Scott OatleySociology / UCL
← Guides

Measuring social class: an interactive guide

A walk-through of the main measures of social class, how each is built, and how to choose between them

Oatley, S. (2027) ‘Measuring social class’, in Encyclopedia of Measurement in Social Sciences, 2nd edn, Vol. 3, pp. 394–402. Elsevier. Read the chapter.

There is no single agreed way to measure social class. Measures differ because they rest on different theories (Weberian, Marxian, Durkheimian and Bourdieusian) and because some are not grounded in theory at all. This guide follows the chapter’s structure. Each section explains how a measure is built and what it costs to use, with a widget to try it out.

The chapter’s advice in brief

  • Avoid subjective and ad hoc measures unless self-perception is the topic.
  • Use validated measures, and the most up-to-date version unless the analysis is historical.
  • Where several validated measures are available, compare them in a sensitivity analysis rather than choosing on theory alone.

1What not to use: subjective and ad hoc measures

Subjective measures ask people to place themselves on a class ladder or name their class. They suit research on class identity, but they are not comparable across people: one person’s “middle class” is another’s “working class”. An occupation question is hard to misread in comparison: a cleaner reports being a cleaner.

Ad hoc measures are home-made indices, often called socio-economic status (SES), that combine income, education, occupation and other items. They lack theory and outside validation. Income and education are epiphenomenal to class: they are related to it, but they are not class. If you use such an index, call it SES, not social class.

Sort these measures
MeasureYour call

2Occupation-based measures

Since the 1970s the dominant view has been that occupations are the backbone of the class structure. Occupation-based measures are reliable and parsimonious. They are built from two pieces of information:

Turning free-text job descriptions into codes is slow, even with tools such as CASCOT, and there are hundreds of codes. Automated tools such as socstrat then turn codes and status into a class measure.

Three families of occupation-based measure

FamilyBasisExamples
A prioriAssumed hierarchy of occupational standing or skill; little theoryRegistrar General’s Social Classes (RGSC), first used 1911
Weberian / GoldthorpianEmployment relations: market situation (security, prospects) and work situation (authority, control)EGP, NS-SEC (official UK measure since 2001), ESeC
Micro-classDurkheimian: detailed occupations themselves are the classes80–100 micro-classes

3Schema explorer

Weberian schemas have many categories, so each comes with official ways to collapse it. Fewer classes give more people per class (more statistical power) but less detail. Pick a schema and a version to see which full classes are grouped together.

Full and collapsed versions

RGSC divides non-manual from manual work. It is easy to understand, but it is a priori and failed attempts at validation; the chapter judges its strengths too weak to outweigh its weaknesses. EGP was the first theory-led schema and works well across countries, though its documentation was long hard to obtain. NS-SEC grew out of EGP after the ESRC review and is the measure to use for UK analysis. ESeC uses ISCO codes and is built for cross-national European comparison.

4Micro-classes

The micro-class approach keeps the belief that occupation is central, but rejects grouping occupations into a few “big classes”, which hides real differences between detailed occupations. The costs are 80–100 categories, which strain statistical power even in large samples, and few ready-made derivations. The gain is detail. Micro-classes also fit naturally as a second level in a random-effects (multilevel) model. Section 7 shows what that many categories does to cell sizes.

5Wright’s neo-Marxian schema

Erik Olin Wright’s schema centres on exploitation rather than market position. It divides owners by how many people they employ, and divides wage labourers along two axes: organisational assets (manager, supervisor, neither) and skill or credential assets (expert, semi-skilled, unskilled). Answer the questions to place a person.

Place a person in Wright’s 12-class schema

Wright’s measure is far harder to build than a Weberian one. It needs eight sets of survey items, turned into seven intermediate typologies, and few surveys ask them all. Anyone with a missing answer on any item cannot be classified, although missing-data methods can fill some of these gaps.

6Inductive measures: capitals, assets and resources

All the measures above are deductive: a fixed schema says which class each answer pattern belongs to. Inductive measures, inspired by Bourdieu, estimate classes from each sample using latent class analysis, factor analysis or multiple correspondence analysis on economic, social and cultural capital. The best known is the Great British Class Survey. One sample may give three classes and another nine, and the classes have no fixed order. They can also be expressed as a scale, which helps when many categories would cost too much power. Like Wright’s schema, these measures need many survey items.

7Categories and sample size

A recurring trade-off in the chapter is detail versus statistical power. This widget spreads a sample across classes using the NS-SEC distribution of the NCDS cohort at age 33 (from the socstrat paper). For each class it shows the expected number of people and the 95% margin of error for a proportion of 50% estimated within that class.

How thin do the classes get?

Try thisAt 1,000 people compare 8 and 3 classes. Then switch to micro-classes and find the sample size at which the average class reaches 100 people.

8Choosing a measure

The chapter’s recommendations depend on the question and on what the data allow. Describe your study to get them.

Is the topic people’s own sense of class?
Where is the comparison?
What does the survey have?
Sample and model

9Check yourself

10Reading