Skip to content

Multiple Methods of Assessment

Principle · All principles
Evidence · 43 claims (23 for, 18 mixed, 2 against) · 49 studies (12 causal, 10 review, 7 theoretical, 6 associational, 5 quant-synthesis, 5 qualitative, 4 design), q1–q4 · 5 of 49 report an effect size · 37 claims rest on one study

Conditional relationship

This page owns how several kinds of evidence about one learning goal are planned, read together and turned into a decision. When a teacher or designer must judge whether a learner has reached a goal, a judgement drawn from two or more methods that call for the same capability in different ways (a written answer and a short oral explanation; a product and the learner's account of the decisions behind it; a test and an observed performance; the teacher's, the learner's own and a peer's reading of the same work against one set of criteria) is expected to be more accurate (assessment-accuracy) than a judgement drawn from one method, when the learner's result on that one method depends on something other than the goal: reading and writing load, a symbol system the learner does not use, test anxiety, an unfamiliar context, or a capability that only shows with help. Where learners also get a bounded choice between formats held to one standard, the expected further gain is in motivation (motivation-affect), not in achievement. Adding peer and self-assessment as methods can also raise later performance (general-achievement), because learners apply the criteria, not because the evidence becomes more accurate.

The relationship is conditional on four things: every method must sample the same goal (a poster and a written explanation of photosynthesis may test drawing and writing more than photosynthesis); the criteria must be common across methods and the people applying them calibrated, or the variation added is noise; the format must not itself be the goal (if the goal is writing an argued essay, an oral alternative removes the construct rather than a barrier); and a second method costs time for the learner and the marker, so it pays only where the first is likely to mislead for this learner or where the decision has stakes.

No claim in the wiki tests the relationship as a whole. None compares decisions made from several methods with decisions made from one, against an independent measure of what learners can do, and none compares a choice of format with an assigned format on achievement. The nearest claims argue or illustrate that one format can hide what a learner knows (a review's argument about verbal tests, a design study of family-grounded tasks with young children, a theoretical argument for assisted assessment, and one small qualitative study in which a post-interview with props surfaced in 12 young children an understanding that a multiple-choice post-test showed in 2), show that two modes of one skill are not interchangeable (silent and oral reading in dyslexia), show that peer assessment raises performance in a meta-analysis, show that autonomy support raises motivation, and limit the relationship (untrained self-assessment is inaccurate; raters differ in severity; matching format to a "learning style" does not help). The design below is therefore a labelled default.

How the converted siblings sit beside this: Universal Design for Learning owns the planning audit that offers several ways to show an outcome, held to one standard, so that a barrier does not stop a learner; this page owns what is done with the evidence once several methods exist: whether they agree, what to do when they do not, and when a second method is worth its cost. Validity, Reliability and Bias in Classroom Assessment owns the measurement properties of any one score; this page uses them and adds the combination of scores. Assessment for Learning owns how evidence is used during instruction, with shared criteria and a revision step; Competency-Based Assessment owns the threshold judgement and assessor agreement for a "demonstrated / not yet" decision. Autonomy owns choice within structure. Summative Assessment and Standardized Test Fairness and Bias hold end-of-sequence judgement and large-scale cases. The Authentic Assessment pattern is one design that uses a performance plus feedback and revision as the unit's assessment; the Formative Assessment pattern is the reusable loop for reading a response and acting on it.

Default design, while the relationship is untested

The earlier page's guidance (clear targets, comparable criteria, intentional evidence design, use of results, and its four examples), kept as a concrete default a designer can act on and revise. Each step is a design proposal; its evidence status is stated beside it.

