Principal-Agent models of the Employment Relationship and Performance-Related Pay by P. A. Lewis Introduction 1 Performance-Related Pay as a Solution to Adverse Selection & Moral Hazard 2 The Costs of Performance-Related Pay and the Optimal Intensity of Incentives 5 Absolute Vs. Relative Performance Pay 7 Limitations of the Theory 9 Conclusion: Performance Pay and the Modernisation of the Public Sector 14 Lewis, P.A. (forthcoming). ‘Principal-Agent models of the Employment Relationship and Performance-Related Pay.’ In N.F.B. Allington and J.S.L. McCombie (eds.), The Cambridge Student Handbook in Economics. Cambridge: Cambridge University Press. Introduction Examines how private-sector agents have tried to develop solutions to some of the problems created by asymmetries of information between employers and their (actual or potential) employees o Asymmetries: employers’ inability to observe characteristics and/or actions of workers Examines merits and limitations of performance-related pay DEFINITION: “performance-related pay can be defined as any linkage between an employee’s earnings and any measure of his or her performance on the job” (p.1) o E.g. tying pay to output, such as piece rates for manufacturing workers; bonuses that reward footballers for scoring goals; commissions; and pay and profit linking, such profit-related bonuses or share-options DEFINITION: “A principal-agent relationship is a situation in which one person (the ‘agent’) is called upon to act on behalf of another (the ‘principal’) and advance the latter’s goals.” (p. 1) o DEFINITION: “The principal-agent problem: the general problem of designing contracts which ensure that the principal’s goals are pursued as efficiently as possible.” (p. 1) 2 types of problems in principal-agent relationships that prevent effective contracts 1. Adverse selection: pre-contractual opportunism; hidden information that affects the net benefit to the principal e.g. workers know more about their ability than employers low ability workers will portray themselves to be higher ability to get a better job 2. Moral hazard: post-contractual opportunism; hidden action that affects the principals when agents pursue their own self-interest and deviate from the principal’s preferred course of action e.g. workers might prefer to shirk their duties than work as hard as the employers wish 2 conditions for adverse selection and moral hazard to occur: 1. Asymmetries of information between principal and agent e.g. workers know their own ability and level of effort but there is no way the employer can prove to a court that workers have failed to fulfil their contractual obligations IF information was symmetrically distributed and employers could observe and verify (to a court of law) workers’ attributes and actions then employment contract would specify precise attributes workers are required to have, actions they’re meant to take and the wage given in return therefore an enforceable contract 2. Divergence of interest between principal and agent e.g. employers want to hire hard-working people but the workers have incentive to exploit their private knowledge of their abilities by exaggerating or shirking their duties and putting less effort than their employer wishes IF interests were aligned, the agent would automatically do what the principal wanted because it would be in their interest to do so Principal-agent problem of the principal (employer) hiring a worker (agent) who is meant to work on behalf of the principal to work as efficiently and effectively as possible o PRP meant to create an employment contract that will: 1) attract workers and 2) motivate employees to pursue employer’s goals in a diligent fashion Performance-Related Pay as a Solution to Adverse Selection & Moral Hazard Principal-agent theory assumes employers can observe some (imperfect) measures/indicators of worker talent and conscientiousness o E.g. employers can observe level of output/ sales made by workers and profits generated (positively correlated with worker ability and effort) Measures are imperfect and inaccurate because above indicators of performance are affected by other factors o E.g. worker output can be affected by equipment breakdowns and profit is affected by managerial talent and effort, competitors’ actions, variations in input availability and price, changes in governmental legislation and consumer tastes Despite variations in indicators they are reasonably correlated with worker effort and ability and provide some information about talents and actions of workforce o Act as observable proxies for unobservables like effort and ability Pay schemes that link workers’ earnings to performance enables the employer to link the interests of employers and workers A well-designed performance-related pay scheme will sufficiently compensate the worker for the disutility of extra effort they put in to work harder/produce more o E.g. compensation schemes that link CEO’s pay to firm profits can be used to align shareholders interests and CEO’s interests Rationale for performance-related pay schemes is that they promote goal congruence, where an appropriately designed scheme changes workers’ interests so that the rational thing for a utility-maximising worker to do is to behave in the way desired by their employer Edward Lazear’s (1999) diagrams illustrate how PRP can overcome adverse selection and moral hazard o Horizontal axis: effort exerted and level of output produced by worker o Vertical axis: level of pay received o Employers assumed to be risk-neutral; workers assumed to be risk averse Diagram (a): The initial situation (fixed wage) Fixed wage = w; employees paid (w) just by turning up to work and