Skip to content

Competency-Based Assessment

Principle · All principles
Evidence · 36 claims (8 for, 26 mixed, 2 against) · 36 studies (8 causal, 8 theoretical, 7 design, 6 associational, 4 review, 2 qualitative, 1 quant-synthesis), q1–q4 · 2 of 36 report an effect size · 31 claims rest on one study

Conditional relationship

Competency-based assessment states a relationship about judgement, not first about learning: when a learner's response to a task that represents the intended capability is scored against explicit criteria and a stated threshold, by assessors whose scoring agrees, the resulting judgement ("demonstrated" or "not yet") is expected to support better decisions about that learner than a norm-referenced rank or a record of time spent. The decisions are what to teach next, whether to advance, and whether to certify. A second, learning relationship rides on the first: when criteria are visible to learners and a "not yet" leads to correction and another comparable attempt, the quality of the learner's work and later examination performance (general-achievement) are expected to improve. Both are conditional on four things a page cannot settle in advance: whether the task elicits the capability itself rather than a proxy for it (most of all for complex-skill goals in workplace-clinical settings), whether the threshold was set by a defensible procedure, whether assessors are consistent and calibrated, and whether the learner's starting response reflects capability or the conditions of the attempt.

The wider instructional model (pacing, progression and pathways around competencies) is Competency-based Learning & Assessment, and the Competency-Based Learning pattern gives a reusable design. Two converted siblings sit inside this page's relationship. Mastery Learning holds the conditional model of the progression rule (gate, correct, recheck); this page owns what the gate rests on, a criterion-referenced judgement, and does not repeat the gate's model. Formative Assessment holds the model of interpreting a response and choosing a next action; a competency judgement is one such response read against a public standard, and the formative model applies to everything below the threshold.

Default design, while the relationship is untested

The page's earlier guidance, kept as a concrete default a designer can act on and revise. Each step is a design proposal with its evidence status beside it; most is untested as written.

  1. State each competency as an observable performance under stated conditions (task, aids, time, setting), specific enough that two assessors would agree on what counts. Untested here; from the earlier page. The bearing evidence is that assessors differ when it is not done well (see Evidence, below), and one design demonstration in which competencies in a new area (students' use of generative AI in introductory physics) were stated so that they could be observed in students' work and scored with rubrics anchored in an expert answer key: AI interaction competencies observable and scorable [+W]. It shows the step can be done in one course, not that it makes judgements agree.
  2. Choose a task that elicits the competence itself, not a correlate of it. For a complex-skill goal, prefer an integrated performance that resembles real use (a live scenario, an observed lesson) over an explanation of the steps on paper. Untested here. The nearest claim, a training-design argument that practising whole complex tasks improves transfer [~W], is about how to train, not how to assess, and is itself an argument from prior research rather than a comparison of whole-task with part-task practice; it is carried here only as a reason to sample the integrated performance.
  3. Write the threshold down, and set it by a stated procedure. Say what separates "not yet demonstrated" from "demonstrated", and if a panel sets the cut, use a method robust to one deviant rating. One standard-setting report bears on the method (see Evidence); none tests whether a given threshold predicts later performance.
  4. Share the criteria with learners before the attempt, with an exemplar and practice in applying them. Evidence: two comparison-group studies of rubric-referenced self-assessment (see Evidence), with a cost noted beside them. A narrative review of goal-setting research holds that specific, difficult goals lead to higher performance than vague "do your best" goals, provided the person is committed and able: Specific, difficult goals lead to higher performance [+W]. That is a reason to make the criterion specific, carried from goals to assessment criteria; it does not test visible criteria.
  5. Calibrate assessors and check their agreement before relying on a judgement. Use a shared rubric, score common samples together, and look at the spread between assessors on the same work. Evidence: two studies show assessors on one rubric can differ substantially; neither tests calibration as a remedy.
  6. Allow reassessment after correction, on a comparable but changed task, so a judgement reflects capability rather than timing. Evidence: the mastery-learning synthesis (see Evidence) supports the correct-and-recheck loop at programme level; the gate's own model is on Mastery Learning. Where an earlier attempt was supported, withdraw the support across attempts before the judgement that certifies: fading support promotes the transfer of responsibility to the learner [~M] (worked-example fading favoured near transfer in three experiments; carried here from instruction to assessment, untested there).
  7. Record what the judgement does not cover: the aids allowed, the attempt number, the assessor, and the horizon. A "demonstrated" today is not evidence of delayed-retention or of performance in the learner's later setting. Untested here; from the earlier page.

