Likert scale

Item versus scale

The distinction is pedantic and it matters. A Likert item is one statement with an ordered agreement response — "The information was easy to understand: strongly disagree … strongly agree". A Likert scale is several such items about the same underlying attitude, whose responses are combined into one score.

Almost everything called "a Likert scale" in day-to-day survey work is a single Likert item. That is fine, but it explains an argument you will eventually have: whether you can take a mean. Individual item responses are ordinal — the gap between "disagree" and "neither" is not known to be the same as the gap between "neither" and "agree" — so a mean of one item is on shaky ground. A summed multi-item scale behaves much more like interval data, which is what Rensis Likert was proposing in 1932 in the first place.

Designing the scale

• Be symmetric. The same number of positive and negative points, with mirrored wording. "Excellent / very good / good / fair / poor" is not symmetric — four of those five are positive, and it will flatter you.
• Decide about the midpoint deliberately. An odd number of points offers a neutral option, which is honest for people who genuinely have no view. An even number forces a lean. Neither is wrong; what is wrong is switching between them mid-survey.
• Label every point, not just the ends. Numbers without words are interpreted differently by different people, and translated differently by everyone.
• Keep "don't know" out of the scale. If you need it, it is a separate option outside the ordered points, or it will be treated as a midpoint in analysis.

A worked example

A patient experience survey wants a "confidence in the service" measure. One item — "I felt confident in the care I received" — gives you a headline. Four items, all on the same five-point agreement scale, give you a scale:

• I felt confident in the care I received.
• Staff explained things in a way I understood.
• I knew who to contact if something went wrong.
• I would feel confident using the service again.

Score each 1–5, total them, and you have a 4–20 scale that is far more stable than any one item. This is naturally a matrix question: four rows, one shared scale.

Common mistakes

• Double-barrelled statements. "Staff were friendly and efficient" cannot be agreed with by someone who met a rude, efficient person. One idea per item.
• Negative wording mixed in carelessly. Reversing some items guards against people ticking straight down the page, but you must remember to reverse the scores. Un-reversed items are one of the most common quiet errors in survey analysis.
• Reporting "top two box" without saying so. Collapsing "agree" and "strongly agree" into one number is a legitimate and common choice. Not disclosing that you did it is not.
• Comparing across differently worded scales. Last year's seven-point scale and this year's five-point scale do not produce a trend.

In NumoForms

There is no question type called "Likert", because a Likert item is a shape rather than a mechanism. Build one as a Linear scale (1–n with end labels), as a Multiple choice question with the five labels written out in full, or — for several items on the same scale — as a Matrix. If you want a total across items, per-choice scores and per-question weights are part of scoring and bands. For the difference between this and a 0–10 recommendation question, see NPS.