A matching question tests a whole set of relationships in the space of one question: eight terms against eight definitions, six causes against six effects, a column of symbols against a column of names. Done well, it covers more ground per minute than any other shape. Done badly, it is a puzzle that a student with no knowledge can half solve by elimination.
The difference is almost entirely in the lists. This guide covers how to build them, the arithmetic of the free last pair, how many options a set should carry, and the extended matching format that medical licensing exams adopted. EdProcto builds the shape as matching pairs, marks each pair by rule, and shuffles the right-hand side for every student.
When is a matching question the right shape?
A matching question is the right shape when you have a family of items of the same kind, each with one partner of another kind. Terms and definitions, dates and events, formulas and their names, authors and works, organs and functions, vocabulary in two languages.
It is the wrong shape when the items are a mixed bag. The standard taxonomy of item-writing rules lists matching among the formats it recommends for classroom tests, and one of its rules for options applies with double force here: keep the choices homogeneous in content and grammatical structure (Haladyna et al., 2002). In a matching set, every answer is a candidate for every prompt, so one answer of a different kind is a gift.
A test for homogeneity: read any prompt against any answer. If the pairing is wrong but not absurd, the set is sound. If half the pairings are nonsense ("1066" against "the capital of France"), the student is sorting categories, not recalling history.
How do you write the two lists?
Write the longer text on the left and the short answers on the right, keep both lists on one screen, and make each side one kind of thing. The student reads a prompt once and scans the short list many times, so the short list should be the one that is scanned.
- Give the set a clear instruction. "Match each process to the stage of the water cycle where it happens" says what the relationship is. "Match the following" does not.
- One kind of item per side. All processes on the left, all stages on the right.
- Five to eight pairs. Fewer than five and elimination does most of the work. More than eight and the question tests patience and scrolling.
- Order the answer list logically: alphabetically, by date or by size. A logical order hides nothing and saves search time. The same taxonomy asks for options in logical or numerical order.
- Make near neighbours. The pairs that teach you something are the ones students confuse: mitosis and meiosis, weather and climate, affect and effect.
- Say whether an answer can be used more than once. If it can, the last pair stops being free.
A set, before and after
Before: left column "Paris, 1945, H2O, Shakespeare"; right column "water, Hamlet, France, end of the Second World War". Each prompt has only one answer of the right kind. Nothing is tested.
After: left column "evaporation, condensation, precipitation, infiltration, transpiration"; right column, in alphabetical order, "liquid water soaking into soil; water falling from clouds; water vapour leaving plant leaves; water vapour turning to liquid; liquid water turning to vapour". Every answer is a plausible partner for every prompt, and a student who confuses evaporation with transpiration shows it.
Why is the last pair free, and how do you stop it?
When both lists are the same length and each answer is used once, the final pair needs no knowledge at all, and the one before it is a coin toss. A student who knows three of five pairs for certain has two prompts and two answers left: a fifty-fifty guess that is worth two marks.
The arithmetic of pure guessing is less generous than students hope, though. Pairing two equal lists at random gets one pair right on average, however long the lists are, and gets all five pairs of a five-pair set right once in 120 tries. The real leak is not the blind guesser. It is the student who knows most of the set and is handed the rest.
Three ways to close the leak:
| Method | What it does | Cost |
|---|---|---|
| More answers than prompts | No pair can be finished by elimination | You must write plausible extras |
| Answers that can be used more than once | The count of remaining answers tells nothing | The set needs prompts that share an answer |
| Credit for each pair | The free pair is worth one small mark, not the whole question | None |
The first is the classic advice and the best where the tool allows it. The third should be on in every case: all-or-nothing marking on an eight-pair set throws away most of what the student showed you.
How many options should a matching set have?
About eight at most. The clearest evidence comes from medical licensing, where the shape was studied at scale. Swanson et al. (2008) presented 96 questions on a United States licensing examination in two formats and with different numbers of options. Questions with more options were harder and took longer, and they separated candidates about as well as questions with fewer. The authors concluded that extended matching sets with eight options, and ordinary single-answer questions with four or five, made the most efficient use of testing time.
That fits what is known about wrong options in general. In one analysis of 514 teacher-written multiple-choice questions, only about half of the wrong options were chosen often enough to count as working (Tarrant et al., 2009). Writing plausible wrong answers is hard, and a long list of implausible ones adds reading and nothing else. For a classroom set, eight answers is a sensible ceiling.
What are extended matching questions?
Extended matching questions are sets of short cases that share one long list of options. Each case is a question on its own, and the candidate picks the best option from the list for each.
Case and Swanson (1993) described the format as it was used in United States medical licensing: several items in a set, one option list, and typically each item describing a patient with the same presenting complaint, for which the candidate chooses the most likely diagnosis. They presented it as a practical middle way between a free-response question and a five-option multiple-choice question, because a long list of real alternatives leaves much less room for recognising the answer.
