Horizontal vs. vertical equating are two core approaches used to place test scores from different forms or grade levels onto a common scale so the results can be interpreted fairly. In psychometrics, scaling refers to the statistical construction of score metrics, while equating refers to the process of adjusting scores so they are interchangeable within a defined use case. I have worked on K-12 assessment programs where confusion between these terms caused reporting mistakes, especially when teams assumed any score conversion table was an equating. It is not. A concordance, projection, moderation, and equating answer different questions, and the distinction matters for validity, accountability, and student decisions.
This topic matters because testing programs rarely stay static. Forms change across administrations for security. Blueprints evolve. Students take assessments across grade levels. Licensure and admissions programs need stable passing standards over time. Without sound scaling and equating, a score of 250 on one form might reflect a different level of proficiency than 250 on another. That undermines trend reporting, growth claims, and fairness. The Standards for Educational and Psychological Testing, along with guidance from professional organizations such as NCME, make clear that score comparability requires evidence, not assumption. The hub topic of Scaling and Equating exists for exactly this reason: to explain how psychometricians build common metrics, choose designs, evaluate assumptions, and communicate limitations in language decision-makers can trust.
At a practical level, the central question is simple: when can scores be treated as comparable, and what statistical method justifies that claim? Horizontal equating answers that question when test forms are intended to measure the same construct at the same difficulty level, usually for the same grade or population. Vertical equating addresses a different problem: linking tests across grade levels or developmental stages so results can support interpretations of growth. The methods overlap in some tools, especially item response theory, but their assumptions, risks, and interpretations differ. Understanding those differences is essential for anyone building score scales, reviewing technical manuals, or using assessment data to make instructional, accountability, or policy decisions.
What horizontal equating means and when to use it
Horizontal equating is used when two or more test forms are built to the same content specifications and are intended to be statistically interchangeable. In plain terms, Form A and Form B should measure the same construct, at the same level, with similar reliability, content balance, and administration conditions. Classic examples include alternate SAT administrations, state summative forms within a grade, or licensure forms delivered in different windows. The goal is not to show growth or progression. The goal is score comparability across parallel forms so that a candidate is neither helped nor harmed by the form they happened to receive.
In operational programs, horizontal equating often relies on common-item nonequivalent groups designs, random groups designs, or common-person designs. The most common in large-scale testing is the anchor-test design, where a stable set of items appears on multiple forms. Those anchor items provide the statistical bridge. If the anchor is representative of the full blueprint, secure enough to remain stable, and calibrated well, equating can adjust for slight differences in form difficulty. In my experience, anchor quality matters more than many nonpsychometric stakeholders realize. A short, speeded, or poorly targeted anchor can destabilize the entire conversion, even when the total test looks strong on paper.
Methods vary by framework. Under classical test theory, linear equating and equipercentile equating are common. Under item response theory, mean-sigma, mean-mean, Stocking-Lord, and Haebara linking methods are widely used before score transformations are applied. The choice depends on data volume, model fit, test length, population differences, and reporting requirements. For example, equipercentile methods can capture nonlinearity between forms, while linear methods are simpler but impose stronger assumptions. In IRT-based programs, I usually expect to see strong evidence for model-data fit, anchor drift checks, and sensitivity analyses before any equating table is approved for reporting.
What vertical equating means and why it is harder
Vertical equating, often called vertical scaling in school testing, links assessments built for different grade levels onto a common developmental scale. A grade 3 reading test and a grade 4 reading test are not parallel forms. They are designed to differ in difficulty because the underlying expectation is that students develop over time. The reason for linking them is to support statements about progress across grades, not interchangeability within a grade. That difference is fundamental. Horizontal equating says scores can be used as equivalents across alternate forms. Vertical equating says scores can be interpreted along a growth continuum, with important caveats about what the scale actually represents.
This is harder because developmental change is not just a statistical issue. The construct itself may shift as students mature and as the test blueprint changes. Early reading may emphasize decoding and literal comprehension, while later grades place more weight on vocabulary, inference, and analysis of complex texts. In mathematics, arithmetic fluency gives way to proportional reasoning and algebraic thinking. If the construct changes too much, a single vertical scale can become misleading, even if the linking model converges cleanly. I have seen vertical scales look elegant numerically while masking a messy content reality that made the growth interpretation much weaker than the technical report implied.
