Job Analysis, Job Description and Job Specification: How Organizations Learn What a Job Actually Requires, and How Reliably They Know It

Job analysis produces the raw material that almost every other human resource function consumes without being able to produce it on its own. A recruiter advertising a role needs to know what the role actually does to write anything beyond a generic list of adjectives. A selection panel designing an interview needs to know which knowledge, skills and abilities actually predict performance in the job, not which ones sound impressive on a competency poster. A compensation analyst comparing two positions for internal equity, the subject of Lesson 33, needs a description detailed enough to judge relative worth, not a job title and a paragraph of boilerplate. A trainer building a program needs to know the actual gap between what a job requires and what its current holder can do. None of these functions can manufacture that knowledge by itself; each borrows it from job analysis, and if the borrowed material is wrong, every function built on top of it inherits the error without necessarily knowing the error is there.

This is also why job analysis carries a legal weight that its routine, administrative reputation inside most organizations does not suggest. In jurisdictions with employment discrimination or disability law, a job description's statement of which duties are "essential" to a role can determine whether a decision to dismiss, decline to hire, or decline to accommodate an employee is defensible. A description assembled hastily, copied from a template, or never updated after the job changed is a weak foundation for a decision that may later be examined by a tribunal, a court, or simply an aggrieved former employee asking an uncomfortable question. The document's ordinariness, something most managers barely think about once it is filed away, is precisely what makes it risky when it turns out to be wrong.

Job analysts collect information about a job through a small number of basic methods, and the methods differ less in what they ask about than in how directly they observe it. Direct observation puts the analyst alongside the jobholder, watching and recording what actually happens during a work cycle. It works best for jobs with short, visible, repeatable cycles, assembly work, retail service, routine processing, where a handful of cycles reveals most of what matters, and far less well for long, irregular or largely cognitive cycles: a strategy consultant's real work is mostly invisible to someone standing in the room, and a sales representative's most consequential moments may unfold over months of a single relationship rather than inside any observed hour.

The interview, structured or semi-structured, asks the jobholder, and usually the jobholder's supervisor, to describe the job directly. It captures tasks an observer would never catch by watching, judgment calls, exceptions, the reasoning behind a choice, across the job's full cycle rather than whatever slice happened to occur during observation. Its weakness is that it depends on memory, self-presentation and the interviewer's skill at distinguishing what a job actually requires from what the interviewee believes it should require, or wants a superior to believe it requires.

The standardized questionnaire, of which the Position Analysis Questionnaire discussed below is the most developed example, asks a large sample of jobholders or analysts to rate a job against a fixed, pre-defined set of items. It scales to hundreds of jobs at a cost no interview program can match, and produces numerical data that supports the statistical comparison across jobs that the compensation and classification work in Lesson 33 needs. Its weakness is that a fixed item set can only capture what its designers anticipated; a genuinely novel element of a job has nowhere to register on a form built for jobs in general.

Edwin Levine, Ronald Ash, Hale Hall and Frank Sistrunk's 1983 study in the Academy of Management Journal asked experienced job analysts to rate several of these methods against a range of practical criteria, cost, reliability, and usefulness for different HR purposes among them. Their central finding is one textbooks often compress into a single misleading sentence: no single method dominated across all the criteria the analysts rated. A method that scored well on cost scored less well on the richness of the information it produced; a method well suited to compensation work was not the same method best suited to training needs analysis. The practical implication, borne out ever since, is that method choice is a trade-off decision tied to the specific HR purpose at hand, not a search for one best technique the field has simply failed to settle on.

Ernest McCormick, Paul Jeanneret and Robert Mecham's Position Analysis Questionnaire, validated in their 1972 study in the Journal of Applied Psychology, took a specific, deliberate approach to the problem of comparing jobs that look nothing alike on the surface: ask about generalized worker behaviors rather than job-specific duties.

