How to Find Dead Questions in a Scorecard and Cut Them
How to find dead questions in a scorecard: when one option holds eighty per cent of answers, the question separates nobody. Here is how to read and cut them.
How to find dead questions in a scorecard: read the answers question by question. If one option holds eighty per cent or more of the answers, the question separates nobody and spends attention for no information. Cut it or rewrite it.
How to find dead questions in a scorecard
You wrote twenty-odd questions. You felt good about every one. Some of them are doing nothing.
A dead question looks fine on the page. It's well worded, it's relevant, it sits neatly in its dimension. And almost everyone answers it the same way. It tells you nothing about anybody. Worse, it costs each person a few seconds of attention and gives them nothing back.
You can't see this from your desk. You can only see it in real answers. So finding dead questions in a scorecard is a job of reading, not writing.
The eighty per cent rule
The test is simple. Look at how the answers to one question are spread across its options. If one option holds eighty per cent or more, the question is dead.
Why that line? Because a question earns its place by splitting the group. If nearly everyone picks the same answer, the question can't tell one person from another. It adds the same amount to almost every score. Remove it and the ranking barely moves.
A split distribution is a question doing its job. Two or three options each holding a real share means people differ here, and your scoring is picking that up.
Don't wait for perfection. The guide suggests doing this read after the first fifty or so completions, then monthly. Before that, you're staring at noise.
Free guide · 28 pagesBuild a scorecard that tells the truth.The Scorecard Playbook: an assessment that separates people and gives you a reason to talk.Get the free guide →Nothing will compute this for you
Here's what might surprise you. You'd assume your tool flags dead questions. It doesn't.
The guide checked the whole analytics layer of a mature assessment product. No item-total correlation. No reliability coefficient. No discrimination index. Not one. Every platform in the category shows you distributions and leaves you to interpret them.
That's good news for you, oddly. It means there's no magic dashboard you're missing. The method is eyeballing one chart per question and asking whether it's lopsided. You can do that in an afternoon.
It also means the skill is knowing what you're looking at. Someone who builds one scorecard guesses. Someone who builds a good one reads.
Four ways dead questions in a scorecard appear
Lopsided answers are the headline sign. But questions die for different reasons, and the fix depends on the cause.
| What you see | Likely cause | The fix |
|---|---|---|
| One option holds eighty per cent or more | The question is too easy, too obvious, or everyone in your audience shares the trait. | Cut it, or rewrite it to ask about something harder. Most dead questions in a scorecard start here. |
| Everyone picks the flattering answer | The question signals its own right answer. | Rewrite it to ask about behaviour and what exists, not opinions or intentions. |
| A slider sitting near the middle | A slider pre-answers itself at the midpoint, so untouched sliders look like deliberate middle answers. | Use a single-choice question, or accept that the data is unreliable. |
| One question seems to swing a whole dimension | A scale mismatch, such as a yes/no scoring a hundred beside neighbours that top out at four. | Rewrite it on the same scale as everything else. |
The slider one deserves a note. A slider always has a position. So it counts as answered and scores half marks before anyone touches it. In your data it's indistinguishable from a person who thought about it and chose the middle. If you used sliders, treat their results with suspicion.
The scale one is quieter. Drop a yes/no question into a scorecard that runs from zero to four and it can silently own almost the whole dimension, because yes scores a hundred next to neighbours topping out at four. Nothing errors. You'd only spot it by noticing that one dimension moves whenever one question does. The fix lives in the design, which is covered in how to weight scorecard questions.
Why cutting is the best edit you'll make
Rewriting feels productive. Cutting is more valuable.
Cutting a dead question does two things at once. It shortens the assessment, which raises completion. And it removes noise, which sharpens every result. Few edits pay twice.
For a solo operator, there's a third benefit: less to maintain. Every question is something you wrote, translated and defended. Fewer questions means fewer things to break.
There's a hidden consideration too. Length decides which kind of scorecard you've got. The guide notes the sizes cluster hard: twenty to twenty-nine questions for something you give away, and a hundred and twenty-five to a hundred and forty-five for something you charge for. If you're building the free one, every question you cut keeps you inside the size that works.