  1. Name one goal per judgement, then decide which formats are part of it. Write the capability as the learner will show it (verb, content, conditions). Mark which response modes belong to the goal (writing for a written argument; speaking for an oral presentation) and which do not (writing for a science concept). Only modes that do not belong to the goal may vary. Untested; from the earlier page ("clear learning targets"; "some skills still require specific formats").
  2. Choose two methods per goal, rarely more than three, chosen to fail differently. Pair a method that depends on reading and writing with one that does not (a short oral explanation, a demonstration, a labelled drawing, an observed task), or a product with the learner's account of the decisions behind it. A single verbal test can underestimate what some learners know (Typical verbal achievement tests may underestimate students' subject-matter knowledge because of a mismatch between the test's symbolic form and students' symbolic encoding [+W], an argument in a review, not a tested result; in one classroom study with young children, an interview with props and gestures showed an understanding that a multiple-choice test missed in ten of twelve children: Post-interviews with props and embodiment revealed understanding of macro–micro particle changes that the multiple-choice post-test did not capture. [+W], descriptive, with no independent check of which method was right). The number of methods is an untested proposal; the earlier page warned against "assessment sprawl".
  3. Write one set of criteria for all methods, with an example of each level in each format. Learners see the criteria and examples before they start. Markers score one example in each format together before marking, and double-mark a sample of about one in five (an untested proportion). Self-assessing against a rubric improved work in two non-randomised studies (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]); raters still differ in severity (Raters differed significantly in severity/leniency in the MFRM analysis [~M], open-ended items in one test).
  4. Read the methods together, with a stated rule. If the methods agree, act on the result. If they disagree, do not average them: ask which demand differs between them (language, speed, context, help available) and give a short third check that removes that demand (2–5 minutes, one item). Record which method each decision rests on. The rule is an untested proposal; what to look for is in the observation table below.
  5. Add self- and peer assessment as methods only after learners have practised judging against the criteria. Show a model of someone assessing work against the criteria, have learners judge two or three examples and compare their judgement with the teacher's, then use their judgements. Untrained self-assessment is inaccurate and tends to overestimate (Self Assessment Accuracy Is Low Without Training [~M]); peer assessment raised performance by a small average amount in a meta-analysis (Peer assessment improves academic performance by a small average amount relative to no assessment or to teacher assessment [+M]). Treat learners' judgements as evidence about their calibration as well as about the work; do not use untrained peer grades in a final mark.
  6. Where learners choose a format, bound the choice. Offer two or three formats, each listed with what it must contain and the common criteria; check before release that each option demands the same capability at the same level, and that none is visibly easier or better supported. Bounded choice can raise motivation (Self-determination theory holds that autonomy support raises intrinsic motivation, and a meta-analysis finds expected tangible rewards undermine it [+M], a narrative review and a meta-analysis on rewards; not a test of assessment choice). Do not assign formats by a learner's "style" (Learning Styles Matching Does Not Improve Learning [-M], tested for instruction, carried to assessment format). From the earlier page ("uneven rigor", "learner agency in assessment").
  7. Feed every method's result into the next instructional step. After each round, name the one goal most learners have not shown in any method and plan the next lesson around it; do not collect evidence no one will use. Untested; from the earlier page ("use of results").
  8. Plan across the unit, not one task. A default arc, all quantities untested proposals:
  9. Early sessions (first third of the unit): two low-stakes methods per session, each 3–5 minutes: a written exit item and a spoken or drawn one ("explain to a partner", a sketch with labels). The teacher chooses the methods; criteria and examples are on display.
  10. Middle sessions: learners practise self-assessment on one piece of work a week against the criteria and compare it with the teacher's reading; one peer review with the criteria; one task where learners choose between two or three formats.
  11. Late sessions: the summative judgement rests on at least two methods for each goal: one common format where the format belongs to the goal, plus one chosen format or an observed performance; a learner whose methods disagree near the cut gets a third, short check before the decision.
  12. Moving on: a learner moves to the next goal once two different methods both show the goal at the criterion; one method at criterion and one below means a targeted third check, not a repeat of the same test.

When a condition is not met: if no marker time is available for a second method, use the second method only for learners whose first result is near a decision point or out of line with their other work; if the format is part of the goal, keep it for everyone and vary only support, time or context.

Fitting the design to a situation

The three facts that most change the decision:

