Why target-based bonuses encourage cheating, and how to motivate instead

Performance bonuses are often said to be a necessary evil. They are necessary to motivate and reward outstanding performance. But they can be evil as they can encourage cheating.

Warren Buffett once wrote: “Managers that always promise to ‘make the numbers’ will at some point be tempted to make up the numbers.”

What to do? A new game-based study addresses this very question. The study is “Are you paying your employees to cheat?” (pdf) written by C. Bram Cadsby, Fei Song, and Francis Tapon, at Ryerson college (via Meena Thiruvengadam at WSJ’s Real Time Economics blog).

The study expands on previous work that shows bonuses encourage cheating versus no bonuses at all. This study takes an extra step by comparing three types of performance structures and analyzing the results of each treatment.

The conclusion may indicate that other types of performance bonuses encourage less cheating and could be preferred methods of compensation.

Different kinds of bonus structures

There are many ways to structure a bonus and each creates different incentives.

In this study, three different incentive structures were analyzed: target-based, tournament-style, and linear pay for performance. They are roughly characterized as follows.

Target-based bonuses sound like this: “anyone who makes 50 sales or more this year will get a $5,000 bonus.” It’s the sales or production goal that defines this compensation type. Someone who sells just one item less and misses the goal will lose out on the bonus. Clearly there is an incentive to lie as one gets close to the target.

Tournament-style bonuses are very similar to target-based. The bonus is a “jump” at a particular level. The difference is tournament style bonuses are based on relative performance: “anyone who makes the top 30 percent of sales this year will get a $5,000 bonus.” There is certainly an incentive to lie about performance, but it’s harder because the goal is relative to other employees rather than a specific sales number.

Linear pay for performance structures are different and are graduated. They sound like this: “each sale will be rewarded with a $100 bonus.” The pay structure is graduated unlike  target-based or tournament-style. There is incentive to lie here too, as each misrepresented sale leads to profits, but it does not focus around a single point as in the other bonus structures.

Note the graph is only meant to be illustrative. It’s not true that linear bonuses pay the most, or that tournament bonuses are higher than target-based ones. The exact targets and bonus levels can vary.

The defining characteristics are the “jumps” in bonus for target-based and tournament-style, the absolute goal for target-based versus the relative goal for tournament-style, and the gradually rising slope of a linear pay-for-performance.

The study’s experiment word game

The study was a controlled experiment of over 200 students at Ryerson University. The students played various rounds of a word game of creating anagrams. At the end, they graded another subject’s work anonymously, and their own work was returned to verify correct grading. The subjects were then paid based on their performance and the bonus type.

The experimenters were careful to control for things like gender and language skills. That’s the basic idea, anyway, and you can read the paper for all the details if you wish. (link)

The results

The first result was that performance was similar regardless of the bonus structure of target-based, tournament-style, or linear pay for performance. The students were trying hard in each structure to create as many anagrams as possible, and their production was statistically at the same level. This is a good sign the experimenters set the bonuses at good levels so effort was comparable.

The second and important result was the evidence of cheating. Subjects demonstrably cheated more under the target-bonus structure compared to the tournament-style or linear pay for performance. There were a couple of ways they measured this, but I only want to mention one.

One measure was the number of overclaimed words, such as incorrect words. These are words that both the grader and the test subject let slip, i.e., a measure of cheating. The number of overclaimed words varied depending on the bonus structure. Not surprisingly, the highest was for the target-based bonus, which had over 2.5 times more overclaimed words than in tournament or linear structures

All in all, the study confirmed the perverse effects of target-based goals and suggested tournament-style and linear pay for performance bonus structures may be better.

Full study available at WSJ public docs (pdf)

What do you think?

I think this paper makes a great point about the problem of target-based incentive structures. It’s too obvious how to cheat, it’s too easy to find a partner in crime, and it’s too tempting to cheat.

Tournament bonus structures are a little harder to cheat at since the goal is relative. The problem here is people often sabotage other’s work to make their own relatively better.

Linear structures seem to encourage a bit more honesty and reward piecework. Sure, it is tempting to lie for a few sales more, but that temptation is there through and throughout rather than at a specific goal where people lie a lot and fudge numbers big time.

I am hoping there is more work done on this in a business environment to expand on this experimental setting.

I worked on a tournament-style bonus structure and often worried I would miss the mark. I would usually play a little bit of office politics around review time just to remind people of my good work. I would have preferred a linear structure so I didn’t have to waste this effort, but at least I didn’t have a target-based bonus which may have led to many ethical dilemmas.

What bonus styles have you worked under? Which one do you prefer? What do you think of this experiment?