The question to ask of each survivor is: could this plausibly split the group? If the honest answer is "it's there for completeness", that's your sign.
Free assessmentHow trapped are you, really?Take the Corporate Suffocation Index, the free solopreneur readiness assessment.Take the assessment →Cut carefully, because nothing recomputes
Now the part that bites. Scores are written once, when someone completes the scorecard. There is no rescore. Whatever you change, past respondents keep the result they were given.
So cutting a question after launch doesn't rewrite history. The people who answered it before still carry its contribution. Anyone who takes the shorter version afterwards is measured differently. Your numbers now come from two slightly different instruments, and nothing records which one a person took.
That leads to a few habits:
- Tune before you drive traffic. The cheapest time to cut is before anyone serious has taken it.
- Batch your changes. Make several cuts in one go instead of one a week.
- Date every change in your own notes. The system won't do it for you.
- Never read a trend across an edit. A shift after you cut a question may be the cut, not the audience.
Question wording you can edit safely. Option wording you can't. Changing an option after launch is a new version of the instrument, not an edit, because it can break translations and anyone midway through. If you want to rewrite a dead question rather than cut it, change the question text and leave its options alone, or treat it as a new version.
If cutting leaves a dimension thin, the problem might be bigger than one question. How to calibrate a scorecard covers reading the spread of a whole dimension and what to do when it will never change anyone's advice.
An afternoon of reading
- Wait for about fifty completions. Less than that is noise.
- Open the answers question by question. Note any with one option at eighty per cent or more.
- Sort each dead question by cause. Too easy, signals the answer, slider, scale mismatch.
- Cut what you can. Rewrite only what you must.
- Write the date and the change in your notes. Then leave it alone for a month.
Before you start cutting, check that the questions you keep are feeding something. A question with a good split and no advice behind it is a smaller version of the same problem, which how to write a scorecard action plan deals with.
Free guide · 28 pagesBuild a scorecard that tells the truth.The Scorecard Playbook: an assessment that separates people and gives you a reason to talk.Get the free guide →What a living question looks like
It helps to know the target. A healthy question has answers spread across its options. Not evenly, necessarily. But enough people pick each of two or three options that the choice means something about them. A good question asks about behaviour and what exists, not opinions or intentions. "Do you have a written process for onboarding a new client?" splits a group, because some do and some don't. "Do you care about client experience?" does not, because everybody says yes. The second one is the commonest source of dead questions: it asks for an attitude when it should ask for a fact.
Check also that no question signals its own right answer. If a person can tell which option you want, they'll pick it, and the question dies by politeness. The cure is to describe what people actually do at each level, in plain terms, so that every option reads like a real situation rather than a good or bad grade.
When you're stuck, ask what you'd want to know about a stranger before you spoke to them. Questions you'd genuinely use to sort one conversation from another are alive. Questions you only added to look thorough are the ones that turn into dead questions in a scorecard.
Free assessmentHow trapped are you, really?Take the Corporate Suffocation Index, the free solopreneur readiness assessment.Take the assessment →Frequently asked questions
How do I know a question is dead?
Look at how answers spread across its options. If one option holds eighty per cent or more, the question separates nobody. A healthy question has answers split across more than one option.
Does my scorecard tool flag dead questions?
Probably not. The guide checked a mature assessment product's analytics and found no item-total correlation, no reliability coefficient and no discrimination index. You read the distributions yourself.
When should I check for dead questions in a scorecard?
After the first fifty or so completions, then monthly. Earlier than that, the numbers are too thin to trust. Do it before you push hard on traffic, because changes later muddy your history.
Will cutting a question change old results?
No. Scores are written once, at completion, and nothing recomputes. People who already finished keep their result. Date the change in your notes so you never read a trend across it.
The full method, with the checklists, is in The Scorecard Playbook, a free 28-page playbook.
How trapped are you, really?
Take the Corporate Suffocation Index, the free solopreneur readiness assessment.