Training Needs Assessment and Competency Gap Analysis: McGehee and Thayer's Three-Level Model, and the Evidence That Most Training Skips It Entirely

Most organizations do not arrive at a training decision through anything resembling the systematic process this lesson describes. A training decision more commonly starts from one visible signal: a complaint that reached a senior manager, a competitor's new product launch, an approaching compliance deadline, a new software rollout nobody has used before, or simply a belief among managers about what a capable person in this role usually knows. None of these signals is useless as information. None of them, on its own, is a sufficient basis for deciding what to train, whom to train, or whether training is the correct response in the first place.

A training needs assessment is the structured process of answering three separable questions before committing training resources: where in the organization is a performance problem serious enough to justify training, what does competent performance on the specific task actually require, and which people currently fall short of that requirement, by how much, and for what reason. This sequence is not bureaucratic thoroughness for its own sake. Training that targets the wrong level, that trains individuals against a task nobody has actually defined, or that assumes a skill deficiency where the real constraint lies elsewhere, consumes a real organizational resource, principally the paid time employees spend away from productive work, while leaving the underlying problem exactly where it was.

The classic framework for structuring this diagnostic work comes from William McGehee and Paul Thayer's 1961 book Training in Business and Industry, which organized training needs analysis into three levels using the terminology current in industrial psychology at the time: organizational analysis, operations analysis, and man analysis. Later writers, including Irwin Goldstein and Kevin Ford across successive editions of Training in Organizations, relabeled the second and third levels as task analysis and person analysis, a change in terminology rather than in substance, and the model is now commonly abbreviated as the OTP model: organization, task, person.

The sequence matters as much as the three categories themselves. McGehee and Thayer's argument was that an organization should move through these levels in order, because a conclusion reached at the person level, that a specific employee or group performs below standard, is not interpretable without first knowing, at the task level, what the standard actually is, and that task-level standard is itself only worth building once the organizational-level analysis has established that the problem is serious enough, and training-addressable enough, to warrant the investment in the first place.

Organizational analysis asks where, within the organization, training is actually needed, and whether the organization's goals, resources, and climate support training as a solution at all. This involves examining indicators already being collected for other purposes, turnover concentrated in specific units, complaints tied to specific functions, safety incidents clustered at specific sites, and asking whether those patterns point toward a skill or knowledge deficiency rather than toward a staffing, equipment, incentive, or supervisory problem that training cannot touch.

It also asks a question many organizations skip entirely: does this organization's culture, budget, and management support actually allow a training intervention to succeed, independent of how well the training itself is designed. A technically excellent program introduced into a unit whose managers visibly do not value it, or whose incentive structure rewards the opposite of what the training teaches, is organizationally undermined before the first session starts, a limitation of training as an intervention that organizational analysis is specifically meant to catch before the expense is committed.

Task analysis, McGehee and Thayer's operations analysis, defines what a job actually requires for competent performance: the specific tasks involved, the standards those tasks must meet, and the knowledge, skills, and attitudes an incumbent needs to meet them. This step is easy to underrate because it sounds like paperwork, writing down what people already do, but its real function is to establish a standard that exists independently of who currently holds the job. Without it, a person analysis has nothing defensible to compare anyone against; a manager's impression of what "good" looks like, shaped mostly by whichever employee the manager happens to find easiest to work with, quietly substitutes for an actual standard.

A properly built task analysis typically draws on direct observation of the work, structured interviews with high performers and their supervisors, and documentation such as job descriptions or standard operating procedures, triangulating across sources rather than trusting any single informant's account of what the job involves. The output is specific enough to be checked against evidence: a list of tasks, the performance standard for each, and the underlying knowledge and skill each one depends on, not a general description of the role's importance to the organization.

Person analysis compares actual performance against the standard the task analysis established, for identified individuals or groups, and asks a further diagnostic question this lesson returns to later: where a gap exists, is it actually attributable to a deficiency in knowledge or skill, as distinct from motivation, feedback, resources, or the design of the job itself. This is the level most organizations reach for first, because it is the most visible and the fastest to act on: a performance review flags someone, and training gets assigned the same week.