exerting bare minimum of effort (e0) Workers’ budget constraints start at the origin (O), moves along horizontal axis to eo, rises vertically to A and stretches right horizontally Workers dislike higher effort but want higher pay, making the direction of all workers’ increasing preference to the northwest corner of the diagram 2 kinds of worker o High ability represented by ICH1; relatively flat indifference curve which represents the fact that they need relatively small increase in income to compensate for extra effort o Low ability represented by ICL2; relatively steep indifference curve to represent the fact that putting in extra effort costs the lowability workers a lot of disutility and must be compensated with large increases in income Where a fixed wage is offered, both utility-maximising workers will choose point A, using the bare minimum of effort (eo) o Point A is the furthest northwest point that both types of workers can reach Diagram (b): The impact of performance-related pay (piece-rates) Guarantees minimum wage (W) for every worker but additionally, workers can earn a bonus of £b for every unit of output produced over level of output q* o Compensation = max [W, bq – K] K = constant term included to ensure that the agent’s individual rationality constraint (of which more later) is satisfied Budget line starts at the origin (O) and moves along horizontally to eo, vertically to point A, stretches horizontally to the right as far as the level of output q* and then rises (blue line) to the northeast as bonus payments are introduced Low-ability workers’ behaviour remains unchanged because even after the change in budget constraint through the bonus payment scheme, the point furthest northwest that a low-ability worker can reach stays point A on ICL o Disutility of the effort required to produce more output exceeds the utility gained from extra bonus payments Change in budget implies high ability workers can reach a higher indifference curve (red curve) because bonus payments compensates them for the disutility of extra effort of producing above q* New equilibrium for high-ability workers is point B where their indifference curve ICH2 is tangential to the upward-sloping portion of the new budget line Effort has increased from eo to eH2 and pay has increased from W to WH2 with utility increasing from UH1 (on old indifference curve) to UH2 (new indifference curve) This particular pay scheme can be considered incentive compatible because the bonus payments are sufficient to ensure that it is in the interest of high-ability workers to increase their level of effort (Lazear 1999: 212-15) PRP can help overcome moral hazard by inducing a higher level of effort from workers; high-ability workers would want to put more effort in rather than in a fixed-wage scheme PRP can also overcome adverse selection; bonus payments can lead to the increase of utility of high ability workers from UH1 to UH2 o Some high-ability workers would be more willing to accept a job with this firm than if they used a fixed-wage scheme because they can receive a higher level of utility than UH1 at any other firm o Attracts more high-ability workers as they are being paid more (wage WH2) at this firm In other words, workers whose reservation utility lies between UH1 to UH2 (that is, workers whose reservation wage lies between W and WH2), and who were therefore initially unwilling to join the firm, now find it in their interests to join this firm’s workforce – their participation constraint is now satisfied - raising the average quality of the firm’s workforce and thereby helping to overcome the adverse selection problem (Lazear 1999: 214-15; Prendergast 1999: 14-15) The Costs of Performance-Related Pay and the Optimal Intensity of Incentives Performance-related pay schemes leads to an inefficient distribution of risk compared to the first-best, full information outcome o Employers must pay workers a higher expected wage under PRP schemes than under a fixed-wage system to compensate for the extra risk employees must bear Where employers assumed to be risk-neutral and workers assumed to be risk-averse, optimal distribution of risk would involve the risk-neutral firms bearing all the risk by giving the risk-averse workers a fixed wage and bearing no risk to themselves o However, if a worker’s wage is fixed they have no incentive to perform better than they need to because it’s not attached to their performance o But under PRP schemes, because performance measures are affected by external factors that lie outside the control of workers, they are exposed to higher risk variations in their income Workers must bear some of the risk, leading to inefficient distribution of risk compared to first-best, full information outcome When effort is not observable, there is a conflict between the goal of providing incentives and the goal of ensuring an efficient distribution of risk o The only way that it is possible to provide the worker with the incentive to work hard is by relating his pay to actual output, the level of which is affected by random factors that lie outside of his or her control Because PRP schemes can expose risk-averse workers to risk, it has important consequences o The risk that PRP exposes risk-averse workers to will reduce their utility When a worker is risk averse he or she prefers a wage of £x for certain compared to a risky wage that has an expected value of £x o Risk-averse workers are worse off with a risky wage than if they would be receiving £x for certain To ensure the workers’ participation constraint is satisfied, employers wishing to adopt PRP schemes must pay workers a higher expected wage under the PRP scheme (which is riskier) than