  • Which response modes belong to the goal. If the goal includes writing, speaking or performing in a particular way, that mode stays fixed and only other demands vary; if it does not, the format is free to change. If unstated, ask: if a learner showed this capability by speaking, drawing or doing instead of writing, would you accept it as the same capability?
  • What the result will decide, and how much rides on it. A check that steers tomorrow's lesson needs one quick method and a second only where it misleads; a grade, placement or certification near a cut needs at least two methods and calibrated markers. If unstated, ask: what decision will this result change, for whom, and can it be reversed?
  • Who can collect and read a second kind of evidence, and how much time they have. An oral check costs teacher time per learner; peer and self-assessment cost learner time and training; a portfolio costs marking time. If unstated, ask: who will look at the second piece of evidence, for how many learners, and how many minutes each?
If the brief says… Then change… Basis
Learners are novice on the goal and new to judging quality Teacher-chosen methods; criteria with an example of each level; postpone learner choice of format and self-assessment that counts until learners have practised judging two or three examples Self Assessment Accuracy Is Low Without Training [~M] (secondary students, one randomised experiment; a higher-education meta-analysis read as an abstract)
Learners are advanced, or adults with applied experience Let them propose a method and the evidence it will produce, checked against the criteria before they start; accept work products from their practice untested; from the earlier page (adults with rich applied experience)
Learners are young children (child, pre-school or early primary) Observe the goal in play or with objects and pictures, with a short interview ("show me", "why there?"), alongside any written task; record what each child did Preschool children demonstrate behavioral knowledge of classification, one-to-one correspondence, and seriation during everyday free play with everyday materials [+W] (one observed play episode, two children); Assessment tasks built on family-identified funds of knowledge elicited computational thinking from K-2 children in a small design-based study that made no comparison with standard assessment [+W] (13 K–2 children, design study, no comparison with a standard test); Post-interviews with props and embodiment revealed understanding of macro–micro particle changes that the multiple-choice post-test did not capture. [+W] (early-primary science in a mixed-reality unit; 12 of the children showed in the interview what 2 showed on the two hardest multiple-choice items; descriptive)
Learners have dyslexia or other reading difficulties, and the goal is reading Assess the mode of reading the goal names: if the goal is reading silently for understanding (most reading from adolescence on), include a silent-reading measure, not only an oral-fluency one; do not carry an oral score over to silent reading For children and adolescents with dyslexia, the silent reading deficit is as large as or larger than the oral reading deficit [+W] (two Dutch samples, children and adolescents; associational; the silent deficit was somewhat larger in adolescents)
Learners have low literacy, are learning in a second language, or have low confidence Pair every written method with a spoken, demonstrated or drawn one that does not need the same reading and writing; read instructions aloud; do not read a low written score as low knowledge until the second method agrees Typical verbal achievement tests may underestimate students' subject-matter knowledge because of a mismatch between the test's symbolic form and students' symbolic encoding [+W] (an argument in a review, not tested)
Learners are anxious about tests, or a timed written test is the only method Add an untimed or low-pressure second method (a conversation, a take-home task, an observed task) before judging; drop time limits unless speed is the goal Math anxiety has a small-to-moderate negative correlation with math achievement (r = −.28) in a meta-analysis of correlations from 223 studies [~M] (meta-analysis of correlations, r = −.28; does not test a change of format)
The goal is a complex-skill or performance (a lab, a clinical task, a design) Combine an observed performance or product with the learner's short written or spoken account of the decisions behind it; use one rubric for both untested; from the earlier page ("performance plus reflection"); the Authentic Assessment pattern holds the task design
The goal is verbal-association or a procedure One well-sampled method (many short items) is usually enough; add a second method only to check that a correct answer is not recognition alone untested proposal
A learner's methods disagree (strong in conversation, weak in writing; or the reverse) Give a short third check that removes the demand that differs; for a capability that shows only with help, record what the learner did with one hint, and judge the unaided result Static individual assessment misses developing functions; ZPD-based assessment reveals potential competencies [+W] (a theoretical argument); Contingent scaffolding improves learning more than fixed or absent support. [~M] (a meta-analysis of 30 dynamic-assessment studies found forms of assisted assessment differ: strategy training above contingent prompting above coaching; it compares forms of help, not assisted with unaided judgement)
The brief names its own assessment routine (its own rubric, portfolio, check-in or exam) Keep it; add only a second method for the goals the routine samples in one mode, and write the routine's criteria so they apply to both untested proposal
Stakes are high: a grade boundary, pass mark or certification Rest the decision on at least two methods per goal; calibrate markers on examples in each format and double-mark a sample; give a learner near the cut a third method before deciding; keep untrained peer judgements out of the mark Raters differed significantly in severity/leniency in the MFRM analysis [~M] (three raters, open-ended items); 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] (one class)
The result decides grade promotion or retention (a reading or test-score gate) Do not let one test on one day decide: give a second route to the standard (an alternative test, a portfolio of classwork against the same criteria) and a third check for a learner near the cut; record which route each decision rests on, and plan the support that follows either decision, since the decision itself is not the intervention Waivers promoted about one-third of third graders and 40% of sixth and eighth graders who failed the 1997 promotional criteria, keeping retention rates well below failure rates [~W] (Chicago: discretionary routes overrode the test for many learners, with no check of whether they chose better); Third-grade retention under the Chicago policy produced no sustained academic benefits or substantial negative effects [~M] (what follows the decision; neither study tests a multi-method gate)
Learners are offered a choice of format Two or three options, one rubric, each option checked for equal demand before release; do not match formats to "learning styles" Self-determination theory holds that autonomy support raises intrinsic motivation, and a meta-analysis finds expected tangible rewards undermine it [+M] (motivation, not achievement); Learning Styles Matching Does Not Improve Learning [-M] (tested for instruction, carried to assessment)
One teacher for a large group, or scarce marker time (volunteers, a busy shift, a tutor who checks in briefly) Keep one quick common method for everyone; use the second method only for learners near a decision or out of line with their other work; use trained peer review for the second reading of products; a 2-minute oral check per learner at most Peer assessment improves academic performance by a small average amount relative to no assessment or to teacher assessment [+M] (meta-analysis of 54 studies; about performance, not the accuracy of peer judgements); the targeting rule is untested
Time is one session only Use two methods inside the session (a written item and a 1-minute spoken or drawn one) for the one goal that matters; use the result to steer the next step, not to grade untested proposal
Mixed modes: some learners in the room, some remote, some on phones only Choose methods every mode can do (a short spoken or recorded answer, a photo of a drawing or worked page, a poll); give remote and phone learners the same criteria and time; note the mode beside each result and compare across modes before combining untested proposal
Adults in work settings, or the goal is a job task Combine observation of the task at work (a checklist) with a short conversation about why each step was done, and a work product; use the workplace's own records where they show the goal untested proposal; from the earlier page (adults with applied experience)
The work is a product or performance for a real audience Judge it against the criteria separately from the audience's response; record the audience's response as its own observation, and add the learner's account of decisions as a second method untested proposal; the audience relationship is owned by Authentic Audiences and Purposes
online-self-paced, no teacher present Pair auto-scored items with one short constructed or recorded response per goal, reviewed by a peer with the criteria or sampled by a tutor; have the learner self-assess against the criteria with an example first untested proposal; self-assessment limit from Self Assessment Accuracy Is Low Without Training [~M]
Little time (one or two sessions) and one assessor Use two methods per decision, not three: the one that shows the performance (an observed task that changes the case, such as a changed order) decides, and the quick one (a quiz) screens; run them as a rotation so the assessor observes while the others work; write an evidence matrix in advance saying what each disagreement leads to (passes quiz, fails task → targeted practice and reassess; the reverse → a short oral check) untested proposal

