# Multiple Response Analysis: Respondents or Selections?

Multiple response analysis explained with a worked example: respondent percentages, selection percentages, explicit none answers, and missing data.

- Canonical URL: https://www.harmate.com/en/blog/multiple-response-analysis
- Author: Harmate Team
- Published: 2026-10-25
- Updated: 2026-10-25T22:48:16.322449+00:00
- Language: en

## Content

# Multiple Response Analysis: Respondents or Selections?

When a survey question allows several choices, separate **the percentage of respondents selecting an option** from **that option’s share of all selections**. The numerator can be identical while the denominator changes. The resulting percentages answer different questions.

This guide covers “check all that apply” survey items, not regression with several outcomes or repeated measurements. The practical output is a descriptive table whose calculations can be reconstructed, with explicit “none” answers and missing answers kept visible.

## Start with the records, not the chart

[Kent State University’s statistical tutorial on multiple response sets](https://libguides.library.kent.edu/SPSS/Multiple-Response-Sets) explains that a checkbox item comprises several variables, one for each option. It distinguishes percentages of cases from percentages of selections. The tutorial uses SPSS, but that distinction matters regardless of the tool used to prepare your table.

Check what the exported values mean. A zero can represent an unselected option under a documented coding scheme; another export may use a blank for the same state. An entirely empty row can also indicate that the question was not shown or that an answer is missing. Do not assign one interpretation to every empty cell without consulting the questionnaire and export settings.

Where appropriate, prepare a working table with one identifiable response per row and one column per option. Preserve the original file. If several choices arrive in one cell, verify their separation and mapping to labels rather than treating every comma as a new option. The [survey-data preparation guide](/en/blog/prepare-survey-responses-before-analysis-a-practical-method-and-example) covers these checks.

## A worked example: resources used after a workshop

**All data below are fictional.** A training provider asks, “Since the workshop, which resources have you consulted? Select all that apply.” Options are a practical guide, a video, and a discussion forum. A separate “none of these resources” answer is available and cannot be combined with the other options.

| Response | Guide | Video | Forum | Explicit none | Question status |
|---|---|---|---|---|---|
| R1 | Yes | Yes | No | No | Answer available |
| R2 | Yes | No | Yes | No | Answer available |
| R3 | No | Yes | No | No | Answer available |
| R4 | Yes | No | Yes | No | Answer available |
| R5 | No | No | No | Yes | Answer available |
| R6 | Unknown | Unknown | Unknown | Unknown | Unexplained missing answer |

For this example, the inclusion rule admits the five records that explicitly answer the question, including “none.” R6 is excluded from the calculation and remains reported as missing for an unknown reason. The example does not assume that everyone invited answered.

There are **five admitted responses**, four of which select at least one resource, and **seven resource selections**: three for the guide, two for the video, and two for the forum. “None” is an answer to the question; it is not a selected resource.

## Calculate both percentages

For each resource:

- **Percentage of admitted responses** = responses selecting that resource ÷ 5 × 100.
- **Percentage of resource selections** = selections of that resource ÷ 7 × 100.

| Resource | Count | % of five admitted responses | % of seven resource selections |
|---|---:|---:|---:|
| Practical guide | 3 | 60% | 42.86% |
| Video | 2 | 40% | 28.57% |
| Forum | 2 | 40% | 28.57% |
| Resource total | 7 selections | 140% | 100% |

“Sixty percent of admitted responses mention the guide” describes three records out of five. “The guide accounts for 42.86% of resource selections” describes three choices out of seven. Writing only “Guide: 42.86%” leaves the reader to guess which base you used.

The 140% total is not a mistake: one response can select several resources. Do not force that column to add to 100%. By contrast, the selection column distributes all seven choices across three resources and totals 100% here.

Report separately: “No resource: one of five admitted responses, or 20%. One record excluded because of an unexplained missing answer.” Do not add “none” to the selection denominator after defining that denominator as actual resources selected.

The count of seven is also not seven people. In this example each positive record can contribute more than one selection. If your collection permits repeated submissions, establish what a row represents before calling a row count a count of unique respondents.

## State which respondent base you mean

Some tables use only cases that selected at least one option as their base. Kent State demonstrates that convention in its example. It is not a universal rule for every questionnaire.

In this example, restricting the base to the four records with a resource selection gives three out of four, or 75%, for the guide. That answers a different question: “Among responses mentioning at least one resource, what share mentions the guide?” It does not replace the 60% among all five admitted responses.

Choose the base according to the question you want to describe, before choosing whichever percentage looks more favorable. State the total row count, exclusions, treatment of “none,” and exact denominator in the table heading or note. When comparing groups, each group needs its own explicit base. Several selections within one response do not become independent observations.

This item is not an [ordered Likert item](/en/blog/likert-scale-examples-response-options-and-analysis-errors). Resources are categories: their numeric codes, if any, do not support a meaningful average. The table describes reported resource use, not the intensity of a single ordered opinion.

## What if every checkbox is empty?

Kent State highlights a useful limitation: when there is no explicit “none” option, it may be impossible to separate someone with nothing to select from someone who skipped the question. Do not rewrite all no-selection records as “none” after collection.

For a future questionnaire, check whether an explicit “none,” “other,” or “not applicable” option fits the question. “None” should be incompatible with selecting a positive option. For data already collected, preserve the uncertainty and explain how it affects the reporting base instead of manufacturing an answer.

Consulting a resource does not establish that it helped or that the respondent preferred it. An open follow-up such as “In what situation did this resource help you, or fall short?” can recover circumstances and exceptions. The [open-response analysis method](/en/blog/analyzing-200-open-responses-without-bias-a-method-for-actionable-decisions) explains how to connect those accounts to source excerpts.

If your file contains those contextual comments, you can [manually import a compatible file into Harmate](/en/product/response-import) as CSV, JSON, or XLSX, then review the column mapping. Defining denominators and preparing the descriptive table remain your analytical work. This import flow does not promise native multiple-response statistical calculations.