Fitting the design to a situation

Ask for four facts when the brief omits them: what the judgement decides (feedback, progression, a credential), how many assessors score each performance, how the threshold will be set, and whether reassessment is allowed.

If the brief says… Then change… Basis
Several assessors will score the same kind of performance Calibrate them on common samples and check their spread before relying on a judgement; raters on one scale differed in severity Raters differed in severity [~M]; Engineering rubric scorers differed by 1.2 of 9 points [-W]
A panel will set the pass threshold Use a cut-score method that resists one deviant rating, and record the method Mean-classification cut-score method more robust to outliers [~W]
AI will score responses Write the required conditions into the rubric explicitly, keep human review of borderline and high-stakes decisions, and check agreement with human scorers Explicit rubric conditions reduced AI disagreement [~W]; AI not yet best practice for competency assessment [~W]
Learners should improve, not only be certified Share the criteria with an exemplar before the attempt and have learners self-assess a draft against them Rubric-referenced self-assessment improved work [~M]
The judgement controls progression to the next unit Allow reassessment after correction on a comparable but changed task Mastery learning improves outcomes [+M]
A test places learners into levels Check measurement precision at the cut scores that matter, not only overall reliability Placement test precision declines above the 40th percentile [~W]
The test was built by picking items that spread learners out Choose items that represent the competency domain, not those that discriminate best between learners Statistical item selection conflicts with criterion-referenced definitions [~W]; Criterion-referenced tests need construct validation [~W]
Checks must be short Use an adaptive mastery test; in one military-training simulation it shortened tests by 30–81% with the same decision for 96% of trainees Adaptive mastery testing reduces length and preserves decisions [~W]
Learners whose capability is still emerging (young children, early stages) Add an assisted attempt beside the unassisted one, and record what help made the difference Static assessment misses developing functions [~W]
Learners are novices on the task being judged Frame early attempts and feedback around the process criteria (the steps and strategy), and judge against the outcome threshold once the process is in place; withdraw prompts across attempts before the judgement that counts Process goals outperform outcome goals for novices [~M] (high-school writers learning one revision skill; a goal-setting finding carried to assessment criteria); Fading support transfers responsibility [~M]
Drafts need frank feedback, but every judgement also counts toward the grade (writing courses) Consider separating feedback on drafts from the grade up to a stated level, for example a grade guaranteed for completing specified work, with judged quality deciding only the grades above it; record that this credits completion, not demonstrated capability Authors' account of their own writing courses, untested: a contract guarantees a B for completed activities [~W]; decoupling evaluation from grades eases feedback [~W]; contract grading yields more student work [~W]; fourteen weeks of contract work warrants a B [~W]
Practising professionals will be certified on portfolios or observed practice (teachers) Say which later outcome the certification is meant to predict, and check it against that outcome; a credential can agree with observation scores and not with student-learning measures Board certification identifies more expert teachers [+W]; Certified teachers differ only on observation scores, not value added [~M]
Graduation or certification will rest on performance assessments Give learners more than one occasion to demonstrate each competency, and record the conditions of each Performance assessments expand opportunities to demonstrate mastery [~W]

Observation, state and explanation

What an assessor observes is a response to one task under stated conditions (task, aids, time, attempt number, assessor, prior exposures), scored against a criterion. "Competent" and "not yet competent" are inferences from that score, and they depend on the task sampling the capability, the threshold, and the assessor. A score from a lenient assessor and one from a severe assessor are not the same observation. "Novice" means novice on this task, not in the field.

Response Plausible explanations Observation that could change the interpretation
Below threshold on a performance task, but explains the steps correctly Capability not yet integrated under task conditions; the task adds demands the competency does not include (time, equipment, language); assessor severity Re-score the same performance with a second, calibrated assessor; give a component probe and a whole task with the extraneous demand removed; record which steps failed.
Meets threshold on one attempt, fails a comparable task later Narrow task sampled only part of the competency; forgetting; first task resembled practice items; first assessor lenient Compare the two tasks' demands; check the first score against a second assessor; re-probe at a stated delay on a changed task.
Two assessors disagree on the same work Criterion wording admits several readings; assessors weight criteria differently; one is more severe; the work is genuinely borderline Have both score shared anchor samples and explain each score against the criterion; see whether disagreement is spread or concentrated on one criterion or one assessor.
Meets threshold only with support (prompts, a checklist, a partner) Capability is developing and shows only when assisted; support carries part of the performance; the competency as written includes the support Decide whether the support belongs to the competency; vary the support while holding the task constant; reassess unaided after a stated interval.
Repeated "not yet", then disengagement or withdrawal Correction repeats the original instruction; a prerequisite gap; the threshold is above what instruction teaches; the judgement is experienced as threat; the learner's goal differs from the competency Change the form of correction and observe the next attempt; probe a prerequisite; ask what the learner wants from the course; check how other learners fare on the same task and threshold.

