, one would consider the maximum interest rate offered, and use that to calculate the present value of $250 one year from today. As of 2013, the real interest rated of the United States as reported by the World Bank is 1.7%. The S&P 500 and NASDAQ have both increased about 1.3% year-to-date, while the federal funds rate remains at less than 1%. Using 1.7% as an optimistic discount rate, you should be indifferent between $100 today and $101.70 one year from today. But do you know anyone who would take such a deal? Probably not.

Before I move on, I’d like to briefly touch on the importance of distinguishing between state and trait determinants of utility. Some of you might be thinking, “but wait! Wanting $100 today is perfectly legitimate! You might need that $100 to fix your car or to pay the rent!” Of course, if a need for money in the short-term exists, then one’s utility curve will be shifted. If you’re poor and would therefore prefer $50 now over $5,000 next year in order to keep from being homeless, that’s perfectly rational. Poverty is technically a temporary state, as opposed to a personality-based trait. Therefore, to truly extract the element of temporal discounting that is due to human bias, we have to ignore all changes in utility function caused by state elements, which is unfortunately impossible in the real world. But this post is entirely a thought exercise, so whatever. The point is, if you’d rather take $50 now over $5,000 a year from now independently of any needs you might have……… please play poker with me.

Subjective biases influence one’s utility function, such that choosing $50 today over $5,000 a year from now really CAN be subjective utility maximizing for some weirdo out there. This is the essence of weak rationality: one seeks to maximize subjective utility. And after all this preamble, this is where the fundamental crux of my argument lies.

From my point of view, optimal decision-making given weak rationality involves minimizing the influence of heuristics and biases. Barring time constraints, in which case it obviously makes sense to use mental shortcuts, I believe that one should always seek to be as objective and formulaic in decision-making as possible. This stems from the fact that we, as a species, are pitifully weak at taking long-distance evaluative views. In constructing our subjective utility functions, we generally do not consider the subjective utility we expect to experience in the future, only considering present subjective utility. If one considers the net utility of having $250 far outweighs that of having $100, one’s subjective utility curve should shift, resulting in less temporal discounting. Indeed, studies have shown that when a long-term orientation is induced in participants (essentially making them consider the future), participants exhibit less temporal discounting behavior.

If utility curves are viewed as constant, we are stuck in a situation where despite being aware of one’s irrational impulses, we are unable to change them.

The opposing point of view would be to consider heuristics and biases as an integral element of decision-making. The goal, then, would be to maximize subjective utility under one’s utility function, whatever it may be. If you’re dumb and receive more utility from $50 today than $5,000 in a year, then it’s rational for you to take the $50!

But there’s no way this is the optimal solution. Sure, this is the utility-maximizing decision given your utility curve, but your utility curve fucking sucks! And herein lies the fundamental point of disagreement. This point of view takes utility curves as fixed, while I view them as yet another element of the system to optimize. Consider strong and semi-strong rationality. Optimizing semi-strong rationality approximates strong rationality by maximizing information. To me, subsequently optimizing weak rationality approximates semi-strong rationality by minimizing subjective bias. Both of these optimizations fundamentally change one’s utility curve. If utility curves are viewed as constant, we are stuck in a situation where despite being aware of one’s irrational impulses, we are unable to change them. That doesn’t make sense to me.

CONCLUSION

In summary, there are 2 components in optimizing decision-making.

1. Attempt to reduce influence of subjective biases, such that your decision-making approximates semi-strong rationality.
2. Maximize the likelihood that the information you have is both correct and complete, such that your decision-making approximates strong rationality.
3. Make the choice that maximizes expected utility under these new parameters.

Strong rationality assumes perfect information, as well as no subjective bias. To optimize under strong rationality, simply find the option that maximizes expected value: EV = P(Success) * Payoff.

Semi-strong rationality assumes imperfect information, but no subjective bias. To optimize under semi-strong rationality, attempt to perfect one’s information as much as possible, and then assume strong rationality.

Weak rationality assumes both imperfect information and the presence of subjective bias. To optimize under weak rationality, first minimize the influence of subjective bias, and then assume semi-strong rationality. Then, follow the steps for semi-strong rationality and assume strong rationality.

To me, this model is sensible. Of course, there are limitations: one elephant in this room, which was briefly mentioned previously, is that the marginal value of finding new information is eventually outweighed by the marginal cost. This means that when trying to optimize semi-strong rationality, there is a point where examining every last option really isn’t paying off. This gets into the distinction between maximizing and satisficing as styles of decision-making. Current research shows that maximizers tend to make better decisions… but satisficiers tend to feel less anticipated regret and stress regarding these decisions. Which of these is more optimal? It’s still something we don’t have an answer yet.

Of course, I am not claiming that my stance is irrefutably correct. There has been evidence to the contrary — research has shown that damage to the orbitofrontal cortex, which impairs processing of emotional information, weakens one’s capability to make proper judgments, due in part to the inability to experience anticipated emotions (e.g. anticipated regret). Emotions are definitely a crucial part to decision-making. However, I still believe that the influence of emotions should be minimized if the goal is to optimize and make decisions that are as close to normative as possible.

Anyways, this is about it for my massive theoretical word-vomit. I usually tend to fire and forget my posts, but I might go back and edit this to account for mistakes and to add clarification. Any thoughts would be appreciated!