McGehee and Thayer's sequencing argument is a direct corrective to that instinct. A person analysis conducted without a prior task analysis measures people against an undefined standard. A person analysis conducted without a prior organizational analysis risks training individuals for a problem that was never really about them, a distinction this lesson's decision exercise is built around.

McGehee and Thayer's task analysis produces a list of tasks and the knowledge, skills, and attitudes each one requires, tied closely to a specific job as it currently exists. A parallel, partly overlapping tradition emerged from a different starting question: not what does this specific job require, but what distinguishes a superior performer from an average one, potentially across many jobs. David McClelland's (1973) paper in American Psychologist, "Testing for Competence Rather Than for Intelligence," is usually credited as the founding document of this competency tradition. McClelland argued that conventional intelligence and aptitude tests predicted academic and test-taking performance far better than they predicted actual job or life success, and proposed instead identifying the specific behaviors, which he termed competencies, that distinguished people who excelled in a role from people who merely held the role adequately, derived from studying role-holders directly rather than from a general theory of cognitive ability.

Richard Boyatzis extended this into organizational practice in The Competent Manager (1982), identifying competencies associated with managerial effectiveness through critical-incident research: studying what high- and average-performing managers actually did differently in specific situations, rather than asking them to describe their own skills in the abstract. Lyle Spencer and Signe Spencer's Competence at Work (1993) then organized this research into the iceberg model familiar in most HR textbooks: skills and knowledge sit near the visible surface, relatively easy to assess and train, while motives, traits, and self-concept sit beneath it, harder to change through training, and often, in Spencer and Spencer's framework, what actually separates a consistently superior performer from a merely competent one. The distinction reframes what a gap analysis can realistically close: a skill or knowledge gap is the part training addresses; a motive or self-concept gap is not, a point this lesson returns to below.

A second, less widely known complication is worth naming, because it explains why competency framework documents produced by different consultancies for the same role can look so different. The McClelland-Boyatzis-Spencer tradition described above is attribute-based: it asks what underlying characteristics of a person predict superior performance. A separate tradition, associated with the United Kingdom's national occupational standards and vocational qualifications system, is output-based: it defines competence as demonstrated ability to meet a specified performance standard in a work role, independent of which underlying trait produced it. Charles Woodruffe's (1991) article "Competent by Any Other Name," published in Personnel Management, proposed distinguishing the two terms directly on this basis: competence describing the demonstrated output or result a job requires, competency describing the underlying personal behavior or attribute believed to produce it.

Most organizational competency frameworks blend both traditions without acknowledging it, borrowing the behavioral language of the attribute-based tradition while scoring people against output-style proficiency levels from the other. That blending is one reason a gap analysis built against one organization's framework can reach a different conclusion than one built against another, for an employee doing functionally the same job.

A competency gap analysis takes a competency model, whichever tradition or blend it draws from, and compares an individual's or a group's current demonstrated level against the level the model specifies as required for the role, usually along a defined proficiency scale rather than a simple present-or-absent judgment. The output is typically a gap profile: competencies where current performance meets or exceeds the required level, and a shorter list where it does not, ranked by the size of the gap or by how directly it connects to a measurable business outcome.

This sounds procedurally similar to McGehee and Thayer's person analysis, and in practice it usually is the same activity under a newer label, with one real difference. Because competency models are typically built once, for a role or a family of roles, and then reused across many individual gap analyses, the quality of the underlying model becomes a bottleneck that affects every gap analysis built on top of it. A model built from a rushed, unrepresentative sample of high performers, or borrowed wholesale from a vendor's generic library without adaptation to this organization's actual work, produces confident-looking gap scores measured against a standard that does not actually describe this job.

Michael Campion and colleagues' (2011) article in Personnel Psychology, "Doing Competencies Well: Best Practices in Competency Modeling," assembled twenty best practices drawn from two major corporations, a consulting firm, a university research program, and a Society for Industrial and Organizational Psychology task force, precisely because the authors judged that a meaningful share of competency modeling work in practice falls short of basic methodological standards: a biased or too-small sample of job incumbents used to define the model, no link established between the competencies and a measurable performance outcome, or a generic, purchased competency library treated as though it had already been validated for this specific organization.