Eleven of the twenty rows cite a claim. None of those claims tests a combination of methods against one method against an independent measure of capability: two rows rest on meta-analyses of neighbouring questions (peer assessment and performance; anxiety and achievement), two on self-assessment training, one on two single rating studies, one on autonomy support and on learning styles (motivation, and instruction rather than assessment), three on arguments or small descriptive studies (verbal tests, assisted assessment with a dynamic-assessment meta-analysis beside it, young children, where one descriptive study did set an interview beside a multiple-choice test), one on a correlational comparison of oral and silent reading in dyslexia, and one on Chicago promotion-policy reports that describe a single-test gate and its waivers without testing a second route. Each is carried to this page's move by extrapolation. The other nine rows are untested proposals.

Observation, state and explanation

What can be observed is a response to a particular method under stated conditions: the task, the mode (written, spoken, drawn, performed), the time allowed, the help available, who scored it and against which criteria, and the score. "Understands the concept", "can do it but cannot write it" and "was helped" are inferred states. Two methods agreeing does not prove the inference; they may share a demand (both in English, both timed).

Response Plausible explanations Observation that could change the interpretation
Weak written answer, strong spoken explanation of the same goal Writing or reading load hides knowledge; the spoken exchange added cues or help; the spoken explanation is fluent but would not survive a precise question Give a written item with minimal reading load (labelled diagram, short answer) and a spoken item with no prompts; ask one precise follow-up question in each.
Strong on the test, weak on the performance task Recognition without use; the performance adds coordination, time pressure or an unfamiliar context; the performance rubric is scored more harshly Give a short performance on a familiar case; have a second marker score the performance; ask the learner to talk through one decision.
Self-assessment much higher than the teacher's reading Untrained or miscalibrated self-judgement; the learner reads the criteria differently; the teacher's reading missed something Ask the learner to point to the evidence for each criterion in their work; compare their judgement of an example with yours.
Peer and teacher readings disagree Peer lacks domain knowledge; criteria are ambiguous; teacher's severity differs from the group's Have both mark a common example; check where on the rubric the readings split.
Learners who chose one format score systematically lower That option is harder or less supported; weaker learners chose it; markers judge that format more harshly Compare the same learners' results in a common format; have the options double-marked across formats.
Weak on multiple choice, strong in an interview with props or gestures (common with young children) The written items' wording or reading load hides understanding; the props and the interviewer's questions supplied part of the reasoning; the interview asked an easier version of the question Ask the interview question without props, then with; give a multiple-choice item read aloud with a picture; compare the two on the same content.
Correct only with a hint The capability is developing; the hint carried the decision; the task was unclear Give a matched item unaided with clearer instructions; record the level of hint needed across two occasions.

