The Likert scale is the workhorse of survey research, appearing in everything from customer satisfaction surveys to political polls to psychological assessments. Its popularity is deserved, but the ease of using it masks subtleties that determine whether the resulting data actually measures what you intend to measure. Getting these details right separates surveys that yield insights from those that yield noise. The classic Likert format presents a statement and asks respondents to rate their agreement from "strongly disagree" to "strongly agree," typically in five or seven points. The key assumption is that responses represent an underlying continuous attitude, and the number of response options represents precision along that continuum. When this assumption holds, you can treat responses as numbers and calculate means. When it doesn't, you can't. Five-point scales suit quick surveys, customer feedback, and situations where respondent attention is limited. They force categorization that seven-point scales allow respondents to avoid. Seven-point scales add two middle positions that people with moderate opinions can occupy naturally, rather than being forced toward either side of neutral. Nine-point scales are sometimes used in academic research for maximum precision. Labeling conventions affect how people respond. Some scales label every point—"strongly disagree," "disagree," "neither agree
Introduction
nor disagree," "agree," "strongly agree." Others label only anchors—the endpoints—with intermediate points left blank or labeled as numbers. Fully labeled scales are more intuitive but can introduce label effects where specific words influence responses differently. Partially labeled scales reduce this bias but may confuse respondents. Balanced scales have equal numbers of positive and negative options, with a neutral midpoint. Unbalanced scales—perhaps three positive options and two negative—introduce acquiescence bias, the tendency for respondents to agree more than they would with a balanced scale. This matters when comparing results across surveys with different scale balances. Psychometric validation determines whether your Likert scale actually measures the construct you think it does. Factor analysis reveals whether items that should cluster together actually do. Cronbach's alpha measures internal consistency—whether the scale's items are measuring the same underlying construct. A scale that fails these tests might be measuring several unrelated things bundled together under one label. Analysis choices matter as much as scale design. Treating Likert data as interval for computing means and standard deviations is standard and usually appropriate for five or more points. For fewer points, non-parametric tests or ordinal logistic regression provide more appropriate analysis. The choice depends on how you'll use
Key Concepts
results, not on what looks simplest.