under the fixed-wage system (riskless) (Prendergast 1999: 8; Mas-Colell, Whinston and Green 1995: 486) o Higher expected wage is the cost to the employer for using PRP schemes o Employer hopes to recover the cost through higher profits expected through workers working harder/producing more/being more motivated o A profit maximising employer will use PRP to extract extra effort to the point where marginal increase in expected revenue from higher output equals the marginal cost of higher expected wages to compensate for risk-averse workers’ disutility of exerting more effort (Gibbons 1997: 3; Burgess and Ratto 2003: 286) If we are thinking about situations in which effort is a continuous variable that can take any one of many values, then the level of effort which it is optimal for a profit-maximising employer to induce obtains when the marginal increase in expected revenue due to the extra output yielded by a slightly higher level of effort is just equal to the marginal cost of the higher expected wages required to induce that effort (pg. 12) The term ‘intensity of incentives’ denotes the rate at which at which a worker’s pay increases with the level of output he or she produces (pg. 12) o i.e. the higher the value of b in the expression for the worker’s compensation (graph) and the more the more the worker’s pay goes up as his or her output increases The optimal intensity of incentives b* occurs when the marginal benefit to the firm of making the worker’s incentives a bit more intense (i.e. the extra profits that the firm expects to earn if the worker puts in a bit more effort) equals the marginal cost of the increased intensity of incentives (i.e. the expected increase in the worker’s wages after the implementation of performance-related pay) (pg. 12) The following factors influence the optimal intensity of incentives (Milgrom and John Roberts 1992: 221-26; Gibbons 1997: 4): 1. Profitability of extra effort: the optimal intensity of incentives b* is greater when the contribution of the worker’s extra effort is expected to increase employer’s profits o The more that extra effort is expected to add to the employer’s profits, the employer is willing to incur greater costs to obtain that effort thus the greater is the optimal intensity of incentives, ceteris paribus 2. Agent risk aversion: the optimal intensity of incentives b* is lower when the agent is more risk averse o The agent suffers higher disutility when their income is unpredictable due to the variations from performancerelated measures and it is more costly to compensate the worker to bear the above risks; thus the more costly it is to generate incentives, the lower the optimal intensity of incentives is, ceteris paribus 3. The accuracy with which the worker’s effort can be estimated: the optimal intensity of incentives b* is lower when the variance of the estimate of worker effort is higher o The less precise a worker’s effort can be measured, the optimal intensity of incentives is lower; inaccurate, noisy, high-variance estimates of worker effort impose higher risks to workers and consequently increases the cost of compensating the workers for bearing the extra risk (relates to point 2) 4. The rate of change of the marginal cost or disutility of effort to the worker: the optimal intensity of incentives b* is lower when the faster the disutility of effort increases as the worker’s overall level of effort increases o As the disutility of effort increases when the worker’s overall effort level rises, employers must increase the expected wage (and therefore the cost) offered by the employer to compensate the worker for the higher level of effort/disutility; the faster the expected wage must grow, the lower the optimal intensity of incentives is ceteris paribus Absolute Vs. Relative Performance Pay Workers’ risk-averse nature and that they must be compensated with higher expected wages implies that profit-maximising firms will try to minimise the costs of implementing PRP to generate the particular intensity of incentives o Use performance indicators that measure effort as accurately as possible Informativeness principle states the following: in designing their compensation formulas, employers should make their workers’ pay depend on any performance indicator that (with appropriate weighting) reduces the overall error with which the employee’s level of effort is estimated o The employer will be able to minimise the costs of offering his or her workers incentives of a given intensity (Holmstrom 1979; Milgrom and Roberts 1992: 214-21; Prendergast 1999: 12-14). Consider the equation w = + (q + y) which describes a linear compensation scheme that makes a worker’s pay (w) dependent not just on his or her level of output (q) but also upon a second measure of performance (denoted by the letter ‘y’) o The Greek letter phi () in this expression is a parameter which denotes the weight accorded to the second performance measure ‘y’ in the formula by which workers’ pay is determined o If is set equal to zero, then the second performance indicator carries no weight and the performance-related component of worker pay is determined solely by the first measure of performance (q) or if it is not set to zero, the second performance measure plays a role in the worker’s pay What weight (if any) should be accorded to the second performance indicator (i.e. what should the value of be)? (Use informativeness principle) Informativeness principle suggests that the value of the weighting factor should be set to minimise the variance of the estimate of worker effort o Doing so will minimise the risks that workers are exposed to, given that it is necessary to give them incentives to work hard, and will therefore minimise the costs of the extra pay that the employer must offer the workers in order to compensate them for bearing risk It can be shown that the weight * which it is optimal to attach to the cov( , y) second measure of performance is given by the equation * = var(y) o cov(θ, y), the covariance of θ and y, is a statistical measure of how θ and y vary together (Milgrom and Roberts 1992: 246) The relation between the second performance indicator (y) and the random factor () which, along with effort, influences a worker’s output If cov(θ, y) 0, so that there is a non-zero correlation between the random variable θ and the second performance indicator y, then a knowledge of y does provide the employer with information about the value of the random variable θ If * is negative then, according to equation (2), high positive values of the second performance indicator (y) will tend to reduce the worker’s pay, while negative values of y will tend to increase the worker’s pay Should employers use absolute or relative (in comparison either to each other or to workers in other firms) measures of performance? o Example: worker is a manager Absolute measure: the level of his own firm’s profits viewed in isolation Relative measure: firm’s level of profits relative to the average level of profits across the industry as a whole 2 possible cases: 1) when the random shocks which impinge upon this industry are pretty much the same across all firms 2) where different firms are affected by different shocks, so that few if any of the random disturbances are common to all firms 1. If most of the random shocks are common to all firms in the industry, then those shocks may well significantly affect the absolute levels of performance of all firms, so that the absolute measure of performance is likely to provide a relatively inaccurate, high-variance estimate of worker effort o But if all firms are facing the same shocks, the random effects won’t have much impact on firms’ performance relative to one another o In this first case, differences in the relative performance of firms are likely to reflect differences in the levels of effort put in by their managers o In such circumstances, a relative measure of managerial performance, such as how one firm’s profits compare to the industry-wide average, will provide an accurate, lowvariance estimate of managerial effort 2. If firms are affected by different shocks, differences in relative performance of firms can reflect differences in the random factors that impinge upon them rather than variations in the levels of the effort put in by the managers o In this case, a relative measure of managerial performance such as our second performance (y) will provide an inaccurate, high-variance estimate of managerial effort because it will incorporate, rather than filter out, the noise from random effects o In situations like this, where the relative measure of performance (y) has a high variance, the weight attached to it in the optimal compensation scheme will be low and pay will depend more on the absolute measure of performance (Prendergast 1999: 14). Limitations of the Theory Explicit performance-related pay schemes are far rarer than principalagent theory would at first glance suggest The schemes which do exist seldom account for a significant part of a worker’s compensation Jensen and Murphy (1990) studied the relation between pay and performance in 1,300 US corporations between 1974 and 1986 o Empirical relation between pay and performance was economically insignificant, with CEO wealth changing by just $3.25 for every $1,000 change in shareholder value Link between pay and performance weaker than principalagent theory predicts; provides very weak incentives for managerial effort Evidence for the UK is provided by Cully et al. (1999), who found that only one in ten non-managerial employees has individual PRP More enjoy the benefits of Share ownership (one in seven) and profitsharing (one in three), but they remain very much a minority Reasons why predictions of principal-agent theory concerning PRP are flawed centre on the fact that the principal-agent models considered make a number of rather unrealistic assumptions concerning: (i) the nature of the jobs that workers perform; (ii) the information available to employers; and (iii) the motivation of workers i) Nature of the jobs workers perform a) Multi-tasking o Has been implicitly assumed that effort is a simple, onedimensional measure of how ‘hard’ a worker works o But most jobs entail a variety of tasks, which include a number of activities/dimensions where a worker can exert effort E.g. workers on a production line can devote effort to quantity and quality of their output; teachers are meant to pursue numerous goals such as teaching, instilling intellectual curiosity etc. o Employers should ensure workers are allocating energy appropriately between the different dimensions of their jobs o Problem of multi-tasking analysed by Holmstrom and Milgrom (1991) E.g. There are 2 dimensions of a person’s job and both are substitutes for each other, where the time and effort devoted to each activity cannot be devoted to the other; assumed that the employer cannot observe how workers allocate their time Holmstrom and Milgrom argue that if an employer bases workers’ pay conditional to the activity where performance is measurable, a rational worker will put all of his effort into that activity to ultimately maximise his pay o When pay is dependent on easily-observable performance indicators, it can lead to the neglect of a desirable but less easily observable goal E.g. workers on a production line whose pay focuses on piece rates would