These rows are proposals for interpreting responses, untested with learners. A probe can shift confidence between explanations; one contrast does not identify a cause, and a reassessment is itself an exposure that can change the response. Record what remains unresolved rather than recording "competent" or "not yet" alone.

Evidence and qualifications

No claim in the wiki tests this relationship as stated: none compares decisions or learning under criterion-referenced competency judgements against norm-referenced grading or seat-time credit. The claims below test neighbouring parts of it; of the six read at conversion, none has been read at full text with an effect size.

  • Self-assessing against a rubric improved the quality of student work in two non-randomised studies, one of which also found more stress and performance-avoidance [~M]: two comparison-group studies (controlled-nonrandom as far as the claim page reports). Middle-school writers (n = 162) who reviewed a model essay, generated criteria and self-assessed a draft against a rubric scored higher on total essay scores and on every rubric criterion than a comparison condition that generated criteria and reviewed drafts without the rubric step. Pre-service teachers (n = 218) who self-assessed a concept map with a rubric produced work an expert rated higher, and self-assessed more accurately, than those with no tool, but also reported more difficulty coping with stress and a higher performance/avoidance orientation, which the authors judge detrimental to learning. No effect sizes; horizon immediate, one task each. One of two entries passes the judge, on abstracts. It supports making criteria visible and practising their use (step 4); it does not show that a competency threshold improves learning, and it carries a motivational cost to watch for.
  • Mastery Learning Improves Outcomes [+M]: a meta-analysis of 108 controlled evaluations (synthesis-experimental) at college, high-school and upper-elementary levels reports positive effects of mastery programmes on examination performance against conventionally paced instruction, apparently stronger for weaker students, varying with procedure, design and content; positive effects on attitudes; possibly more time on instructional tasks; and, for self-paced college programmes, often reduced completion. No pooled effect size recorded; checked by the judge on the abstract only. It bears on judging against a criterion and re-attempting rather than averaging, not on the quality of the judgement itself, and it does not predict one learner's gain or a delayed outcome.
  • Raters differed significantly in severity/leniency in the MFRM analysis [~M]: in a many-facet Rasch analysis of three raters scoring open-ended items in a crossed design (correlational), the rater facet was significant, so the raters differed in severity, while item and person reliabilities stayed above .80. Not yet checked against its source. It qualifies the relationship: a judgement depends on who scored it, and high reliability in one facet does not remove differences in another.
  • In one community-college lab class, six scorers rating six teams with an engineering rubric differed by a mean of 1.2 of 9 points, which the authors read as a sign the rubric may not be reliable [-W]: six scorers rated six teams on one inquiry-lab rubric; the mean class standard deviation was 1.2 of 9 points, which the authors took to mean the rubric may not have been reliable. One small design report, not yet checked. It limits the relationship: having a rubric does not by itself make judgements consistent.
  • A mean-classification cut-score method outperformed a contrasting-groups-style mean method because it was more robust to outliers [~W]: a standard-setting report for alternate portfolio assessments of students with significant disabilities applied two methods of computing a cut score to panelists' ratings of 15 portfolios per grade and content area; one was heavily influenced by deviant ratings, the other more robust. Not yet checked. It bears on how a threshold is computed from panel judgements (step 3), not on whether any threshold is valid for a decision.
  • Static individual assessment misses developing functions; ZPD-based assessment reveals potential competencies [~W]: a theoretical argument in a review paper, attributed to Vygotsky, that unassisted individual testing shows only matured functions, while assisted performance reveals developing ones. No data. It qualifies single-attempt, unassisted judgements: a "not yet" may misread a developing capability, which is a reason to record support and allow reassessment, not evidence that assisted scores should certify.

