Parallel test forms are not shuffled questions

Parallel test forms use different items built to one specification, with scores equated so a result means the same on each form. Why shuffling is not enough.

Parallel test forms are versions of one exam built from different items to the same specification, with their scores equated so that a result means the same whichever form a candidate sat. They are not shuffled questions. Shuffling changes the order and makes two papers look different; it does not make two different sets of items equivalent. On NUADU, item-order and option-order randomisation is the default for lower stakes, and parallel forms drawn from item banking with calibration statistics are the high-stakes option — because comparability is a delivery decision with an appeals consequence.

What are parallel test forms?

Parallel test forms are two or more versions of an exam that use different items but are built to one specification — the same content coverage, the same mix of item types, a comparable spread of difficulty — and whose scores are equated before anyone compares them. Exam boards and certification bodies often call these versions variants.

The specification is the part that can be written down in advance. Often called a blueprint, it fixes what every form must cover: the topics and their weight, the item types, the number of items and the time allowed. Two forms that answer to different blueprints are two exams, whatever they are called.

The rest has to be observed. In classical test theory an item's difficulty is the proportion of candidates who answer it correctly, its discrimination is how well it separates stronger candidates from weaker ones, and a form's reliability is how consistently it would rank the same candidates again. None of that can be read off the page; it comes from candidates answering the items, which is why item analysis sits underneath any claim that two forms are comparable.

The practical consequence is blunt. A parallel form cannot be assembled on the morning of the exam from items nobody has seen perform. The evidence has to exist before the form does.

How do you create multiple versions of a test?

There are two ways to create multiple versions of a test, and they make different claims. Shuffling the same items into a different order is a security measure. Building versions from different items is a measurement claim, and a measurement claim needs evidence.

Randomising the order of items, and of the options within them, makes copying from the next seat harder and makes each paper look different from the one beside it. The content is identical, so the difficulty is essentially the same, and there is little to prove. For a weekly test, shuffling exam questions is usually enough.

Writing a second set of questions — by hand, by changing the numbers in a calculation item, or by drawing a different selection from a bank — produces different items, and different items carry different difficulty until someone shows otherwise. Two candidates who sat two such versions sat two exams, unless the versions were built to one specification and their scores equated.

On NUADU the choice is a per-assessment setting under exam scheduling. Item-order and option-order randomisation is on by default, and for a classroom test or a progress check it is often all the variation a programme needs. Where the stakes justify it, the exam officer switches to parallel forms. Which of the two an assessment uses is one of the four decisions of online exam administration, settled in configuration rather than at the door.

Variants from item banking with calibration statistics — not just shuffled questions.

How does NUADU draw a parallel form?

NUADU draws parallel forms from item banking with calibration statistics, with equating logic to keep difficulty comparable across forms. The calibration comes first: items earn their place in the pool before any form is drawn from it.

A certification programme might run item development like this. Subject-matter experts author the items. Each item goes through psychometric review, calibration statistics and equating before it enters the live pool, and the live pool is the bank parallel forms are later drawn from — so item quality is evidenced, not assumed. The item bank keeps its own record — every item versioned, with a named approver and its change history — ready for QA and external review.

In delivery, the options line up like this:

  • Lower stakes: item-order and option-order randomisation, on by default.
  • Higher stakes: parallel forms from calibrated items, equated so difficulty stays comparable.
  • Either way: configurable per assessment, a different form per cohort where appropriate, every operational action logged.

The per-cohort option is what makes parallel forms practical across sites. A programme can give each campus its own form, cohort by cohort, decided once in configuration, so the question of who sat what is answered by the setup rather than by memory.

NUADU does not claim that two parallel forms are statistically identical. No equating does that. The claim is narrower and more useful: forms are drawn from calibrated items, equating logic keeps difficulty comparable across forms, and the item evidence is kept ready for QA and external review. Comparable, evidenced and logged is a standard a high-stakes programme can defend. Identical is a word it should never have to use.

What will an appeal ask about the forms?

An appeal will ask whether the paper a candidate sat was the same exam as everyone else's, and with parallel forms that is a question about evidence, not about order. The Certification path lists what has to hold when a candidate appeals; one link in that chain is "that variants are properly equated".

Picture the case. The same certification exam runs in several sessions. A candidate from a later session says her paper was harder than the one a colleague sat earlier. If both sessions drew the same items in a different order, the answer is easy. If they drew different items, the panel needs the calibration statistics behind each form, the equating that joined them, and the record of which cohort was given which form. A programme that can show those three has an answer. A programme that shuffled questions and called the result parallel forms has only a score.

Variants are one link in that chain, not the whole of it. Identity verification, integrity controls, marking and the candidate's route to contest a result each carry their own weight, and the exam appeals process follows them link by link. Decide the variants carelessly, though, and the chain breaks before marking has started.

Frequently asked questions

What is the difference between parallel forms and alternate forms?

In classical test theory, parallel forms are the strict case: every candidate would have the same expected score on each form, with the same error variance. Alternate forms — sometimes called equivalent forms — are built to the same specification without assuming that strict equality, which is why their scores are equated before they are compared. In practice no two forms meet the strict standard exactly, so the forms a programme calls parallel are equated too, and the two terms are often swapped.

What does test equating mean?

Test equating is the statistical adjustment that puts scores from different forms of the same exam on one scale, so that a given score means the same whichever form was sat. It relies on calibration data from real responses. It corrects small differences in difficulty between forms built to one specification; it cannot turn two different exams into one.

What is an example of a parallel form test?

The Assign page sketches one, illustratively: a midterm run across several campuses, where each campus gets a parallel form and results aggregate to a single programme view. The campuses sit different items, but every form comes from the same calibrated bank and is equated, so difficulty stays comparable between sites.

→ See how certification programmes run on NUADU