Leanne Markus, Helena Cooper-Thomas, and Keith Allpress's (2005) critique in the New Zealand Journal of Psychology, "Confounded by Competencies?," goes further, arguing that the gap between the benefits competency models are marketed as delivering and the benefits actually demonstrated in published research is substantial, partly because the two traditions above get blended inconsistently within one framework, and partly because many models are adopted for symbolic value, signaling systematic talent management, rather than built with the rigor Campion and colleagues describe. None of this means competency gap analysis is not worth doing. It means the method's output is only as trustworthy as the underlying model, and that trustworthiness has to be established separately, at the point the model was built, by checking whether it was actually derived from this organization's own high performers and tied to measurable performance, rather than assumed from a template.

Robert Mager and Peter Pipe's (1970) short book Analyzing Performance Problems: Or, You Really Oughta Wanna addresses a question McGehee and Thayer's model assumes but does not itself answer: once a performance gap has been identified at the person level, is it actually the kind of gap training can close. Mager and Pipe's diagnostic distinction is blunt, and still widely taught because it remains genuinely useful: ask whether the person could perform the task correctly if their circumstances required it immediately. If the honest answer is no, the problem may be a skill or knowledge deficiency, and training is a plausible remedy, provided the organizational and task analysis described earlier have already established what "correctly" means and that closing the gap is worth the cost. If the honest answer is yes, the person already knows how but is not doing it, the cause lies elsewhere: unclear performance expectations, feedback that never actually reaches the person, a job or incentive structure that quietly punishes the desired behavior, or resources and tools the person simply does not have. No amount of well-designed training fixes a "will not do it" problem, because the constraint was never an absence of capability in the first place.

The practical difficulty is that a "will not do it" problem and a "cannot do it" problem can look identical from the outside, both surfacing as a missed target, which is exactly why training is so often commissioned before this distinction is checked. Training also has an attractive property that has nothing to do with whether it is the correct response: it is visible, schedulable, and easy to present to a board as decisive action, in a way that redesigning an incentive structure or fixing a chronic resourcing problem often is not. That asymmetry, between how satisfying training looks and how often it is actually the right response, is a structural reason to expect needs analysis to be skipped, not an occasional lapse by a few careless organizations.

How often does the sequence this lesson describes actually happen before training is commissioned. Winfred Arthur, Winston Bennett, Pamela Edens, and Suzanne Bell's (2003) meta-analysis in the Journal of Applied Psychology, built from 397 training-effectiveness data points drawn from the published literature, reported a specific and uncomfortable number: only 6 percent of those data points, 22 out of 397, came from studies that reported conducting any training needs assessment before the training was implemented. The authors' own caution is worth repeating rather than smoothing over: it is entirely possible that a larger share of the underlying programs were preceded by some needs assessment the published study simply did not mention, because the researchers were not treating it as a variable of interest. The figure measures what got reported in the literature, not a direct census of organizational practice. Even allowing for that caveat, a literature in which barely one in twenty training evaluations mentions a needs assessment step is difficult to reconcile with how central that step is treated in every training textbook, including this one.

The second finding from the same analysis is quoted less often and is arguably more interesting. Among the small number of studies that did report a needs assessment, Arthur and colleagues tested whether training following a more comprehensive assessment, covering multiple levels of the McGehee and Thayer sequence, produced larger effects than training following only one level. They found no clear pattern. The honest explanation is almost certainly statistical rather than theoretical: these comparisons rested on four or fewer data points in most cells, far too few to detect a real effect even if one exists. It is not evidence that organizational, task, and person analysis together produce no more benefit than any single level alone; the study could not answer that question either way. What the finding does establish is that the strong claim sometimes made in practitioner training, that rigorous multi-level needs assessment has been empirically proven to improve outcomes, is not something this specific, frequently cited meta-analysis actually supports.

Put together, the honest picture is specific rather than simply discouraging. McGehee and Thayer's sequence, with Mager and Pipe's diagnostic question layered on top of it, describes a genuinely sound way to decide whether, where, and what to train. The evidence on whether organizations actually follow that sequence before committing training budget is thin, and what little evidence exists suggests the sequence is the exception in the published record, not the norm, for reasons that have more to do with how visible and schedulable training looks as a response than with any considered judgment that needs analysis is unnecessary. Treating a textbook's confident three-level diagram as a description of standard practice, rather than as a standard most training decisions do not actually meet, is the misreading this lesson is built to correct.