As a result, responsible vertical equating requires a documented theory of progression, carefully designed overlap in content and difficulty, and explicit statements about what inferences are justified. A vertical scale is most defensible when adjacent-grade forms share meaningful commonality, the item pool spans a broad difficulty range, and the model supports stable ordering of students across grades. Even then, equal intervals on the reported score scale do not automatically mean equal learning gains in the classroom. A ten-point gain from grade 3 to grade 4 may not represent the same educational change as a ten-point gain from grade 7 to grade 8. Users need that warning stated plainly.
Key differences between horizontal and vertical equating
The quickest way to distinguish the two approaches is to ask what kind of comparability the program needs. If different forms are meant to be substitutes for one another within the same target population, horizontal equating is the appropriate framework. If different tests are designed for successive levels of development and the program wants to describe growth, vertical equating is the better term. The psychometric machinery may share linking constants, anchor items, and scale transformations, but the interpretive claim is different, and that claim drives every technical decision.
| Dimension | Horizontal Equating | Vertical Equating |
|---|---|---|
| Primary purpose | Make alternate forms interchangeable | Support interpretation of growth across levels |
| Test relationship | Same construct, same grade, similar difficulty targets | Related construct across grades or developmental stages |
| Typical design | Common-item or random-groups forms | Adjacent-grade links with overlapping content and difficulty |
| Main risk | Form unfairness if difficulty differs | Overstating growth on a scale with shifting constructs |
| Interpretation | Scores are exchangeable within intended use | Scores indicate location on a developmental continuum |
Another difference is error structure. All equating introduces uncertainty, but vertical scales often accumulate more complexity because links span multiple grade transitions. Programs sometimes chain grade-to-grade links, and each link can contribute error. Scale drift, blueprint changes, and curriculum shifts can also distort long-range interpretations. That is why technical manuals should report conditional standard errors, subgroup analyses, and evaluations of scale stability over time. In short, horizontal equating asks, “Did two forms land on the same ruler?” Vertical equating asks, “Is this ruler itself meaningful across developmental stages?” The second question is broader and more fragile.
How scaling and linking methods support both approaches
Scaling and equating are often discussed together because both rely on a score metric, but they are not identical tasks. Scaling creates the metric. Equating places scores from different administrations or levels onto that metric under specific assumptions. In classical test theory, raw-to-scale conversions may be produced after equating through observed-score or true-score methods. In IRT, item parameters are estimated first, then aligned through linking, and finally transformed into reported scales such as 200 to 800 or 1000 to 2000. The reporting scale is often chosen for interpretability, not because the underlying theta metric has direct meaning to users.
For horizontal equating, common-item designs remain standard because they balance practicality and statistical control. Anchor items should mirror the total test in content and difficulty, avoid unusual format effects, and remain secure across administrations. For vertical equating, item pools need broader coverage so the model can locate lower-performing older students and higher-performing younger students on the same continuum without excessive extrapolation. Rasch, 2PL, and 3PL models may all be used, but model choice should reflect the item type, sample size, and operational purpose. IRT is not automatically better than classical methods; it is better when its assumptions fit the program and when parameter invariance is supported well enough to justify the links.
One practical issue rarely explained well outside technical circles is that linking is not always equating. A statistical transformation that places two calibrations onto a common metric may support score comparisons, but only equating justifies interchangeability, and only when the required conditions hold. Similarly, a vertical scale may be linked across grades without proving that every score difference represents equivalent educational growth. Careful programs separate these claims in documentation. They state what was linked, what was equated, what evidence supports the decision, and where interpretation should stop. That distinction protects users from taking a mathematically tidy scale too literally.
Design choices, quality checks, and common mistakes
Good equating begins long before any data file is analyzed. It starts with test design, blueprint alignment, representative anchors, administration consistency, and a predeclared analysis plan. For horizontal equating, the anchor set usually needs enough items to stabilize the link but not so many that exposure risk becomes unacceptable. For vertical equating, adjacent-grade overlap must be deliberate, not accidental. Content experts and psychometricians need to review whether the tests share enough substance to justify a common developmental interpretation. When that collaboration is weak, statistical convenience can outrun construct validity.
Quality checks should include item fit, differential item functioning, anchor drift, subgroup stability, standard error evaluation, and comparison of alternative equating methods. If results change materially when a handful of anchor items are removed, the link is fragile. If subgroup conversions differ sharply, the program may have fairness concerns. If equipercentile and IRT-based results diverge beyond operational tolerance, teams should investigate why rather than choosing the preferred answer. I have found that sensitivity analysis is one of the most persuasive tools in technical review because it reveals whether conclusions are robust or dependent on one narrow modeling choice.
