Chapter 5 — Performance Appraisal and Management

Chapter 5 opens with eight learning goals, numbered 5.1 through 5.8, reproduced verbatim below since later chapters (6, 7, 8) build directly on the measurement vocabulary this chapter establishes.

1. 5.1 Design a performance management system that attains multiple purposes (e.g., strategic, communication, and developmental; organizational diagnosis, maintenance, and development).
2. 5.2 Distinguish important differences between performance appraisal and performance management (i.e., ongoing process of performance development aligned with strategic goals), and confront the multiple realities and challenges of performance management systems.
3. 5.3 Design a performance management system that meets requirements for success (e.g., congruence with strategy, practicality, meaningfulness).
4. 5.4 Take advantage of the multiple benefits of state-of-the-science performance management systems, and design systems that consider (a) advantages and disadvantages of using different sources of performance information, (b) situations when performance should be assessed at individual or group levels, and (c) multiple types of biases in performance ratings and strategies for minimizing them.
5. 5.5 Distinguish between objective and subjective measures of performance and advantages of using each type, and design performance management systems that include relative and absolute rating systems.
6. 5.6 Create graphic rating scales for jobs and different types of performance dimensions, including behaviorally anchored ratings scales (BARS).
7. 5.7 Consider rater and ratee personal and job-related factors that affect appraisals, and design performance management systems that consider individual as well as team performance, along with training programs to improve rating accuracy.
8. 5.8 Place performance appraisal and management systems within a broader social, emotional, and interpersonal context. Conduct a performance appraisal and goal-setting interview.

The chapter defines performance management as a “continuous process of identifying, measuring, and developing the performance of individuals and teams and aligning performance with the strategic goals of the organization” (Aguinis, 2019, p. 4) — not a one-time event tied to an annual review. Performance appraisal, by contrast, is the systematic description of job-relevant strengths and weaknesses within and between employees or groups: one component nested inside the larger process.

The field's track record is mixed: decades of research have not resolved appraisal's problems, leading one reviewer to call it the “Achilles heel” of HRM (Heneman, 1975), a label the authors say still applies (DeNisi & Murphy, 2017). Yet organizations will not abandon it: about 90% of 278 organizations across 15 countries use a company-sanctioned system (Cascio, 2011). Performance management is a process and dialogue with social, motivational, and interpersonal dimensions (Fletcher, 2001), and it fails if not linked to organizational goals.

Well-designed systems serve several purposes: strategic (linking activity to organizational goals), communication (what's valued), administrative (promotion, termination, merit decisions), HR research (validation criteria), developmental (feedback that fuels training), organizational diagnosis (appraisal as “the beginning of a process rather than an end product,” Jacobs, Kafry, & Zedeck, 1980), and legal documentation.

Every organization faces five realities (Ghorpade & Chen, 1995): appraisal is inevitable and central to legal defense; it carries real consequences; accuracy gets harder as job complexity rises; raters face temptation toward political bias — rewarding allies, punishing rivals (Longenecker & Gioia, 1994); and implementation takes real buy-in or the system backfires. These produce political challenges (giving average ratings to weak performers rather than confront them — “damning with faint praise,” Benedict & Levine, 1988) and interpersonal challenges (managers resisting the “judge” role, Meyer, 1991).

For any performance management system to succeed, it should have nine characteristics (Aguinis, 2019; Aguinis, Joo, & Gottfredson, 2011) — a favorite source for discussion prompts asking you to evaluate or redesign a system.

Evidence backs several of these directly: perceived meaningfulness predicted system adoption among 176 Australian government workers (Langan-Fox, Waycott, Morizzi, & McDonald, 1998), and participation correlated .61 with favorable reactions across 27 studies (Cawley, Keeping, & Levy, 1998).

Systems meeting these nine requirements become a source of competitive advantage (Aguinis, Joo, & Gottfredson, 2011): employees gain self-esteem and clarity; managers gain a more motivated workforce and sharper differentiation between performers; organizations gain more appropriate decisions, less misconduct, better legal protection, and higher engagement. The dilemma from Section 2 is solved not by discarding appraisal but by implementing it against this evidence base.

