Un Mesur i Reol Nhw Pob

We encourage interested readers to skim a very short essay on Mae ein Fframwaith Rheoli to gain familiarity with the way that we use certain terms in this essay. Wrth i amser yn caniatáu, we plan to post additional introductory and reference material.

A Common Performance Measure Fallacy:

″My subordinates should be rewarded the same way that I am. That will align their incentives with mine, and they will do what they are supposed to do.”

We often observe senior managers rewarding their subordinates on the same basis for which they, eu hunain, are held accountable. Our response in those situations is always the same: “why?” It’s a rhetorical question because such policies are almost never optimal, and seem to be selected and implemented due to a simplistic assumption that identical performance measures align interests despite the fact that subordinates may have very different jobs (and therefore signals of their performance) than their bosses.

Included in our criticism of overly-general performance measures is the common policy of rewarding relatively low-level employees based upon firm-wide profits or share appreciation, especially cases of granting employees restricted stock in the gobaith that the grants somehow, miraculously motivate employees to act in an incentive-compatible way.

The problem with such performance measures, and the reason that they are so inefficient is that most employees can’t determine which of their possible actions or decisions will maximize such global, diffuse, and noisy random variables, and the small effect they may have can be easily overwhelmed and hidden by events outside their control–kind of like spitting in the ocean.

In this essay, we explain how such fallacious beliefs can be very expensive.

Risk Aversion:

Before proceeding, note our working premise: individuals are risk-averse with respect to compensation.

That means that we can represent satisfaction or happiness as an increasing, ceugrwm swyddogaeth; in the graph below, boddhad, represented by the red utility function, increases as compensation does, but it increases at a decreasing rate. Felly,, more is better, but not proportionally better.

If everything else were equal, such an employee would prefer a certain, non-risky salary to the same ddisgwylir and risky level of compensation. Byddwn yn esbonio.

Consider two pay schemes: (1) a certain salary, a (2) a lower, base salary plus a chance at a bonus. (Suppose that the bonus is related to something like a sales goal, and the entire bonus is awarded when the goal is attained, and the goal is expected to be attained 50%…as you can see it is very easy to complicate the problem and lose sight of the concept.)

In our simple example, the certain salary is half-way between the the base pay and the base pay plus the bonus. Felly,, if the bonus is earned one-half of the time, Yna, ar gyfartaledd–if the situation were repeated many, nifer o weithiau–average pay would be the same under (1) a (2). (That makes the two schemes easy to compare.)

Receiving the salary would give satisfaction at the red dot. Under the other compensation plan, half of the time, the person would receive only the base and the other half of the time, he would enjoy satisfaction from getting both the base and the bonus. In the later state, the person would (yn amlwg) receive more satisfaction than from the salary, but the margin satisfaction from the higher compensation is less than the marginal satisfaction between the salary and the lower base. Felly,, and here is the very important part, if a risky compensation scheme pays the same as a certain one, then the person’s average satisfaction–which matters far more than pay levels–is lower than with a fixed salary.

Mewn geiriau eraill, for a given level of expected (but risky) pay, the firm could pay a lower salary and keep the person just as satisfied (on that dimension). To see that, draw a horizontal line from the x to the red utility function, and then draw a vertical line from that point to the horizontal axis. That value on the axis is known as the certainty equivalent of gamble, or the fixed amount that gives the same satisfaction as the risky scheme, and for risk-averse individuals, it has to be less than the expected value.

That’s important because if the firm can satisfy (and retain and motivate) the employee with lower, fixed pay, it can lower its expenses and earn greater profits.

Moesol Peryglon:

The problem is that when the employee’s effort is unobservable or the reasons for his decisions are unknown, it may be difficult or impossible to motivate him without providing incentive pay that varies with respect to one or several variables (that reflect whether he did what he was supposed to do). Huckleberry Finn said it best: “Wel, Yna,, Dywed yr wyf yn, beth yw'r defnydd chi dysgu i wneud yn iawn pan ei fod yn drafferthus i wneud iawn ac nid yw'n unrhyw drafferth i wneud o'i le, ac mae'r cyflog yn unig yw yr un fath?”

Now if effort is unobservable then any available signal of effort (worth considering) has to be noisy and mixed with uncontrollable factors to generate the signal. (Yn amlwg, “unobservable” means that you can’t see it, but it also means that you can’t measure another variable and then (perfectly) invert that variable to determine the effort or decision taken. (As we’ll see below, some signals are “gwell” than others.)

What is important to know is that if a variable does not perfectly signal effort, there is a change the signal is wrong; felly, the person may be erroneously punished–dweud, by having their bonus withheld–despite the fact that they did what they were supposed to do.

That error imposes risk on the person, ff, he did the right thing but may not be rewarded for it. As we saw in graph above, that imposition of risk hurts the personby reducing his or her expected satisfaction. Felly,, yr holl bethau gyfartal, they must be compensated for the risk, and that additional compensation is the risk premium or the average, additional pay required to provide the person with the same satisfaction as, dweud, a fixed, cyflog.