Common mistakes recur across programs. Teams sometimes call a raw-to-scale conversion an equating when no alternate-form evidence exists. They sometimes use vertical scales to claim year-to-year growth even when the content progression is poorly defined. Others overinterpret scale score gaps as if the intervals have obvious instructional meaning. Another frequent problem is weak documentation. Decision-makers need to know the equating design, the anchor composition, the sample characteristics, and the standard errors. Without that information, a conversion table may look authoritative while hiding assumptions that materially affect fairness. Clear documentation is not a bureaucratic add-on; it is part of score validity.
How to interpret results in real testing programs
In real programs, the best interpretation depends on the assessment purpose. If a state uses alternate forms of a grade 5 science test across spring windows, horizontal equating supports fair proficiency decisions and trend reporting within that grade. If a district reports student growth from grade 4 through grade 8 mathematics, vertical equating can provide a common developmental metric, but the district should still avoid simplistic claims that equal score gains mean equal instructional impact at every grade. Growth percentiles, benchmark expectations, and content-level evidence often need to complement the vertical scale.
For score users, three questions are essential. First, were the tests intended to be interchangeable or developmental? Second, what design and evidence support the link? Third, what interpretations are explicitly allowed in the technical documentation? When those answers are clear, scaling and equating become powerful tools rather than opaque statistical rituals. The benefit is practical: fairer form comparisons, more credible trend lines, and growth reporting that respects how learning actually develops. If you work with assessment data, use this hub as your starting point, then review your program’s technical manual and ask whether the claimed score comparisons are truly justified.
Frequently Asked Questions
What is the difference between horizontal equating and vertical equating?
Horizontal equating and vertical equating both place scores onto a common scale, but they are designed for different comparison purposes. Horizontal equating is used when test forms are intended to measure the same construct at the same difficulty level for the same population. A common example is multiple forms of a Grade 5 mathematics assessment administered in the same year. In that case, the goal is score interchangeability: a student who takes Form A should receive a score that means the same thing as a student who takes Form B, even if the forms are not identical.
Vertical equating, by contrast, is used when assessments are built for different grade levels or developmental stages and are intended to reflect growth over time. For example, a vertically scaled reading program may connect scores from Grade 3, Grade 4, and Grade 5 onto one reporting scale. The purpose is not to claim that the tests are identical, but to support interpretation of progress across levels. Because the tests differ in content targets, difficulty, and expected performance, vertical equating carries stronger assumptions and requires careful design to ensure the scale reflects meaningful developmental change.
In practical terms, horizontal equating answers the question, “Are scores from these parallel forms comparable right now?” Vertical equating answers, “Can scores from these different grade-level tests be interpreted along a developmental continuum?” That distinction matters because using a horizontal method for a vertical purpose, or vice versa, can lead to inaccurate score reports, flawed growth conclusions, and confusion among educators and families.
How is equating different from scaling in psychometrics?
Scaling and equating are related, but they are not the same process. Scaling refers to the statistical construction of a score metric. It is the step where raw item responses are translated into a latent scale, often through classical test theory or item response theory methods. The scale defines the numerical framework on which scores will be reported, such as a score range, unit size, and relationship between performance and reported values.
Equating happens after or alongside scaling and addresses comparability. It is the process of adjusting scores from different test forms, administrations, or levels so they can be used interchangeably within a defined context. If scaling builds the ruler, equating makes sure that different versions of the ruler line up correctly. Two forms may each be scaled well on their own, but without equating, a score of 250 on one form may not represent the same achievement as a 250 on another form.
This distinction is especially important in operational assessment programs. In K–12 reporting, confusion between scaling and equating often leads stakeholders to assume that any common-looking score scale automatically guarantees comparability. It does not. A shared scale label or score range is not enough. Equating requires evidence that the forms or levels are linked appropriately, that the statistical assumptions are met, and that the intended score interpretations are supported. When programs blur the line between scaling and equating, reporting errors can follow quickly, particularly when technical documentation is abbreviated or when score users are not trained on what the reported scale actually means.
When should horizontal equating be used instead of vertical equating?