Keep the population, setting, comparator, instrument and horizon with each finding. Do not combine these into one strength: two are about learners' work, one about programme outcomes, and three about the judgement itself.

Read since conversion (2026-10-15), none of which tests the relationship either:

How far each may be carried: the rubric studies are self-assessment with one rubric on one task, so they do not show that assessor-scored competency rubrics help learning; the mastery synthesis is about unit gates in courses, not certification of workplace competence; the two reliability claims are single instruments and do not estimate agreement for any other rubric; the cut-score report is one population and one panel procedure.

The evidence by question

Further evidence, not yet read against this model

Re-read 2026-10-15: eleven of the claims listed here were worked into the model above. These two were left because they bear only on the comparator, not on the relationship.

Objective and learner-valued goal

The designer's objective names the competency, the task that represents it, the criterion and threshold, the permitted aids and the horizon at which the judgement is meant to hold. Elicit separately what the learner wants from the assessment: a credential for a job, recognition of experience they already have, confidence that they can do the work unsupervised, or simply to finish. Agreement and divergence are both observations, and a learner may value the credential more than the capability or the reverse.

For example, a nurse returning to practice may value being trusted to run a medication round alone, while the programme's objective is a pass on a simulated medication-safety scenario scored on a checklist. A pass serves the programme's decision; whether it serves the learner's aim depends on whether the scenario includes the interruptions and judgement calls of a real round, and on a later observation in practice. Recording both keeps a scenario pass from being read as the capability the learner came for, and tells the designer where the task may be a proxy.

What would revise this model?

The model should weaken if criterion-referenced judgements, made with calibrated assessors on representative tasks, predict later performance in the target setting no better than norm-referenced grades or time-based credit, or if they lead to no better decisions about what to teach next. If assessor disagreement on well-specified rubrics stays large after calibration, the threshold cannot carry the weight the principle puts on it, and judgements should be reported with their uncertainty or replaced by multiple assessors and attempts. If making criteria visible reliably raises stress or performance-avoidance enough to lower learning, as one rubric study suggests it can, the default design's step 4 needs a different form. Do not protect the model by calling every inconsistent judgement a poorly written rubric.

A score on a rubric, agreement between assessors, a threshold's defensibility, learning gains from the assessment process, and performance in the learner's real setting are separate claims. The present evidence does not establish how to set a threshold for a given decision, how much calibration produces acceptable agreement, or whether competency judgements transfer across settings.

Claims

Examples

Key Sources

  • Le, C., Wolfe, R. E., & Steinberg, A. (2014). The past and the promise: Today's competency education movement. Jobs for the Future.
  • Henri, M., Johnson, M. D., & Nepal, B. (2017). A review of competency-based learning: Tools, assessments, and recommendations. Journal of Engineering Education, 106(4), 607-638. doi:10.1002/jee.20180
  • Collins, J. B., Harsy, A., Hart, J., Haymaker, K. A., Hoofnagle, A. M., Kuyper Janssen, M., & O'Shaughnessy, J. (2019). Mastery-based testing in undergraduate mathematics courses. PRIMUS, 29(5), 441-460. doi:10.1080/10511970.2018.1488317
  • Shearer, R. L. (2017). Adaptive learning, competency-based education, and personalization: Implications for distance education and adult learners. Journal of Lifelong Learning Society, 13(1), 49-71.
  • National Board for Professional Teaching Standards' National Teacher Certification and a Performance-Based Assessment System. (2000). ERIC Digest. https://eric.ed.gov/
  • Fricke, B. G. (1976). Grading by Contract. Memo to the Faculty, n57. https://eric.ed.gov/?id=ED127854
Page metadata
Field Value
type principle
id competency-based-assessment
aliases competency-based-learning-assessment, performance-based-teacher-certification-assessment
title Competency-Based Assessment
description Scoring a learner's response on a task that represents the intended capability against explicit criteria and a stated threshold, by consistent assessors, is expected to support better teaching, progression and certification decisions than a rank or seat time, though no claim here tests that comparison and assessor agreement is not guaranteed.
status review
moved_from pattern/grading-by-contract-fricke
canonical True
generated by claude/unspecified on 2026-10-15
trust unverified — no one has confirmed this page's content
sources 3 — henri-2017, national-board-for-professional-teaching-standards-national-teacher-certification-and-a-performance-based-assessment-system-2000, fricke-1976

Frontmatter is the page's machine-readable record — see the Schema & Guide.