20 November 2012

The Limits of “Carrots and Sticks”

When attempting to influence or guide human behavior in a certain direction the general belief is that there should be rewards for complying, punishments for not complying and a control mechanism aimed at identifying and punishing those who do not comply. In brief it is the “Carrots and Sticks” view. 

In economic terms this “Carrots and Sticks” system is called “The principal and agent problem”. To briefly describe it: there are two actors – the principal and the agent. The principal wants the agent to behave in such a way that the principal’s self-interest is met. At the same time the agent has its own self-interest which is different from the one of the principal. A very common example is the situation of an employee. In this situation the principal is the company (employer) and the agent is the employee. At least in theory, the self-interest of the employer (principal) is for the employee (agent) to work, while the self-interest of the employee (agent) is to not work, but still get the salary.

In order to make the employee (agent) work the company (principal) offers a reward – the salary – and at the same time it threatens the employee with a punishment (penalty) for not working. This penalty is applied if the employee (agent) is caught not working and for this a control mechanism is put in place. Since the control mechanism is not perfect or the employee is cunning there is a probability of getting caught (a probability smaller than 100%).  In some forms of the “principal and agent problem” the agent (employee) has an “outside option” which in our example could be unemployment benefit.

In this economic view the principal has a “production function” which determines what the principal gets if the agent works and the agent has a “utility function” which determines the benefits of the agent. In typical theoretical principal and agent problems these functions are known (don’t ask me how to do that in real life) and the questions are to determine the amount of the reward (salary), the amount of the punishment and the optimal probability of getting caught (calibrating the control mechanism).

I’ve presented this model of carrots and sticks using the example of a working contract, but there are countless examples of it or variants of it being used in day to day life. Take traffic regulations for example. There is no reward for respecting the regulations, but there are punishments for not doing so and there is a control mechanism and inherently probabilities of getting caught that applies these punishments.

From a rational thinking perspective this is the way human behavior should be guided in the desired direction. People as rational agents should be happy with the reward (if any), be afraid of the punishment and comply in order to not get the punishment.

This model is so profoundly embedded in popular culture that every time one hears about something undesirable going on the first reaction is to “Increase penalties (fines and years in jail)”, to “put more police on the street” (which means increase probability of getting caught) and so on. Similarly to encourage certain behaviors (financial) rewards are offered or increased. Just think how much people in top positions in the financial sector receive as bonuses…

The truth is that this model has certain validity in real life, but by far is in no way perfect or even adequate for being used exclusively to guide human behavior. One point of validity of the model is that rewards are useful in stimulating people to behave in a certain way. For example students study harder for an extra point in their final grade and probably would do not do so if there would be no reward - increase in the final grade.  Another point of validity is that fear is a very powerful motivator of human behavior. Fear is deeply rooted in our evolutionary past and we act in certain ways in order to avoid negative consequences.

The week points of the model are four-folded. First, it ignores a vital learning from prospect theory, namely that people underweight average and large probabilities. To put this in easier to understand language people perceive probabilities differently from the objective (numerical) probabilities. For example an objective probability of 40% is perceived (and subsequently used in judgment) as a 30% (subjective) probability. Similarly an objective probability of 90% is perceived as roughly 70% subjective probability. Certainty, namely a probability of 100%, is perceived as 100% subjective probability.

The implications for the principal and agent model are highly significant. The model is based on the agent knowing (or at least guessing) the probability of getting caught. If people underestimate objective probabilities it means that for the agent to perceive for example a (subjective) 60% probability of getting caught the actual (objective) probability has to be around 85%. Moreover, there is research done and or presented by Dan Ariely in his book “The honest truth about dishonesty” that shows that the probability of getting caught plays no real role in cheating behavior. Now, this could be taken with a grain of salt. No one (sane) would commit a crime in front of a police patrol. In front of a police patrol can be translated as probability of getting caught of 100%. Similarly if the getting caught probability is 99% the undesired behavior would be virtually inexistent.

In real life, however, most control mechanisms do not function with 99-100% accuracy, meaning that the probability of getting caught is usually not very high. In my opinion this probability gravitates around 50-60%. Adding to this that people underweight (underestimate) probabilities in human perception the 50-60% chance of getting caught is perceived roughly around 40%.

Second, the principal and agent model makes one major assumption, namely that the control mechanism is costless. In my opinion this is a major mistake for an economic model. In real life control mechanisms cost money and usually a lot of money. Assuming that the current control mechanism costs X Euros and gives an objective probability of getting caught of 40% which is perceived subjectively as about 30%, what would be the cost of increasing the perceived probability to 70%? Remember that the subjective (perceived) probability of 70% means an objective probability of roughly 90%. 

Changing a control mechanism from 40% to 90% accuracy usually involves huge costs. Moreover there is a question of availability of resources. For example “putting more police on the streets” implies that there are people who are trained to be police officers (or agents) available on the labor market… now this is quite hard to believe, right? Doubling the police presence on the streets of a city is not very easy unless one would agree to have police people who are not properly trained.

One might say that the costs of increasing the accuracy of the control mechanism would be covered by the increase in revenues from penalties. Of course, this is a reasonable idea, but there are several limits to it. Increasing the probability of getting caught is meant to make people comply and not actually pay fines. Moreover, penalties should actually be paid by the people who get them and this is not always the case. In addition to this, in some cases fines paid by wrong-doers do not go into the budget of the organism that enforces the control. For example in my home-country traffic fines (when paid) go to the central or local budget of the government and not to the traffic police budget.

Third, the principal and agent model is myopic when it comes to setting the amount (level) of penalty or punishment. The theoretical model takes into account the individual’s “utility function” when setting the level of the penalty. However, in real life this is virtually impossible since regulations are made for everyone and subsequent penalties applied are similar across individuals. For example traffic fines are set based on formulas that have as starting point the minimum wage in the country. This might be a good or adequate formula for countries where the differences between rich and poor are small, but in polarized societies the difference in the “pain” of loss for a “rich” and for a “poor” could be considerable.

It is true that prospect theory states that gains and losses are not relative to one’s wealth (as the normative expected utility theory says), but to one’s expectations. At the same time, paying 100 Euros as a fine has different consequences for a person who earns 600 Euros per month as for a person who earns 8000 Euros per month.

Fourth, the principal and agent model completely ignores other motivations that people have to behave in a certain way. Let’s take the example of a student who is studying hard for a better grade (reward) and that is afraid of failing the exam (punishment). Even if these rewards and punishments are in place, there are other motivations such as liking what she is studying or studying hard because her friends are studying hard and she wants to not be perceived as falling out of the group or she is doing so with the hope of getting the attention of her handsome colleague. Examples of motivations for human behavior are countless and usually they don’t include rewards punishments and probabilities of getting caught.

I would like to end this post in a more philosophical note. Apart from the pros and cons of the Principal and agent model as the proper way of inducing or guiding behavior, there is a deeper question here. Do we actually want to live in a world of carrots and sticks? Would we behave in a way only because we want to get the carrot and avoid the stick?

Like it?  Spice Up  Your Business

No comments: