---
title: Writing Likert items that survive a pretest
description: Five rules for agreement scales: one dimension per item, a true midpoint, a labelled scale, no negation, and a reversal test before you field.
---

# Writing Likert items that survive a pretest

Published 2026-09-05

Agreement scales are the default in survey research and the place where most item defects are born.
The rules below are not stylistic. Each one corresponds to a failure a pretest will otherwise catch
after the field, at the cost of the field.

## One dimension per item

An item that asks two things cannot be answered when the two halves disagree, and the respondent
answers something else: the first half, the last half, or the average of the two, depending on how
the sentence is parsed. "The price is fair and the quality is high" is two items. Split it, or you
will not know which one moved the answer.

## A true midpoint

A five-point agreement scale has a neutral point only if the middle label means neither agreement
nor disagreement. "Neither agree nor disagree" is a position, not a remainder. Use it, and check that
the two sides are symmetric: two agreement points and two disagreement points, with equal
intensity, otherwise the midpoint is not in the middle.

## Every point labelled

Numeric-only scales invite interpretation. A respondent who sees 1 to 7 builds their own meaning for
4, and the meaning is not the same across respondents. Label each point with a phrase a respondent
could defend in a sentence.

## No negation

Negated items cost a working-memory step, and the error they introduce is not random: it concentrates
among respondents who read quickly, which is a large and non-random share of any sample. If you need
the opposite direction, write the opposite statement, not the same statement with "not".

## The reversal test

Before you field, run the item and its reverse. If the two answers do not move in opposite
directions, the item is not measuring a single dimension, or the scale is not being read the way you
intended. This is the cheapest test available, and it is the one most often skipped.

## A note on balance

Balance is about the scale, not about the topic. A scale can be balanced and still be biased by the
question stem, and an item can be neutral in tone and still collapse. The reversal test catches the
second case; only a careful stem catches the first.

None of this requires a population. It requires the item, the scale, and one run against respondents
who will answer it as written. That is what a pretest is for, and it is why the most useful output of
a pretest is a list of items to rewrite, not a number to report.