It has been tested as a whole exam. Beullens et al. (2002) reported on a final-year medical test built from extended matching questions, with option lists of 7 to 26 answers. Final-year students, 251 of them, answered series of 100 questions in about four hours. After the questions that correlated negatively with the total were removed, reliability reached a Cronbach's alpha of 0.80. Students grew more familiar with the format as the test went on, and a questionnaire on how they saw the test indicated poor face validity, which is worth knowing before you introduce the format to a class.
Using the idea outside medicine
The structure travels well. A theme, an option list, and a run of short scenarios: "For each situation, choose the statistical test you would use", "For each sentence, choose the tense", "For each fault, choose the most likely cause". The scenarios are where the thinking is, so write those with care and reuse the option list.
How does EdProcto build and mark matching?
EdProcto is the best tool for matching when the set comes from your own lesson, because it marks every pair on its own and never puts the right pairing on the student's device.
Build the set in the matching pairs template: the left items, and the partner for each. Or paste a chapter into the quiz maker on the home page and choose matching, and Procto AI drafts the pairs from the terms your text defines, for you to edit before anyone sees them.
When a student opens the quiz, the right-hand side is shuffled with a seed the student does not have, so the page carries no key and two students side by side see different orders. Marking is by rule, at once. With partial credit switched on, each pair is its own small question, so a student who has six of eight is scored for six. That setting is the single most useful one for this shape.
Where you want many items sorted against a few answers, which is the surest cure for the free last pair, the sort into groups template drops items into named groups and marks each placement. One to many, where one item on the left has several partners, is among the templates too. For young children there is picture match, with large targets.
The results then show which pairs the class got wrong together. Two terms that half the class swapped are a confusion to teach, and the full analysis, on the Educator plan, names the question that separated the students who had it from those who did not.
The Free plan carries the core templates, marking by rule and results question by question, and Tutor opens all 111 shapes. The pricing page lists what each plan carries.
What are the common mistakes?
The common mistakes are all failures of the lists.
- A mixed set. Different kinds of answer in one column, so most pairings are absurd.
- Equal lists with all-or-nothing marking. The student who knows most is handed the rest, and the student who slips once loses everything.
- A set that runs over the screen. Scrolling between prompt and answer measures working memory.
- Clues in the grammar. A prompt ending in "an" with one answer starting with a vowel.
- Definitions copied from the textbook. They are matched by remembered wording, not by meaning.
- Pairs that are a matter of opinion. If two answers are defensible for one prompt, it is not a matching question.
If what you want to test is an order and not a pairing, use a sequence shape instead. If there is one relationship and several tempting wrong answers, a multiple-choice question will tell you more, because each wrong option can name a different misconception.
Quick questions
How many pairs should a matching question have?
Five to eight. Below five, elimination does too much of the work. Above eight, the set stops fitting on one screen and the question starts to measure searching.
Should the two lists be the same length?
Ideally not: extra answers stop the last pair being free. Where the lists are equal, give credit for each pair so that the free pair is worth little. EdProcto's partial credit setting does this for matching pairs.
Is matching just a memory test?
It tests recognition of relationships, which is recall of a kind. It reaches further when the prompts are new examples and not the textbook's definitions: match each unseen sentence to its tense, each scenario to its principle.
What is the difference between matching and extended matching?
Ordinary matching pairs each prompt with one partner from a short list. Extended matching gives a long shared option list and a series of separate cases, each answered from that list, as Case and Swanson (1993) described for medical licensing.
Can a student guess a matching question?
Blind pairing gets about one pair right on average, whatever the length of the lists. The real gain is from elimination once most pairs are known, which is why extra answers and credit for each pair matter. EdProcto also shuffles the answer side for each student, so a neighbour's order is no help.
References
- Beullens, J., Van Damme, B., Jaspaert, H., & Janssen, P. J. (2002). Are extended-matching multiple-choice items appropriate for a final test in medical education? Medical Teacher, 24(4), 390–395. https://doi.org/10.1080/0142159021000000843
- Case, S. M., & Swanson, D. B. (1993). Extended-matching items: A practical alternative to free-response questions. Teaching and Learning in Medicine, 5(2), 107–115. https://doi.org/10.1080/10401339309539601
- Haladyna, T. M., Downing, S. M., & Rodriguez, M. C. (2002). A review of multiple-choice item-writing guidelines for classroom assessment. Applied Measurement in Education, 15(3), 309–333. https://doi.org/10.1207/S15324818AME1503_5
- Swanson, D. B., Holtzman, K. Z., & Allbee, K. (2008). Measurement characteristics of content-parallel single-best-answer and extended-matching questions in relation to number and source of options. Academic Medicine, 83(10 Suppl.), S21–S24. https://doi.org/10.1097/ACM.0b013e318183e5bb
- Tarrant, M., Ware, J., & Mohammed, A. M. (2009). An assessment of functioning and non-functioning distractors in multiple-choice questions: A descriptive analysis. BMC Medical Education, 9, Article 40. https://doi.org/10.1186/1472-6920-9-40