Horizontal equating should be used when the tests being linked are alternate forms designed for the same examinee population and the same intended interpretation. The forms should measure the same content domain, to the same specifications, at roughly the same level of difficulty, and under similar testing conditions. This is the standard situation for spring-to-spring alternate forms in a single grade, multiple secure versions of an end-of-course exam, or different forms used in large-scale administration windows to maintain test security.
The key requirement is that the scores are meant to be interchangeable. If a student takes Form X and another student takes Form Y, and both forms target Grade 6 science under the same blueprint, horizontal equating can adjust for small form differences so the resulting scores mean the same thing. In these settings, the emphasis is fairness across forms rather than measurement of long-term developmental growth.
Vertical equating should not be substituted just because assessments happen in adjacent grades. If the forms differ by grade-level expectations, cognitive demand, or content structure, then the interpretation shifts from comparability within a level to progress across levels. That is a fundamentally different psychometric problem. Choosing the wrong approach can distort score meaning. For example, treating grade-level tests as if they were merely alternate forms may understate genuine developmental differences, while forcing a vertical interpretation onto weakly connected forms can create misleading growth trends. The right choice depends on the test design, the population, and the intended score use—not simply on whether the tests share subject matter or administration timing.
What are the biggest challenges and assumptions in vertical equating?
Vertical equating is appealing because it supports growth interpretation across grades, but it is also one of the more complex linking tasks in educational measurement. The first challenge is construct continuity. For a vertical scale to be meaningful, the assessments at different levels must measure a construct that is stable enough across grades to justify a single developmental continuum. That is not always easy. Reading in early elementary school may emphasize decoding and foundational skills, while later grades rely more heavily on comprehension, vocabulary, and analysis. If the construct changes too much across levels, a single vertical scale may oversimplify what students actually know and can do.
A second challenge is test overlap. Vertical equating usually depends on common items, overlapping content, or a well-designed linking plan to connect adjacent grades. If the tests are too far apart in difficulty or too different in content emphasis, the statistical link can become weak. That can make growth estimates unstable or overly model-dependent. In practice, good vertical designs often include overlapping item pools, adjacent-grade linking, and careful blueprint alignment to support a defensible continuum.
Another major issue is interpretation. Even when the technical linking is strong, users may overread the scale. A gain of 20 points from Grade 3 to Grade 4 does not automatically mean the same thing as a gain of 20 points from Grade 7 to Grade 8 unless the program has clear evidence that scale units behave consistently across the range. Vertical scales are powerful tools, but they do not eliminate the need for grade-specific expectations, proficiency standards, and caution in communicating results.
Finally, vertical equating relies on assumptions about model fit, population comparability, and the relationship between the linked tests. If those assumptions are not monitored, the resulting scale can appear precise while masking important validity concerns. That is why responsible programs pair vertical equating with strong content design, empirical checks, and careful guidance for score users. The technical success of the equating process matters, but so does the clarity of what the reported scores are allowed to mean.
Why does it matter to educators and reporting teams whether scores were horizontally or vertically equated?
It matters because the type of equating defines the claims a score report can support. If scores were horizontally equated, educators can compare results across forms within the same grade or administration context with confidence that form difficulty differences were addressed. That supports fair decisions about student performance, accountability, placement, and trend analysis within the intended population. If those same scores are interpreted as evidence of year-to-year growth across grades without a valid vertical framework, the conclusions may be inaccurate.
When scores were vertically equated, reporting teams can describe performance across grade levels on a shared developmental scale, but they still must communicate limits carefully. A common scale does not mean every point gain has identical instructional meaning, and it does not remove the need for grade-level context. Teachers, parents, and administrators often want a simple answer about whether a student “grew enough,” but the reporting language has to reflect what the equating design actually supports.
This distinction becomes operationally important in score reporting systems, technical manuals, and stakeholder communications. Labels such as “scaled score,” “equated score,” and “growth score” are often used loosely, and that is where preventable mistakes happen. A district may compare scores across years assuming continuity that was never established, or a vendor may report a shared score range that looks vertically linked even though only horizontal form equating was performed. In real assessment programs, these misunderstandings can affect accountability interpretations, student placement decisions, and public trust in results.
For educators and reporting teams, the takeaway is simple: before interpreting comparability, ask what kind of equating was done, what populations and forms were linked, and what uses the program explicitly supports. That one step can prevent serious reporting errors and lead to more accurate, responsible decisions about student performance.