So varying compensation based upon some signal of effort, g, elw, cost control, sales volume, ac ati. makes sense only if the firm is attempting to motivate the employee to behave in a particular way when their effort or actions or decisions are unobservable. It’s costly, because the firm has to pay the risk premium, but it is worth doing if it (the imposition of risk) motivates the employee to take actions that, ar gyfartaledd, (1) maximizes his expected pay, a (2) increase expected profits more than the expected increase in pay increases costs.

Felly,, if monitoring the employee is too expensive or infeasible, the control problem that must be solved involves determining the best way to structure incentive pay to motivate the person to cooperate. Erbyn “best way” we mean the cheapest way to motivate the desired behavior, and that depends upon the environment, the available signals of the person’s efforts, and certain characteristics of those signals. It would be truly remarkable if the solutions to all incentive problems within an organization were the same, ff, pay everyone based upon, dweud, the appreciation in share price.

It’s not a complicated argument, but there are a variety of parts to it.

Isod, we discuss a few statistical characteristics, but there are others not mentioned here. (Hey, we have to have some secrets, dde?)

Desirable Characteristics of Performance Measures:

To find the best performance measures for a particular task, we advise using a relatively simple, five-step process. ¹ Fel arfer, those measures are variables or combinations of variables that (braidd yn) inform about the person’s decisions and efforts. Fel arfer, the measures are something that the person can influence, ond nid oes angen eu, and uncontrollable variables can be used, rhy, if by considering such variables, one can get a better estimate of the person’s behavior.

When we analyze a random variable’s suitability as a performance measure, two of the characteristics that we consider are its sensitivity a precision to the employee’s effort. (If they exist.)

• Sensitivity considers how the employee’s effort influences the performance measure; on average how will the variable change as the employee’s effort changes? Even with multiple factors, it may be possible to understand the relationship between the noisy variable and the person’s actions or decisions. Visually, one can think of sensitivity as the slope of line relating the expected value of the measure to the employee′s effort. Felly,, in a graph, a variable whose expected value changes faster with respect to effort (has steeper slope) represents a more sensitive measure than another with a flatter slope.

• Precision is the inverse of statistical variance or dispersion. It measures the accuracy of the variable as an indication of effort. (Presuming those statistical values exist for the distribution under analysis.) Precision can be thought of as the narrowness of the (probabilistic) range of values that the variable can take for a given level of effort. The higher the precision, the better the inference about the employee’s actual effort, or anything else, o ran hynny. (A perfectly precise measure is the same as observing the employee’s action.) Move your mouse over the graph below to see the difference between a precise (glas) and imprecise (red) measure.
  

The blue and red lines above represent the average or expected values of the two performance measures. The bell curves attempt to represent the third dimension, or the dispersion of possible values around the mean. That difference is that for the red one, unrhyw (unknown) level of effort generates–mewn rhyw ystyr–a wider range of possible outcomes or values than the blue one does. With everything else equal, that means that if one observes a value from the red signal, like the red dot, isod, then it is more difficult to determine if the person did what he or she was asked to do.

Felly,, if the superior is less sure about the person’s actions, Yna, (eto, all else equal) more risk is imposed on the subordinate, and if that subordinate is risk-averse, he doesn’t like that uncertainty and must be compensated for it.

Mewn geiriau eraill, yr holl bethau gyfartal, when the employee is risk-averse, increases in sensitivity and/or precision (of a performance measure) allow for more accurate inferences. More accurate inferences by the performance evaluator permits the payment of a smaller risk premium while still motivating the desired behavior.

Felly,, in most situations, it would be highly unlikely for one measure to be optimal for all employees because it is rarely the case that subordinates perform exactly the same tasks in exactly the same (risky) environment as the superior. Fel arfer, workers are asked to perform different tasks for which different signals that different degrees of sensitivity and precision (and other important characteristics) for the work they perform.

That means that using of a single gross (and less personally-informative measure) for all employees will either (1) increase compensation costs by making pay more risky (than it need be) a (2) destroy or reduce employees’s motivation to act cooperatively because they see less of a relationship between their efforts and the performance measure(â).

In that context, we analyze using overall corporate profits or share prices as performance measures.

Corporate Profit or Share Price as Performance Measures:

Consider corporate profit or share price change in terms of sensitivity and precision for a typical worker.

Yn awr, exactly how sensitive is corporate profit or share price to any individual′s effort? We′ll answer that one: not very. How will corporate profits change with respect to the actions of, dweud, one minion in human resources or accounting or data-processing or production or sales, o ran hynny.

How precisely does corporate profit or share price reflect an individual′s effort? Unwaith eto, not very, especially for lower-level employees.

Share prices for large corporations change almost continuously during a trading day (and are influenced by many exogenous factors). Can those changes be attributed to any single employee′s effort or lack of effort throughout the day? Over a month, a quarter, y flwyddyn? Yn, usually not. “Andy is drinking a soda, our share price is down because he is not working.”

Felly,, is there a compelling reason to reward employees on such general measures? Anything is possible, but one would have to tell a very special story about a very special environment to think that either macro measure informs about a particular worker’s efforts.