Raters must be cooperative, trained, and have direct knowledge of the person being rated. Five perspectives each supply a different piece of the puzzle.

The immediate supervisor is usually best positioned to judge performance against organizational objectives and link it to reward/punishment — a linkage the chapter calls essential. Limits: in jobs like teaching, law enforcement, or self-managed teams, supervisors may rarely observe performance directly (Hogan & Shelton, 1998).

Peer assessment takes three forms — nomination (extremes), rating (feedback), ranking (discrimination) — and shows favorable reliability and validity overall (Kane & Lawler, 1978), though friendship bias lowers acceptance while developmental use raises it (McEvoy & Buller, 1987), and peer ratings carry more common method variance than other sources (Conway, 1998).

Subordinates directly observe delegation, planning, leadership style, and communication — variance no other source captures (Conway, Lombardo, & Sanders, 2001), validly predicting later supervisory ratings over multi-year spans (McEvoy & Beatty, 1989); they stay more honest when anonymous, since disclosed identity inflates ratings (Antonioni, 1994).

Self-appraisal, paired with goal setting, improves motivation but runs more lenient and less in agreement with others' judgments — the Western norm (Atkins & Wood, 2002); Taiwan shows the opposite modesty bias (Farh, Dobbins, & Cheng, 1991). Leniency drops when self-ratings serve development rather than administration (Meyer, 1991). Fixes: relative rather than absolute scales (Farh & Dobbins, 1989), multiple self-appraisal opportunities, guaranteed confidentiality, and a future-behavior orientation.

Clients served — customers of jobs with heavy public interaction — supply useful information despite not identifying with organizational goals; their input can inform employment decisions, HR research, or self-development.

Appraisal is rarely a purely individual act: 98.5% of 135 surveyed raters used at least one secondhand source of performance information (Raymark, Balzer, & De La Torre, 1999). Could it be formalized as a group task? Groups remembered behaviors better than individuals but showed greater response bias (Martell & Borg, 1993); a later study found individual ratings were swayed by a prior performance cue while four-person consensus groups were not (Martell & Leavitt, 2002). Groups help when the task has a demonstrably correct answer and the biasing cue isn't too large; for highly subjective judgments (e.g., “management potential”), group ratings can amplify rather than reduce individual bias.

Since sources aren't mutually exclusive, 360-degree feedback systems combine self, peers, subordinates, and sometimes clients, offering improved reliability, a broader information range that reduces criterion deficiency, coverage of contextual and counterproductive behavior, and reduced bias (Campion, Campion, & Campion, 2015). Four conditions must hold for them to work (Bracken & Rose, 2011): relevant content, data credibility, accountability, and organization-wide (not unit-by-unit) participation.

Because raters at different organizational levels see different facets of a job, cross-source agreement runs modest: subordinate–self correlates only .14, subordinate–supervisor and self–supervisor both .22 (Conway & Huffcutt, 1997). The chapter's key move: expecting full interrater agreement may be the wrong question. What should hold first is measurement equivalence — whether the underlying trait relates to observed scores the same way across sources. Checking agreement before establishing equivalence risks mistaking non-equivalence for disagreement (Cheung, 1999), and real non-equivalence can make cross-rater differences misleading (Maurer, Raju, & Collins, 1998).

Equivalence has been found for managers' team-building skills across peer and subordinate raters (Maurer, Raju, & Collins, 1998), and for overall job performance, effort, and job knowledge rated by supervisors and peers (Viswesvaran, Schmidt, & Ones, 2002) — but NOT for interpersonal competence rated by supervisors versus peers, a reminder that equivalence must be checked, not assumed. Once established, a hybrid multitrait-multirater matrix (Borman, 1974) lets raters judge only dimensions they're well positioned to assess, using within-level agreement as the convergent-validity index.