be tempted to merely focus on the quantity aspect of their jobs and neglect quality E.g. teachers are meant to pursue numerous goals; while they aren’t mutually contradictory, given the finite resources they compete for time and attention and are not precisely measurable If teachers were paid solely according to the examination results achieved by their pupils, they would have an incentive to concentrate exclusively on equipping their students with the narrowly defined basic skills that are tested in standardised examinations, and to neglect the other aspects of their job just because doing so would be the most effective way for them to maximise their pay o The general point made by Holmstrom and Milgrom is that where there are multiple tasks associated with a particular job, incentive pay serves not just to allocate risks and to motivate people to work hard; but also to determine how workers allocate their time and effort among the various duties associated with their job o When there are multiple dimensions to a job, inappropriately designed performance-related pay schemes can lead to dysfunctional behavioural responses or ‘gaming’ People substitute effort away from those aspects of their job for which they are offered the least intense incentives and towards those activities for which the incentives are most intense Gaming, then, might be defined as consisting in strategic responses to performance measures that increase performance outcome but result in costly misallocations of resources o Equal-compensation principle: if a worker’s allocation of time between two different activities cannot be monitored by the employer, then either the worker must be offered the same intensity of incentives for each of those activities, or the activity from which the worker receives the lower expected marginal rate of return for her efforts will be completely neglected Illustrates the constraint on PRP that multi-tasking imposes o When a worker has 2 dimensions in their job, where the first is less accurately measured than the other aspects of the job o If the employer wants the worker to put effort into both dimensions of her job, she must reduce the incentives offered for the second, more accurately measured activity below what would be optimal if that activity were the only aspect of the worker’s job until the incentives have the same intensity for both jobs. Only then will a rational worker devote time and effort to both activities o o o o o When performance cannot be measured at all in an activity that the employer wants the worker to devote time and effort into, they should not offer any incentive pay If one dimension isn’t measurable, you cannot offer positive incentives to those who work hard at it; if you offer positive incentives to another aspect of their job where performance is measurable, then you give the worker incentive to allocate all their time and effort to that one dimension Holmstrom and Milgrom (1991): inability to measure workers’ performance in some aspects of their job can undermine the case for PRP to obtain a high level of effort in other dimensions of the job Can explain why PRP is not used as much as principal-agent theory predicts Employers fear “gaming” and dysfunctional behaviour responses when jobs have multiple activities b) Team production o Workers may have to contribute efforts into a joint product which makes the output/contribution of the individual harder to distinguish o An individual will not have a well-defined marginal product in team production o Employers could use group incentives Making each individual’s pay dependent on the group’s output would be beneficial because it can provide incentives for individuals in the group to work hard themselves and gives them the incentive to monitor one another Team members are better placed than managers to monitor each others’ efforts and would be able to deter shirking by threatening those who do not pull their weight (e.g. ostracising them) o Size of group limits the usefulness of group incentive schemes Free riding: in a group of “n” workers, where an individual worker bears the disutility of an increase of their effort, he or she only gains (1/n)th of the bonus generated because the group bonus is shared by all “n” members Where “n” is larger, the rewards received by an individual for working harder are smaller compared to the costs of extra effort, therefore causing free riding on the effort of their team members’ effort o Not always easy to find clear, viable and meaningful targets/standards for teams especially when there is no tangible output ii) Information available Informational problems that employers face associated with PRP are more severe than principal-agent theory supposes o Principal-agent theory is built on asymmetric information where they assume that employers are unable to observe the worker’s level of effort but they assume that employers know quite a bit Assumes that there is an objective measure of output and that employers know the distribution of effort for a given level of worker effort, assuming that employers know how output is related to worker effort o a. The absence of objective performance measure o Not always the case that there is a verifiable, ‘objective’ measure of performance o Use of share options/other equity-linked incentive schemes has been the most commonly used way of aligning interests of the company’s managers (agents) and shareholders (principals) Making managers’ wealth depend on share price of the firm is meant to motivate them to pursue the interests of their shareholders by maximising profits instead of pursuing company growth and engaging in other nonprofit-maximising activities However, this