Corporate profit or share price may reflect upon the combined efforts of all employees, but it also depends upon environmental factors that are outside of their control and internal factors that are outside of their control–like the quality of senior management, the ability of product designers, the efforts of the sales force, ac ati. Each functional area can only influence a small piece of the pie, er-mwyn-siarad. Total profits are maximized when each of those areas meets its goals (or accomplishes it tasks) in the most cost-effective manner, and imposing excessive risk on risk-averse workers is not the way to either (1) minimize compensation for a particular level of effort, neu (2) maximize effort for a particular level of expected compensation.

Yn awr, before we describe the motivational problems associated with using the wrong measures, let′s consider a simpler analogy that avoids those messy human behavior and motivation issues.

Optimal Performance Measures for a Car:

Consider an automobile. Most of us are familiar with these devices used for swift transportation; presumably their purpose is to travel either a certain distance or at a certain rate of speed. Either purpose provides a natural performance measure: an odometer for distance and a speedometer for speed.

Suppose one′s purpose for operating an automobile is speed. From the control perspective of the driver—say, a race car driver—would one be satisfied with the speedometer as the sole measure of performance? “Don′t need a gas gauge, we have a speedometer; don′t need thermometer, we have a speedometer; don′t need tire pressure monitors, we have a speedometer… nauseum ad."

Unless the reader is particularly ornery, he or she should readily agree that the use of only a speedometer to measure the efficacy of each sub-function of an automobile would not provide enough specific feedback to allow the driver to maximize his speed or average speed over the long-term. “Hey, we′re going very fast right now There′s no reason to stop for gas!"

Such sentences seem extremely silly, yet they are analogous to a policy of ignoring individual signals of effort and imposing one’s own particular performance measure on subordinates and no different than rewarding all subordinates on, dweud, divisional economic profit when they may have responsibility for only costs or only revenues or for neither?[4. Yeah, yeah, yeah, any information economist worth his salt should be able to easily construct counter-examples, but that is clearly not the point of the essay.]

Felly,, if you want to get an indication of whether someone did what they were supposed to do, measure and account for (o leiaf) some variables in their particular (decentralized) amgylchedd, including ones that they control or influence.

Using overly-general ones impose unnecessary risk on subordinates, and if those subordinates are risk averse—and who′s not—that imposition leads to either higher expected compensation costs and possibly lower effort and the subsequent reduction in output or higher employee turnover. ²

Yn ogystal,, we’ve tried to explain in a non-technical way why artificially constraining the number of measures reduces transparency and the ability to control tasks and processes. (Unwaith eto, think of a car with only a speedometer or an “idiot light” that simply warns without describing the problem.) It is the basic notion behind responsibility accounting: is the employee responsible for revenues, Treuliau, both or neither? Mewn geiriau eraill, what do you want the person to do, and what signals are available about the efforts and decisions?

Behavorial Implications:

As yet we haven′t described the more damaging qualitative aspects of such policies. Here are a few behaviors that we have observed:

• Distracted employees with an unnatural fixation on share price rather than their task at hand.
• The sense of alienation that follows from these types of realizations.
  • “My job’s not important enough for anyone to care about (or understand or measure) what I do; felly, I get paid for what my boss does.” (And if he is a fool, why bother.)
  • “I’ll get paid when he (or she) gets paid; fel arall, it doesn’t matter.” This might include the follow-up: “So, I’ll spend my time sucking up. That will increase my chances of getting paid more than actual work will.”
• The sense of being forsaken that follows from the conclusion, “It doesn’t matter what I do, I can’t influence those things; felly, pam trafferthu?"
• The sense of the organization is arbitrary as in the following sentiment: “They only pick those things because they can manipulate them. That way, if they don’t want to pay us they don’t have to. They have eleven ways to win anyway, and we have none. Felly,, pam trafferthu?"
• Cases where employees will take greater risks because they think that no one is watching. (Remind you of any financial crisis?)

Through the years, we′ve heard many other similar comments and sentiments.

Felly,, where should one begin?

By asking: What do you want the employee do? A, how can you tell if they′re doing it? (Not perfectly, but probabilistically or statistically.)

To do that, one needs to understand their subordinates’ jobs, including how their jobs create value and what private information the subordinates might possess about the firm, the task, or the environment. If you need more help than that, please give us a call at 1.412.779.9028 neu cysylltu â ni. That is why we are here.

Gyda llaw, if after reading our essay you are still thinking: “We’re a team, we have a common goal and should a common measure,” then we recommend that you try coaching a sports team with that singular control mechanism. (Anyone who has ever followed any team sport knows that there are different positions and tasks and the performance of athletes in those positions are measured on the same basis, g, in baseball, the batting average of an outfielder is an important measure, but it is not for a pitcher. If such considerations are too onerous, then we suggest you reread the above essay.

Yn olaf, if you do agree with our analysis, then you might also wonder why many consulting firms sell a single performance measure as the unqualified best in every case. If one doesn’t (gyffredinol) exist within a single firm, how could it exist across the entire economy–across many firms? Our hunch is that what is best for those consultants, isn’t necessarily best for you or your subordinates or shareholders.

© Hawlfraint 2008 2010, Rwy'n gobeithio Ymgynghori.

---

Troednodiadau: