7  Incentives

In the previous part we considered how the principal can motivate their agent employee. That employee balances the marginal cost of their effort against the marginal benefit of the return to their effort.

However, experimental evidence shows that this picture of motivation is incomplete.

7.1 Experiment 1

Ariely et al. (2009) provided experimental participants with financial incentives to undertake a series of tasks that required creativity, concentration, memory and problem solving skills. Some participants could earn a bonus equivalent to about one day’s pay. Others could earn a medium sized bonus equivalent to about two weeks pay. And a third group could earn a bonus worth around five months of their regular pay. (The experiment was done in rural India, so although the amounts were relatively large for the participants, they did not break the experimental budget.)

The result was that those who received the small or medium level bonus did not differ much from each other in performance. But those who had a large bonus on the line were the poorest performers. They choked under pressure.

7.2 Experiment 2

Hedblom et al. (2019) created their own firm and hired actual workers to perform data entry tasks. Some workers were told about the firm’s corporate social responsibility in the job advertisement.

When advertised as socially orientated, the advertisements attracted workers who were more productive and produced higher quality work. This effect was driven by both a selection effect, in that it attracted a slightly different pool of workers, and a treatment effect on the workers in that it made them more productive. The social advertising also increased the number of applicants by 25%, which was the equivalent effect of increasing wages by 36%.

7.3 Experiment 3

List and Momeni (2021) hired workers on Amazon’s online labour market platform, Amazon Mechanical Turk, and asked them to complete a short task for payment. They were then exposed to various corporate social responsibility messages.

List and Momeni found that the use of corporate social responsibility messaging increased cheating - both the number of workers who shirk their primary job duty and the level of the shirking. The share of cheating was highest when the corporate social responsibility message was framed as a prosocial act on behalf of workers.

This cheating is consistent with a higher level of “moral licensing”, whereby good behaviour in one domain liberates us to behave unethically in another while maintaining our self-image as a good and moral person.

For a popular discussion of the paper, Dubner (n.d.) discusses the paper with List.

7.4 Experiment 4

Much of the behavioural literature rests on experiments with low or no stakes. This is somewhat different from that experienced in corporate decision making, where stakes for both the individual and firm can be large.

There is a dearth of evidence of whether cognitive biases are directly reduced in a high-stakes corporate environment. More generally, there is a general lack of high-stakes experiments, with most direct examinations of incentives being an investigation into the difference between no and small stakes.

An exception is a recent working paper by Enke et al. (2021), which examined the effect of high stakes on base rate neglect, anchoring, the failure of contingent thinking, and intuitive reasoning.

You have come across base rate neglect (in the context of the outside view) and anchoring earlier in this unit. The test of intuitive reasoning used the cognitive reflection test, with the following two questions:

  1. A bat and a ball cost 110 KSh in total. The bat costs 100 KSh more than the ball. How much does the ball cost?(Intuitive answer is 10, correct answer is 5).

  2. It takes 5 nurses 5 minutes to measure the blood pressure of 5 patients. How long would it take 10 nurses to measure the blood pressure of 10 patients?(Intuitive answer is 10, correct answer is 5).

(The currency is KSh as this work was done in Kenya.)

The test of contingent reasoning used the Wason selection task. One version used in the experiment runs as follows:

Suppose you have a friend who says he has a special deck of cards. His special deck of cards all have numbers (odd or even) on one side and colors (brown or green) on the other side. Suppose that the 4 cards from his deck are shown below. Your friend also claims that in his special deck of cards, even numbered cards are never brown on the other side. He says: “In my deck of cards, all of the cards with an even number on one side are green on the other.”

Unfortunately, your friend doesn’t always tell the truth, and your job is to figure out whether he is telling the truth or lying about his statement. From the cards below, turn over only those card(s) that can be helpful in determining whether your friend is telling the truth or lying. Do not turn over those cards that cannot help you in determining whether he is telling the truth or lying. Select the card(s) you want to turn over.

Which cards would you turn over?

(I am showing some of the particular tests here not because they are vital to the question of incentives, but rather that these tests appear again and again in the behavioural literature. It is worth becoming familiar with them.)

The headline of the experiment was that the large stakes did little to reduce the bias of the experimental participants, with the exception of the cognitive reflection test. Response times increased by around 40% with the large stakes, suggesting that the performance increase comes from a reduction in the reliance on intuitions.

As for why the experimental participants did not do better, the paper authors write that “some problems are sufficiently complex for people that the binding constraint is not low effort and reliance on intuitions but instead a lack of conceptual problem solving skills”

7.5 Experiment 5

Gneezy and List (2006) hired experimental participants to perform either data entry to computerise the holdings of a small library, or to undertake door-to-door fundraising. The advertisement for the position included the hourly rate they would be paid for the six hours of work, $12 and $10 per hour for the two roles respectively.

In both cases, a treatment group of employees was then told at the beginning of their employment that they would be paid a higher hourly rate of $20. The employees doing data entry boosted their effort in response markedly over the first 90 minutes, after which their effort became indistinguishable from those still paid $12 per hour. Similarly, the fundraisers who received the surprise pay increase raised much more money in the first few hours of the task, but after that were indistinguishable from those still paid $10 per hour.