McCormick, Jeanneret and Mecham's original study factor-analyzed PAQ ratings across 536 jobs and identified a set of underlying job dimensions that recurred across occupations with no obvious surface similarity, and showed that pay rates and aptitude requirements could be predicted reasonably well from the quantified job-analysis data those dimensions produced. That is the PAQ's genuine contribution: it gives compensation and classification work, the subject of Lesson 33, a numerical basis for comparing jobs rather than a manager's unaided impression of which role deserves more. It does not replace the specific, duty-level writing a usable job description still requires; an HR team that stops at a PAQ profile has measured the job's general behavioral demands without yet producing the document an employee can actually read to understand what they were hired to do.

Sidney Fine, working for the U.S. Department of Labor from the late 1940s onward, built a different kind of job analysis around a specific premise: that every job, regardless of industry or title, requires a worker to function in relation to three things, Data, the information and ideas handled, People, the other humans, or at times animals, the worker deals with, and Things, the physical machines, tools and materials the worker handles. Functional Job Analysis, documented fully in Fine and Steven Cronshaw's 1999 book Functional Job Analysis: A Foundation for Human Resources Management, rates every task a worker performs on separate ordinal scales describing how complex that task's demand on Data, People and Things actually is, from the simplest end of each scale toward its most complex. The method also pushes the analyst to write task statements in a controlled, checkable form, naming the action, its object and the result it produces, rather than the vague duty language, "responsible for...," "assists with...," that fills so many job descriptions in practice and says almost nothing a reader could verify against actual behavior.

Fine's worker-function ratings became the organizing scales behind the United States' Dictionary of Occupational Titles for decades, a legacy that lasted until the Department of Labor replaced the Dictionary with O*NET in the late 1990s, an effort documented in Norman Peterson and colleagues' 1999 account of O*NET's development for the American Psychological Association. O*NET kept job analysis's basic ambition, standardized, comparable information about what jobs require, but organized it around a broader content model covering worker characteristics, worker requirements, experience requirements, and occupational and occupation-specific information, rather than Fine's narrower Data-People-Things triad. For a practicing HR manager today, O*NET is a genuinely useful starting reference for an unfamiliar job family. It is a poor substitute for analyzing a specific employer's own version of that job, because O*NET describes a generalized occupation across the whole economy, not the particular duties, tools and context of one organization's role.

Job analysis is a process; the job description and the job specification are its two outputs, and confusing them is one of the most common, and most consequential, mistakes in HR practice. The job description answers a question about the job: what does this role actually do, and under what conditions? It lists duties, responsibilities, reporting relationships and working conditions, written, ideally, as the specific, verifiable task statements Fine's method was built to produce rather than a vague summary. The job specification answers a different question, this time about the person: what does someone need to bring to this role to perform it competently? It lists the KSAOs the job demands, a required qualification, a minimum period of experience with a particular kind of transaction, the ability to read a technical drawing, the physical capacity to lift a stated weight repeatedly.

The distinction matters practically, not only definitionally, because the two documents fail in different ways and feed different decisions. A job description that is vague or outdated misleads whoever is trying to understand the role, including a new hire trying to understand what they were actually hired to do. A job specification that is inflated, listing requirements the job does not genuinely need, narrows the applicant pool for no performance benefit and can create the kind of disparate impact employment discrimination law exists to catch; a specification that is too thin lets through candidates who cannot actually do the job. Writing a specification well requires tracing each listed requirement back to a specific duty in the description it was derived from. A requirement that cannot be traced to an actual duty is a candidate for removal, not a line item to defend out of habit.

A second distinction, often blurred with the description-specification split but actually a separate axis, is between task-based and competency-based job analysis. Task-based analysis, the PAQ and Functional Job Analysis traditions above both belong here, asks what a specific job currently requires, grounded in observable behavior tied to one job at one point in time. Competency-based analysis, popularized through consulting practice from the 1990s onward, asks a broader question: what underlying characteristics, often framed as competencies such as "strategic thinking" or "customer focus," distinguish higher performers across a role, or an organization, now and as it expects to need them in future.