Judgmental biases are traditionally treated as systematic measurement error, though definitions have been inconsistent (Saal, Downey, & Lahey, 1980), and not every manager agrees these are “errors” — a team culture emphasizing contextual over task performance may be capturing a real variable, not a mistake (Lievens, Conway, & De Corte, 2008).

Leniency/severity is a stable rater tendency (Kane, Bernardin, Villanova, & Peyrfitte, 1995) that senior managers take seriously — 77% of a Fortune 100 sample called lenient appraisals a threat to validity (Bretz, Milkovich, & Read, 1990); administrative ratings run a third of a standard deviation higher than research/development ratings (Jawahar & Williams, 1997). Remedies: forced distributions (e.g., 20/70/10), required rank-ordering, regular feedback, and rater accountability — IBM, Pratt & Whitney, and Grumman adopted forced distributions to break leniency blocking downsizing (Kane & Kane, 1993).

Central tendency clusters everyone near the midpoint — “everybody is average” — destroying appraisal's value for decisions or feedback; the remedy is specifying what each anchor means and convincing raters merit ratings matter.

Halo — assigning ratings on all dimensions from one general impression rather than distinguishing across dimensions (Lance, LaPointe, & Stewart, 1994) — is the most actively researched bias. Two critical reviews (Balzer & Sulsky, 1992; Murphy, Jako, & Anhalt, 1993) concluded it is less common than believed, doesn't necessarily hurt rating quality, and can't reliably be separated from true halo (a genuine underlying performance factor) in field settings. An experimental study found rater-error halo stayed uniform across dimensions while true halo varied (Solomonson & Lance, 1997) — intercorrelated ratings may reflect real performance, not bias.

Objective measures include production data (sales volume, units, errors) and employment data (accidents, turnover, absences) — usually tied to results but often partly outside the employee's control. This is why cognitive ability predicts supervisor ratings of sales performance (r = .40) far better than objective sales performance itself (r = .04) (Vinchur, Schippmann, Switzer, & Roth, 1998); objective and subjective measures correlate only .39 (Bommer, Johnson, Rich, Podsakoff, & Mackenzie, 1995). Subjective measures depend on human judgment and are prone to the biases above; they may be relative (comparing ratees) or absolute (describing a ratee alone).

Relative systems are easy to explain and control leniency, severity, and central tendency — but yield only ordinal data and lose reliability mid-distribution. Rank ordering (simple, or alternation) orders all ratees; paired comparisons rank each by how often judged superior to every other; forced distribution slots a fixed percentage into categories, made famous by GE's “vitality curve” (top 20/vital 70/bottom 10) — most beneficial for autonomous jobs, riskier where tasks are interdependent (Moon, Scullen, & Latham, 2016).

Absolute systems: essays give rich detail but are hard to quantify. Behavioral checklists have raters check descriptive statements rather than evaluate — reporters, not judges, raising reliability (Stockford & Bissell, 1949) but weakening diagnostic feedback. The forced-choice system (Sisson, 1948) groups equally desirable statements on a hidden key so raters can't inflate ratings — but this breeds rater resistance. Critical incidents record specific effective/ineffective things employees did (Flanagan, 1954a) — concrete but time-consuming, hence BARS. Graphic rating scales are most widely used; purely qualitative anchors (“outstanding,” “unsatisfactory”) are nearly worthless since meaning is left to the rater, while concrete anchors are stronger — and tests found simple graphic scales just as reliable as forced-choice, and more acceptable to raters (King, Hunter, & Schmidt, 1980).

Smith and Kendall (1963) proposed improving graphic scales not by tricking the rater (as forced-choice does) but by anchoring them to concrete, observable behavior: “We should ask him questions which he can honestly answer about behaviors which he can observe” (p. 151). The retranslation procedure: a group identifies the job's key performance dimensions; a second group generates critical incidents illustrating effective, average, and ineffective performance on each; a third group sorts incidents back into the dimensions they best represent — incidents without 60–80% judge agreement are dropped; and a fourth group scales each surviving incident (e.g., on a 7- or 9-point scale) to build the final instrument.