has also motivated managers to manipulate the earnings of their companies to artificially inflate measured profits and therefore share price to increase the value of managers’ shares beyond the company’s true performance e.g. Enron’s use of accounting and financial manoeuvres to misrepresent its true financial performance o Examples are meant to cause doubt that performancerelated pay schemes involving share options are the right way to motivate managers faithfully to pursue the interests of their shareholders (Healy and Palepu 2003) o Another example of performance manipulation is the case of teachers where they cheated by providing students with correct answers to public tests or altering their answers to improve their scores (Propper and Wilson 2003: 252) o Jacob and Levitt (2003) move beyond anecdotal evidence by reporting the results a large-scale study of cheating by teachers on tests in the Chicago state school system Between 4% and 5% of teachers cheat each year The incidence of cheating increases the more the government attempts to reward teachers and schools on the basis of pupils’ performance in tests, especially at the lower end of the performance scale b. Lack of knowledge of the conditional probability density function that relates a worker’s output to his effort o Principal-agent models also assume that the principal knows the ‘conditional probability density function’ that describes how output is (stochastically/randomly) related to the level of effort exerted by the worker o Various techniques used to determine how much output workers can produce when they’re exerting a reasonable level of effort, thereby allowing employers to set a standard of where PRP should be placed However, these techniques may not elicit the relevant information because it is in the workers’ interest to work slowly when standards are being set to exaggerate the difficulty of their jobs Workers hope that by doing so it will lead to lower standards being set, enabling them to obtain large bonuses when they actually start putting in a reasonable amount of effort (Milgrom and Roberts, 1992: 232-36) E.g. study of the British metalworking industry in early 1970s which found that workers’ ability to thwart attempts to use time and motion studies to uncover the relationship between worker effort and output proved to be so costly to employers that they eventually led to the abandonment of performance-related pay in UK metalworking (Brown 1973) Conclusion: Performance Pay and the Modernisation of the Public Sector Reform of the public sector in the UK has focused on the improvement of incentives so the use of PRP to improve public sector efficiency is becoming increasingly important o New Labour’s program for modernising public services and increasing accountability and demonstrating value for money The nature of the work undertaken in the public sector and the character of the workforce employed may render PRP highly problematic o Difficulty of defining and measuring output o Prevalence of multi-tasking o Importance of teamwork o Existence of norms of professional conduct and capacity to providing workers with the intrinsic motivation to avoid shirking and to work hard Theory of incentive contracts suggests that the optimal compensation scheme is one without PRP or at most gives rise to very low-intensity incentives (Dixit 2002; Burgess and Ratto 2003) Example: teaching i) Teacher’s output is hard to observe and quantify (Propper and Wilson 2003: 258-61; cf. Kessler 1994: 484-85; Back 2000: 250); if output can only be measured vaguely (so that only high-variance estimates of teacher’s efforts are available) then the optimal payment scheme will contain performance based incentives with weak to low-intensity ii) Teaching is a multi-divisional job so compensation based on easilyobservable and quantifiable outputs will distort incentives and lead to dysfunctional behaviour (Dixit 2002: 718; Wiggins and Tymms 2002; Propper and Wilson 2003: 259-60) iii) Teaching involves a lot of teamwork and so contribution of any one teacher may be hard to isolate; where teamwork is important, the use of individual PRP can be counterproductive because it might underline the incentives for individual teachers to cooperate with one another. E.g. competition for individual merit awards can lead to opportunistic and non-cooperative behaviour among teachers (Marsden and French 1998; Adnet 2003) iv) Teachers are professionals that might not be motivated by economic considerations but also by commitment to the norms of professional conduct; the use of PRP may be counterproductive if the extrinsic, financial incentives erodes or crowds out the intrinsic motivation generated by teachers’ commitment to their profession The main lesson to be learned is that the nature of the public services – both of their output and also of the people who work in the public sector – renders a naïve, blanket application of performance-related pay extremely unwise Dixit, 2002: o Public-sector agencies have special features, most notably a lot of dimensions/tasks that make it inappropriate to apply PRP and could generate dysfunctional behaviour o Traditional incentives have to be applied selectively to easilymeasureable, clear and defined performance indicators o PRP should relate the probability of success to specific characteristics of the agencies of organisations, and theoretical research on the design of new reforms should likewise pay attention to these specifics (Dixit 2002: 697) Incentives should be tailored specifically to each different public sector agency by looking at: i) Significance of output measurement problems ii) Multi-tasking iii) Team-working iv) Intrinsic motivation