---
url: "https://questionstar.com/blog/aggregating-multi-item-likert-scales/"
title: "Aggregating multi-item Likert scales — when to use the mean, and when not to"
description: "When several Likert items measure the same construct: mean, sum score, factor score — when each form of aggregation makes sense."
lastmod: "2026-07-08"
type: "blog"
---

# Aggregating multi-item (Likert) scales

*How do I combine several items into a single overall value?*

When you use several items to measure a construct, one question comes up: how do I combine the responses into a single overall value?

Imagine you measure job satisfaction not with a single question but with five. That's methodologically sound — because a construct like satisfaction can rarely be captured in a single question. But then you're sitting in front of five values per person and have to decide: what do you do with them?

There are several ways, and each has its place. The simplest — and, for many use cases, perfectly sufficient — is the mean. The most complete is factor analysis. In between lies a range that we'll sort out here.

## What are the options?

Three approaches are the most commonly used in practice. We'll go through them one by one — from the simple to the involved — and each time we'll tell you when the effort is worth it.

- Mean or sum per person — quick, robust, sufficient in most cases.
- Simple index construction with weights — when items differ in importance.
- Factor scores from a factor analysis — when you want to test the structure of the items or cleanly separate several subscales.

## Mean and sum — the standard approach

For each person, you take the average across all items. If item 3 is missing, QUESTIONSTAR computes the average from the available values. That's robust against individual gaps — and it's exactly what most research teams do. Even when they don't spell it out that clearly in the methods section.

> 💡 **By the way**
> 
> When all items point in the same direction, the mean and the sum are mathematically the same result — just scaled differently. Which form you choose is a matter of taste. We recommend the mean, because it stays on the original scale.

## Simple index construction with weights

Sometimes items are not equally important. In that case, you multiply each item value by a weight and divide the sum by the sum of the weights. The result is a weighted average.

The weights themselves, however, are a delicate matter. You can justify them substantively ("safety matters three times as much to us as comfort") or derive them empirically. With substantively justified weights, transparency is mandatory — write down where the numbers come from.

## Factor scores from a factor analysis

If you have many items — say 12 or more — and suspect that several constructs lie behind them, then a factor analysis is the right tool. It looks at which items vary together and forms groups from them.

The result is factor scores per person — essentially standardized aggregates. They have a mean of 0 and a standard deviation of 1. That's statistically clean, but for reporting to non-statisticians it can sometimes be unwieldy.

### When is the effort worth it?

If you're using the result internally and no one asks about factor loadings, the mean is enough. If you're publishing or validating the scale, factor analysis and reliability (Cronbach's alpha) are part of it. That's the point where it's worth investing an extra hour.

## What we recommend

> When in doubt: the mean. In all other cases: think first, then compute. The more involved method is rarely also the better one — it's just the more involved one.

If you're unsure what fits your case, write to us. We're happy to answer methods questions, even without a contract.

## Conclusion

Aggregation isn't a science unto itself, but it deserves a deliberate decision. Using the mean isn't worse — as long as you know you're using it, and don't pretend in the methods section that these are factor scores.

## Related articles

- [likert-scales-explained](/blog/likert-scales-explained)
- [common-likert-scales](/blog/common-likert-scales)

*Author: Dr. Paul Marx — see [About](/about).*