Juan Sanchez and Edward Levine's 2009 article in Human Resource Management Review set out the distinction carefully rather than simply declaring one approach superior. Traditional job analysis, in their account, is oriented toward the past and the specific job, describes what incumbents actually do, and typically supports operational decisions such as selection and compensation with a defensible paper trail. Competency modeling is oriented toward the organization's future and its strategy, is more likely to be used to try to shape incumbent behavior toward where the organization wants to go rather than simply document where it already is, and tends to produce broader, more abstract descriptors that are harder to verify against observable behavior. Sanchez and Levine's own recommendation, worth taking at face value rather than reading past, is that the two approaches should be used together rather than treated as competitors, each covering a weakness in the other rather than one simply superseding it.

This matters for a manager deciding how to analyze a role today. A pure competency list, "collaborative," "innovative," "results-driven," reads well on a careers page and says almost nothing a selection panel can actually assess against evidence. A pure task list can miss precisely the forward-looking judgment a competency framework exists to capture: how the role will need to change as the organization's strategy changes. Lesson 33 picks this up directly when it turns to job evaluation, which needs job analysis data detailed and comparable enough to rank roles against each other, a demand task-based analysis is generally better equipped to meet than a competency model built mainly for development conversations.

Job analysis is usually presented in HR textbooks as a technical, almost mechanical exercise: follow the steps, collect the data, write the description. The research literature tells a more complicated story, and an honest treatment of this lesson's subject has to include it rather than smooth it over for a tidier narrative.

Frederick Morgeson and Michael Campion's 1997 article in the Journal of Applied Psychology, "Social and Cognitive Sources of Potential Inaccuracy in Job Analysis," laid out sixteen distinct mechanisms, grouped into social sources and cognitive sources, through which a job analysis rating can drift from an accurate account toward something else. Social sources include social desirability, a jobholder inflating a task's apparent importance because it reflects well on them, conformity pressure within a group of raters, and demand effects created by how a question is framed. Cognitive sources include the plain limits of human information processing, reliance on a stereotype about what a job "should" involve rather than what it does, and the availability heuristic, rating a task as more significant simply because a vivid recent example of it comes easily to mind. None of this requires anyone to lie; it describes how ordinary social and cognitive processes distort a rating even when every rater is answering in good faith, which is harder to fix than dishonesty would be, because it cannot be solved by asking people to try harder.

Erich Dierdorff and Mark Wilson's 2003 meta-analysis in the Journal of Applied Psychology, aggregating forty-six studies and 299 separate reliability estimates, puts a number on how much this matters in practice. Analysts and job experts, trained analysts or supervisors among them, produced the highest inter-rater reliability; job incumbents, the people who actually do the work being rated, produced the lowest, for both tasks and broader work activities. Task-level data proved more reliable than ratings of broader work activities, a weighted correlation of roughly .77 against roughly .60. Put plainly: the people with the most direct, lived knowledge of a job are, on average across this literature, the least reliable source of ratings about it, and the broader and more abstract the rating category gets, the less two raters tend to agree even when rating the exact same job. Neither finding fits the common assumption that simply asking the jobholder is the most accurate way to learn what a job involves.

This does not mean job analysis data is worthless; a correlation of .60 to .77 across pooled studies, while far from perfect agreement, is also far from random noise, and well-designed job analysis still outperforms collecting no systematic data at all. It means findings should be used with the epistemic caution any measurement with documented, moderate reliability deserves, triangulated across more than one rater and, where possible, more than one method, rather than taken as a single incumbent's or supervisor's word treated as settled fact. An HR department that absorbed this evidence would build in a verification step, comparing an incumbent's description against a supervisor's independent rating, rather than filing whichever version arrived first.

