Why a spelling mistake should not cost a student marks

Written questions are the ones worth asking. They are also the ones most teachers quietly stop asking, because marking thirty of them by hand costs an evening and auto-marking them costs your credibility the first time a student who clearly knew the answer gets a zero for a missing letter.

That trade-off is not actually forced. It exists because most quiz software treats a written answer as a string to compare, and a string comparison has exactly two outcomes. Real answers do not.

Why exact matching fails on written answers

Consider one question: "Who is credited with the theorem relating the sides of a right-angled triangle?" Here are four answers a class will actually give:

• Pythagoras — the key.
• Pythagorus — knows it, misspelled it.
• pythagoras — knows it, typed on a phone with autocapitalise off.
• Pythagoras of Samos — knows it better than the key does.

Exact matching scores the first one and fails the other three. The student who wrote more than you asked for is punished for it. That is not strictness — it is measuring typing rather than knowledge, and it produces a mark you cannot defend to a parent.

The second-order cost is worse. Once you know the auto-marking is unreliable, you have to read every answer to find its mistakes, including the ones it got right. You now do more work than if you had marked the whole thing by hand.

The principle: decide what "correct" means per question

There is no single right strictness. On a spelling test, recieve is wrong and must be wrong. On a history quiz, a misspelled name is a spelling slip in a question that was not about spelling. The same tolerance cannot serve both, so the choice has to sit on the question, not on the app.

In practice that gives five useful settings.

1. Exact match, normalised

Compare letter for letter, but first fold away case and punctuation. New York, USA and new york usa become the same answer, which they obviously are. This is the sensible default for single-token factual answers.

2. Case-sensitive exact

Keep capitals and punctuation and require both. This is the spelling-test rule, and the rule for anything where HCl and hcl genuinely differ — chemical formulae, code identifiers, case-marked notation.

Use it deliberately and rarely. It is the setting most likely to produce an unfair zero, which is fine when the capital letter is the assessment and indefensible when it isn't.

3. Typo tolerance with a threshold you control

Score how close the answer is to the key rather than whether it equals it, and accept anything above a line you set. Pythagorus against Pythagoras is one character out of ten — obviously the same answer. Plato is not.

The threshold matters. Set it too tight and you are back to exact matching; too loose and genuinely different words start matching each other. A useful instinct: loosen it for long proper nouns, tighten it for short words, where a single character is a much larger share of the whole.

4. Keyword coverage, for explanations

Some questions want reasoning, and reasoning cannot be compared to a key at all. What you can do is list the ideas that have to be present and set how many are enough.

Ask "Explain what plants need for photosynthesis" and list chlorophyll, sunlight, carbon dioxide, water. An answer that hits three of four is clearly not a blank page, and treating it as one is absurd. Set the bar at "most" and it passes; set it at "all" when the omission genuinely matters.

This is the setting that makes "explain why" questions viable at scale, and it is the one most teachers do not realise exists.

5. Numeric equivalence

Numbers should be compared as numbers. 7, seven and 7.0 are one answer, and so are 1500 and one thousand five hundred. Any tool comparing those as text will mark two of the three wrong, which is purely an artefact of the implementation and has nothing to do with the student.

The decision that matters more than all five

This single decision changes the economics completely. A quiz of settled answers costs you nothing to mark and you can trust the totals. A quiz of arguments costs you an evening, and no threshold will change that — but it is an evening you chose to spend rather than one you spend undoing a computer's marking.

In a thirty-student class only a handful of answers are genuinely unusual — the odd phrasing, the partly-right reasoning, the answer that is better than your key. Those are the ones worth opening after the marks are in, whoever or whatever awarded them. They are the answers that deserved a teacher in the first place.

Practical advice for writing the questions

• List every acceptable answer, not just your favourite. If H2O and water are both fine, say so up front rather than discovering it on script eleven.
• Ask for one thing. "Name the process and explain why it matters" is two questions wearing a coat, and no single rule can grade it.
• Say how much you want. "In one word" removes an entire category of near-miss.
• Reuse your review decisions. When you accept an answer you did not anticipate, add it to the accepted list for next year.
• Prefer several loose rules to one tight one. A question can carry typo tolerance and numeric equivalence and accept whichever fires.

When you should mark by hand instead

Typo tolerant grading is a tool for questions with a settled answer. It is the wrong instrument for a whole category of written work, and reaching for it there is how a quiz quietly stops measuring the thing you cared about.

Mark by hand when:

• The answer is an argument. Anything where two students can be equally right in different words is a judgement, not a comparison. No threshold saves you here.
• Partial credit is the point. If the interesting question is how much of the reasoning a student has, you want to read the reasoning.
• Spelling is the assessment. On a vocabulary or spelling test, a misspelling is the data. Turn tolerance off and say so on the question.
• The stakes are high and the cohort is small. For a handful of scripts that carry real weight, reading them costs less than defending an automated mark.
• The question is new. The first time you use a question you do not yet know what the class will write. Read that round, then set the rules from what you saw.

The practical arrangement is to sort it at the quiz level: let the rules mark the papers whose answers are settled, and put the ones that turn on judgement into a quiz you mark yourself. Quizoma works the same way — a quiz is either graded automatically, using whichever of the matching rules each question carries, or switched to manual marking, in which case nothing in it is auto-graded at all.

What you get back

Done properly, the marking pile for a twenty-question class test becomes a handful of answers rather than six hundred, and the marks you hand back are ones you can explain. More usefully, written questions stop being the expensive option — which means you can go back to asking them, even in a mostly multiple-choice paper like the one in how to create an online MCQ test for your class.

Quizoma implements all five rules per question, with typo tolerance as the one a new question starts on, and lets you switch a whole quiz to manual marking when you would rather read everything. There is a full breakdown, with worked examples, on how to auto-grade short answers.