Select a specific, real or realistically disguised organizational performance problem that someone proposed solving with training, drawn from your own organization, a role you have observed closely, or a documented public case. The problem should allow a genuine, defensible case to be made for more than one underlying cause, a skill gap, a motivational or incentive issue, a resource or process constraint, so that the diagnosis itself has real analytical work to do.

Prepare a report of 2,000 to 2,500 words that:

1. conducts an organizational analysis identifying where the problem is concentrated and whether any non-training organizational factors plausibly explain part of it, using whatever performance data you can reasonably access or construct;
2. conducts a task analysis defining, independently of current incumbents, what competent performance on the specific task actually requires, drawing on observation, interviews, or documentation as appropriate to your case;
3. conducts a person-level or competency gap analysis comparing the relevant individuals or group against that standard, stating clearly what evidence the comparison is based on;
4. applies Mager and Pipe's diagnostic question explicitly to state whether the identified gap is genuinely a "cannot do it" problem that training can address, or a "will not do it" problem that requires a different intervention, and justifies that conclusion rather than asserting it; and
5. if a competency model is used anywhere in the analysis, evaluates its own validity against Campion et al.'s (2011) best-practice standards rather than assuming the model is trustworthy because it exists.

Use at least five academic or documented sources, including at least three from this article's reference list. State explicitly where your own evidence for the specific organization is incomplete or inferred rather than directly observed.

Assessment rubric (100 marks)

1. Arthur, W., Jr., Bennett, W., Jr., Edens, P. S., & Bell, S. T. (2003). Effectiveness of training in organizations: A meta-analysis of design and evaluation features. Journal of Applied Psychology, 88(2), 234–245.
2. Boyatzis, R. E. (1982). The Competent Manager: A Model for Effective Performance. New York: Wiley.
3. Campion, M. A., Fink, A. A., Ruggeberg, B. J., Carr, L., Phillips, G. M., & Odman, R. B. (2011). Doing competencies well: Best practices in competency modeling. Personnel Psychology, 64(1), 225–262.
4. Goldstein, I. L., & Ford, J. K. (2002). Training in Organizations: Needs Assessment, Development, and Evaluation (4th ed.). Belmont, CA: Wadsworth.
5. Mager, R. F., & Pipe, P. (1970). Analyzing Performance Problems: Or, You Really Oughta Wanna. Belmont, CA: Fearon Publishers.
6. Markus, L. H., Cooper-Thomas, H. D., & Allpress, K. N. (2005). Confounded by competencies? An evaluation of the evolution and use of competency models. New Zealand Journal of Psychology, 34(2), 117–126.
7. McClelland, D. C. (1973). Testing for competence rather than for "intelligence." American Psychologist, 28(1), 1–14.
8. McGehee, W., & Thayer, P. W. (1961). Training in Business and Industry. New York: Wiley.
9. Spencer, L. M., & Spencer, S. M. (1993). Competence at Work: Models for Superior Performance. New York: Wiley.
10. Woodruffe, C. (1991). Competent by any other name. Personnel Management, September 1991, 30–33.

Verification notes: Arthur, Bennett, Edens, and Bell (2003) was fetched in full and read directly via pdftotext rather than a secondary summary, to confirm the exact 6% (22 of 397) needs-assessment figure and the "no clear pattern" finding together with the authors' own caveats about under-reporting and small cell sizes. McGehee and Thayer (1961), Goldstein and Ford (2002), McClelland (1973), Boyatzis (1982), Spencer and Spencer (1993), Woodruffe (1991), Mager and Pipe (1970), Campion et al. (2011), and Markus, Cooper-Thomas, and Allpress (2005) were checked against publisher, journal-index, and indexed bibliographic or archive records via web search rather than a full-text read of each source; no statistic or finding beyond what is reported in these sources is claimed. The Oyshee Home Appliances case, the decision scenario, the Arthur et al. summary table, and the closing synthesis are original teaching material for this series.

Lesson 43, next in Module 09, picks up once a genuine training need has actually been established: ADDIE as a design process, Bloom's taxonomy for classifying what a learning objective is asking for, and how to write a learning objective specific enough to be designed against and evaluated later, the evaluation question Lesson 44 takes up in full through the Kirkpatrick model.