A second honest limitation sits beside the reliability problem: even a job analysis done carefully has a shelf life, and most organizations have no process that tracks when it expires. A job description is almost always produced at a discrete moment, a role's creation, its recruitment, or an occasional multi-year classification review of the kind Lesson 33 covers. Between those moments, jobs change incrementally and informally: a colleague leaves and their remaining tasks get absorbed by whoever is available rather than through deliberate redesign, a new system replaces an old manual task that never formally leaves the description, a manager asks someone to take on "just this one thing" that quietly becomes routine. No single change feels large enough to justify reopening a formal job analysis, so the accumulation of small, undocumented drift is the normal condition of most job descriptions on file, not an unusual failure.

Sanchez and Levine's 2012 review in the Annual Review of Psychology frames this as part of a larger argument about why the field has had to adapt. Their claim is not that job analysis has become useless, but that the stable, well-bounded job the classical methods above were built to describe is less common than it once was; work is increasingly organized around fluid teams and roles redesigned around the person rather than fixed in advance, harder to capture with a method built on the assumption that a job is a fixed thing waiting to be measured. Their recommended response is not to abandon job analysis but to re-analyze jobs on a shorter cycle than most HR departments currently budget for, and to be explicit about how current the underlying data actually is.

What follows from that, stated plainly because it is advice most organizations do not follow, is this: a job description without a visible date, or without a named owner responsible for flagging when the role has materially changed, is not a stable record of the job. It is a snapshot whose age nobody is tracking, used as though it were current. Treating the document that way, rather than as a permanent file closed the day it was written, is the more defensible practice the evidence above actually supports.

Job analysis answers what a job involves and what it requires; it deliberately does not answer what the job is worth relative to other jobs in the organization, or what it should be paid. That question, job evaluation, and the related work of designing pay grades and a position classification structure, is Lesson 33's subject, not this one's, and the two should not be collapsed into a single exercise even though they draw on the same underlying data. A job evaluation committee ranking roles for a pay structure needs the job description and specification this lesson has covered to be detailed and comparable enough to support that ranking, which is exactly why the reliability and currency problems discussed above are not a narrow technical footnote. A compensation decision built on an unreliable or outdated job analysis inherits that unreliability directly; job evaluation can be run carefully and still produce a questionable ranking if the job analysis feeding it was not. The handoff runs one way: job evaluation cannot correct a bad job analysis, it can only rank whatever job analysis handed it, well done or badly.

Two limitations deserve to be stated directly rather than folded quietly into the sections above. First, nearly all of the reliability and accuracy research reviewed here, Morgeson and Campion's sixteen sources of inaccuracy and Dierdorff and Wilson's reliability meta-analysis included, comes from job analysis practice in large, formalized, mostly North American organizations with dedicated HR functions capable of running a multi-method study. Whether the same reliability patterns, incumbents rating least consistently, task data proving more reliable than broader activity ratings, hold in smaller organizations, informal labor markets, or settings with no trained analyst available to design the exercise in the first place, is not something this body of evidence was built to answer, and this lesson does not claim otherwise.

Second, a genuine tension sits underneath the method-choice material this lesson covers, and it does not resolve cleanly. The multi-method, cross-checked approach the evidence here supports, several raters, more than one method, periodic re-analysis rather than a one-time snapshot, is also the most expensive and slowest option, the fourth option in the Shetu Microfinance exercise above being the clearest illustration. An HR department given six weeks and a modest budget, which describes most HR departments most of the time, faces a real choice between a defensible process and a fast one, and nothing in the research reviewed here resolves that trade-off for them. It only clarifies what is being given up when speed wins, which is a more honest place to leave the question than a confident recommendation this lesson's evidence does not actually support.

Select a real job you can examine closely: your own current or a recent past role, a role held by someone you can interview with their consent, or a role documented in enough published or organizational detail to analyze responsibly without inventing information you do not actually have.