These rows are proposals, untested with learners as diagnostics. Each extra check is itself practice and changes what it measures, so record it, and keep explanations that remain open rather than choosing the one whose name fits.

Evidence and qualifications

No claim in the wiki compares judgements made from several methods with judgements made from one, or compares a choice of assessment format with an assigned one on learning. The claims below bear on parts of the relationship.

Learners, settings and outcomes differ across these claims (school and university peer assessment, pre-university self-assessment, children in play and in an embodied science unit, readers with dyslexia, a city's promotion gate, reviews of testing), and only four print an effect size, each on a different outcome, so they cannot be ranked against each other.

The evidence by question

Further evidence, not yet read against this model

Relevant claims read against the model that do not change it, and off-topic claims kept from merged pages. None tests multiple methods of assessment.

Objective and learner-valued goal

The designer's objective is a decision about a learner (what to teach next, a grade, a certificate) that matches what the learner can do with the goal, in the conditions that matter. Elicit separately what the learner wants from the assessment: to show what they know in a way that feels fair, to get the grade with least risk, to have a product they can use or show, to avoid speaking in front of others. Agreement and divergence are both observations. A learner who values a safe grade may pick the format they think is easiest, which is evidence about the options' equality as well as about the learner; one who dreads speaking may do worse orally for reasons that are not the goal.

For example, an adult apprentice electrician may want a portfolio of real installations to show future employers, while the course objective is passing a written regulations exam. A photographed installation with a short spoken account of the regulations it follows serves the apprentice's goal and gives a second method for the regulations goal; the written exam stays where the qualification requires it. Record both, so that a strong portfolio is not read as exam readiness, or a weak exam as weak practice.

What would revise this model?

The expectation should weaken if studies comparing decisions from several methods with decisions from one, against an independent measure of capability, show no gain in accuracy, or show that the added rater variation outweighs the added evidence. If learners given a choice of format, held to one rubric, reach the goal no more often and are no more motivated than learners assigned one format, the choice step is a preference only. If calibrated peer judgements prove as accurate as teachers' for a kind of task, peer review can carry more of the second reading. If a single well-designed method with low language load predicts later performance as well as two methods, the second method is a cost to cut. Do not protect the model by calling every disagreement between methods "evidence of hidden knowledge"; sometimes the lower result is the accurate one.

Agreement between methods, accuracy of a decision, later achievement and motivation are separate outcomes. The present evidence establishes no number of methods, no rule for combining them and no evidence that a choice of format improves achievement.

Claims

Examples

Key Sources

  • Andrade, H. L., & Brookhart, S. M. (2020). Classroom assessment as the co-regulation of learning. Assessment in Education: Principles, Policy & Practice, 27(4), 350-372. https://doi.org/10.1080/0969594X.2019.1571992
  • Shepard, L. A., Penuel, W. R., & Pellegrino, J. W. (2018). Using learning and motivation theories to coherently link formative assessment, grading practices, and large-scale assessment. Educational Measurement: Issues and Practice, 37(1), 21-34. https://doi.org/10.1111/emip.12189
  • Rao, K. (2021). Inclusive instructional design: Applying UDL to online learning. The Journal of Applied Instructional Design, 10(1). https://doi.org/10.51869/101/kr
  • Costley, K. C. (2012). An Overview of the Life, Central Concepts, Including Classroom Applications of Lev Vygotsky. https://eric.ed.gov/?id=ED529565
  • Comprehensive Early Literacy Policy: Fundamental Principles. (2024). ExcelinEd Policy Toolkit. https://www.excelined.org
Page metadata
Field Value
type principle
id multiple-methods-of-assessment
aliases assess-alone-and-assisted-performance, test-based-retention-multiple-promotion-options
title Multiple Methods of Assessment
description When one assessment method can mislead about a learner (through reading load, an unfamiliar symbol system, anxiety or a capability that shows only with help), a judgement drawn from two or more methods sampling the same goal under common, calibrated criteria is expected to be more accurate than one drawn from a single method, and bounded choice of format to raise motivation rather than achievement; no claim here tests the relationship as a whole.
status review
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content
sources 5 — andrade-2020, shepard-2018, rao-2021, costley-2012, comprehensive-early-literacy-policy-fundamental-principles-2024

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