Following up on Joseph's recent posts on corporate incentive plans that use rank to assign nice carrots and spiked sticks, here are a few disorganized thoughts.
1. The law of large numbers tells us that the distributions of random sample will converge on the distribution of the population as the size grows large. If you're assuming small, nonrandom samples will have nice distributions, this theorem is not applicable;
2. Most of these incentive schemes assume a linearity that neither experience or first principles support (I blame Heinlein). These relationships are not monotonic. A sufficiently strong penalty can, by prompting anger or panic or paralysis, produce the very behavior being penalized. A sufficiently large reward can and often does encourage people to game the system;
3. Complex systems can pretty much always be gamed. This is particularly true for systems where all parties are acting solely out of self interest with no emotional investment in the larger goals of the institution;
4. As incentives (positive and negative) increase, people will be more likely to come up with new behaviors that satisfy the letter of the conditions.
5. Even if we put aside the concerns in item 2, the potential for misalignment of incentives in peer interaction here is huge. When you interview potential co-workers, is it in your best interest to see the strongest candidate get the job? What about when a co-worker asks your advice on a modeling problem?;
6. From the management side this might even be worse. There are certain people managers are inclined to protect, either for personal reasons or because they full an immediate need in team. Sometimes a manager would rather hold onto an adequate SAS programmer even at the cost of losing a gifted financial analyst;
7. "Because that's how Jack Welch did it" has gotten to be a less and less convincing reason over the years.
The gaming piece is really tough to combat, which is why we often have professional managers who are supposed to try and use judgment.
ReplyDelete