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

1. collects job analysis data using at least two distinct methods covered in this lesson (for example, a structured interview with the jobholder and a separate interview with their supervisor, or an interview combined with direct observation), and states explicitly where the two sources agreed and where they diverged;
2. produces a written job description, using specific, verifiable task statements rather than vague duty language, and a separate job specification listing KSAOs, with each listed requirement traced back to a specific duty it is derived from;
3. states whether your approach was closer to task-based or competency-based job analysis, using Sanchez and Levine's (2009) distinction, and justifies that choice against the purpose your analysis was meant to serve;
4. discusses, using Morgeson and Campion's (1997) framework and Dierdorff and Wilson's (2003) findings, at least two specific, plausible sources of inaccuracy or unreliability in your own data collection, rather than asserting your findings are accurate; and
5. assesses how current the role's existing job description, if one exists, actually was relative to what you found, and what that gap, or its absence, implies about the organization's practice of maintaining job documentation.

Use at least five academic sources, including at least three from this article's reference list. State explicitly where information was unavailable or where consent limited what could be examined, rather than filling the gap with assumption.

Assessment rubric (100 marks)

1. Dierdorff, E. C., & Wilson, M. A. (2003). A meta-analysis of job analysis reliability. Journal of Applied Psychology, 88(4), 635–646.
2. Fine, S. A., & Cronshaw, S. F. (1999). Functional Job Analysis: A Foundation for Human Resources Management. Lawrence Erlbaum Associates.
3. Levine, E. L., Ash, R. A., Hall, H., & Sistrunk, F. (1983). Evaluation of job analysis methods by experienced job analysts. Academy of Management Journal, 26(2), 339–348.
4. McCormick, E. J., Jeanneret, P. R., & Mecham, R. C. (1972). A study of job characteristics and job dimensions as based on the Position Analysis Questionnaire (PAQ). Journal of Applied Psychology, 56(4), 347–368.
5. Morgeson, F. P., & Campion, M. A. (1997). Social and cognitive sources of potential inaccuracy in job analysis. Journal of Applied Psychology, 82(5), 627–655.
6. Morgeson, F. P., & Campion, M. A. (2000). Accuracy in job analysis: Toward an inference-based model. Journal of Organizational Behavior, 21(7), 819–827.
7. Peterson, N. G., Mumford, M. D., Borman, W. C., Jeanneret, P. R., & Fleishman, E. A. (Eds.). (1999). An Occupational Information System for the 21st Century: The Development of O*NET. American Psychological Association.
8. Sanchez, J. I., & Levine, E. L. (2009). What is (or should be) the difference between competency modeling and traditional job analysis? Human Resource Management Review, 19(2), 53–63.
9. Sanchez, J. I., & Levine, E. L. (2012). The rise and fall of job analysis and the future of work analysis. Annual Review of Psychology, 63, 397–425.

Verification notes: every reference above was checked against publicly indexed bibliographic records (publisher and journal pages, APA PsycNET, Wiley, SAGE, ResearchGate-hosted abstracts, and Google Scholar citation records) via web search rather than a full-text read of each paper. Findings attributed to McCormick, Jeanneret and Mecham (1972), Fine and Cronshaw (1999), Levine, Ash, Hall and Sistrunk (1983), Morgeson and Campion (1997, 2000), Dierdorff and Wilson (2003), Sanchez and Levine (2009, 2012), and Peterson, Mumford, Borman, Jeanneret and Fleishman (1999) are reported at the level of detail available in abstracts, publisher summaries and secondary accounts, cross-checked across at least two independent sources where possible; no numeric figure in the article text, the PAQ's 195 items and six divisions, Dierdorff and Wilson's 46 studies and 299 estimates with weighted reliabilities of approximately .77 for tasks versus .60 for broader activities, or Morgeson and Campion's sixteen sources of inaccuracy, goes beyond what surfaced in those sources. The Shetu Microfinance case, the comparison table, the decision exercise, and the synthesis connecting this lesson to Lesson 31 (workforce planning) and forward to Lesson 33 (job evaluation and position classification) are original teaching material for this series.

Lesson 33 turns to what job analysis data is used for once it exists: ranking and pricing jobs against each other through job evaluation, and building the broader position classification structure a pay system depends on.