BARS development is long and labor-intensive, and separate scales are needed for dissimilar jobs, yet it remains common — about 40% of surveyed hotels used it (Woods, Sciarini, & Breiter, 1998).

Idiosyncratic (rater) variance is actually larger than true ratee-performance variance — 1.21x larger for supervisory ratings, 2.08x for peer ratings, 1.86x for subordinate ratings (Scullen, Mount, & Goff, 2000). Raters may also inflate ratings for reasons unrelated to true performance — avoiding confrontation, pushing out a problem employee, or appearing competent (Spence & Keeping, 2011).

Team-based organizations don't automatically outperform non-team ones (Hackman, 1998), but systems increasingly must capture both individual contribution and team performance — ignoring the former invites social loafing, since once teammates spot a free rider, they withdraw effort too (Heneman & von Hippel, 1995). Individual and team incentives have comparable, complementary effects (g = 0.32 and 0.34; Garbers & Konradt, 2014). Three team types need different measurement (Scott & Einstein, 2001): work/service (intact, routine), project (disband once done — Hewlett-Packard measures during rather than after, since end-of-project data arrives too late), and network (dispersed, telecommunications-coordinated, most nonroutine).

Rater training has three objectives: sharpen observational skills, reduce judgmental biases, and improve how raters communicate performance information. Traditional training targeted leniency, central tendency, and halo directly (Bernardin & Buckley, 1981), but this backfires: raters trained to produce lower averages and more cross-dimension variability actually become less accurate (Hedge & Kavanagh, 1988), and bias-elimination training has only short-term effects (Fay & Latham, 1982). Rater error training (RET) — simple exposure to common errors — similarly doesn't guarantee they disappear (London, Mone, & Scott, 2004).

Frame-of-reference (FOR) training is the more effective alternative, with meta-analytic evidence of d = .50 on accuracy (Roch, Woehr, Mishra, & Kieszczynska, 2012). Following Pulakos (1984, 1986): participants learn they'll rate three ratees on three dimensions; they read dimension definitions and scale anchors together; the trainer discusses behaviors at each effectiveness level, building a shared “theory of performance”; participants practice-rate a videotaped manager, ratings are posted and discussed as a group, and the trainer explains the correct target score for each dimension.

FOR training works by giving raters a coherent theory of performance — how dimensions map to behaviors, how to judge effectiveness, how to integrate judgments into an overall rating (Sulsky & Day, 1992) — through hands-on practice with standards, examples, and accuracy feedback (Stamoulis & Hauenstein, 1993). One caveat: peers, supervisors, and subordinates hold genuinely different frames of reference (Hauenstein & Foti, 1989), so training should address content and process alike, flagging idiosyncratic raters for follow-up. This style of training especially improves accuracy for individual ratees on separate dimensions (Day & Sulsky, 1995).

Any rating format has only limited impact on appraisal by itself (Banks & Roberson, 1985) — outcomes emerge from context, format, and motivation together. Fairness splits into interactional justice (the supervisor-employee exchange) and procedural justice (the system's structure): a review of 295 U.S. circuit court decisions found fairness and due-process issues the most salient factor in appraisal rulings (Werner & Bolino, 1997). Aguinis and Pierce (2008) flagged four related constructs: social power/influence, trust (especially during downsizing), social exchange, and group dynamics (including supervisor-subordinate romances).

Feedback doesn't automatically help: a meta-analysis of 131 studies found it improves performance on average, but 38% of feedback interventions actually reduced it (Kluger & DeNisi, 1996) — feedback that shifts attention to the self rather than the task backfires. Without a formal system, feedback-seeking is uneven — stereotype threat reduced it among African American managers who were the only one in their workplace (Roberson, Deitch, Brief, & Block, 2003). EPM generates continuous data but doesn't remove the need for face-to-face interpretation; most supervisors still find the formal interview trying, since people resist “playing God” (McGregor, 1957).

