Analyse Your Likert Items The Right Way

How it works

The response scale is detected from the data itself: numeric codes are read in ascending order, and text labels are ordered by matching them to a known survey wording family (agreement, satisfaction, frequency, quality, likelihood, importance) rather than alphabetically. Every item is then summarised on that shared scale — the full distribution, the top-box share (the highest option), the top-2-box share, the bottom-box share, the modal category, and the median category, defined as the lowest option whose cumulative share of answers reaches 50 percent so that it stays well defined under ties and even sample sizes. Items are ranked by top-2-box and drawn as a diverging stacked bar with shares at or below the midpoint to the left of zero. When a grouping column is mapped, each item's ranked answers are compared with Mann-Whitney (two groups) or Kruskal-Wallis (three or more), with Holm correction across items and a rank-biserial or epsilon-squared effect size. The mean of the codes is reported alongside, always with the equal-spacing assumption stated.

Use it for any set of survey questions answered on a rating scale — employee engagement, customer satisfaction, product feedback, course evaluations, agreement batteries — especially when you want to rank items or report them to an audience that will read an average as a measurement.

Not for a single continuous measurement (use the group comparison or regression tools), not for unordered categories such as "which channel did you use" (use the categorical association tool), and not for judging whether several items form one reliable scale (use the reliability tool for Cronbach's alpha).

What data do you need?

One row per respondent, one column per rating question, optionally a grouping column. Answers may be numeric codes or text labels:

Minimum 10 rows · Best with 100-20,000 respondents and 3-20 scale items

What's in the report?

Standard-library analysis: your Likert-scale survey items analysed as the ordinal data they are, not as if they were numbers. Map the columns holding your scale answers and get, for every item, the full response distribution, the top-box and top-2-box shares, the bottom-box share, the median response category and the modal category, plus a diverging stacked bar across all items so you can rank them at a glance. The scale itself is detected and stated from the data — numeric codes (1-5, 1-7) or text labels ("Strongly disagree" through "Strongly agree") ordered by their wording rather than alphabetically. An optional grouping column compares the groups with a rank-based test (Mann-Whitney or Kruskal-Wallis), Holm-corrected across items, never a t-test on the raw codes.