A few practices deserve elaboration. Frequent feedback beats once-a-year appraisal (Cederblom, 1982); subordinate preparation predicts greater satisfaction and improvement (Burke, Weitzel, & Weir, 1978); self-review casts the manager as counselor rather than judge. Lead with positive feedback on minor issues before major ones, since early praise reduces the “blocking” effect criticism otherwise triggers (Stone, Gueutal, & McIntosh, 1984); destructive criticism reliably damages self-efficacy (Baron, 1988). Specific-difficult goals outperform specific-easy ones (Erez, Earley, & Hulin, 1985), and top-management commitment matters enormously — 56% productivity gains versus 6% in MBO programs (Rodgers & Hunter, 1991). Microsoft uses the SMART label: Specific, Measurable, Achievable, Results-based, Time-specific (Shaw, 2004).

Chapter 5 ends with eleven discussion questions, each paired below with a concise model answer.

1. Why do performance management systems often fail?

Systems fail when political barriers (self-interested raters, damning with faint praise) and interpersonal barriers (resistance to the “judge” role) go unaddressed, or when treated as a one-time event disconnected from strategy.

2. What is the difference between performance management and performance appraisal?

Performance management is the continuous, strategy-aligned process of developing performance (Aguinis, 2019). Performance appraisal — the systematic description of strengths and weaknesses — is one component nested inside it, not a synonym for it.

3. What are the most important purposes of performance management systems and why?

Strategic alignment, communication, employment decisions, HR research, development, organizational diagnosis, and legal documentation — each serving a different stakeholder: strategy for the organization, growth for the employee, defensible decisions for managers.

4. Under what circumstances can performance management systems be said to 'work'?

A system works when it satisfies the nine fundamental requirements and delivers documented benefits to employees, managers, and the organization — and when trust, fairness, and participation are managed as carefully as the measurement mechanics.

5. What kinds of unique information about performance can each of the following provide: immediate supervisor, peers, self, subordinates, and clients served?

The supervisor links performance to objectives and rewards. Peers discriminate performance levels well but risk friendship bias. Subordinates uniquely see delegation and leadership style. Self-appraisal boosts motivation but runs lenient. Clients served provide an external view internal raters can't replicate.

6. Consider an organization you know and rank the fundamental requirements for that organization.

There's no universally correct ranking — the nine requirements must be weighted against the organization's context. A strong answer names its most acute pain point (e.g., legal exposure argues for reliability/validity and fairness) and explains that ranking logic rather than reciting the list in order.

7. When is agreement across performance information sources desirable? When might disagreement be expected or even desired?

Agreement is most desirable when raters observe the same directly comparable behavior under established measurement equivalence. Disagreement is expected, and can be desirable, when sources occupy different vantage points — low correlation then signals complementary information, not error, which is exactly what a 360-degree system captures.

8. What interpersonal and social interaction dimensions should be considered in implementing a performance management system?

Social power and influence, collective trust, social exchange, and group dynamics including close interpersonal relationships. Procedural and interactional justice — the system's structure versus the supervisor's face-to-face treatment — both shape whether it's accepted at all.

9. Under what circumstances would you recommend that performance measurement be conducted as a group task?

Group rating helps when the judgment has a demonstrably correct answer and any biasing cue is small; it's a poor choice for highly subjective judgments, since groups can then amplify rather than correct individual bias.

10. What key elements would you design into a rater-training program?

Frame-of-reference (FOR) training rather than simple error-awareness training, since FOR shows a meaningfully larger accuracy effect (d = .50): dimension definitions, practice ratings, group discussion, and trainer feedback on target scores, while accounting for the fact that different rater sources legitimately hold different frames of reference.

11. In a team-based organization, what role would individual-level performance measurement play alongside a team-based system?

Individual measurement remains essential even in team structures: without it, social loafing emerges and free-riding erodes team effort once noticed. Individual and team incentives produce comparable, complementary effects, so the two levels should run side by side.

Every bolded or explicitly defined term in Chapter 5, in one line each, roughly in the order the chapter introduces them.

A single table capturing the chapter's core distinctions, named frameworks, and most quotable findings.