CampusDown

Home › Guides

Concept

Nominal, Ordinal, Interval or Ratio? Levels of Measurement in Social Work Research

CampusDown Editorial · 2026. 10. 28. · Updated 2026. 11. 13.
Generic AI wording will not get you ahead.Find original angles for your work on CampusDown.
Key takeaways
  • Measurement assigns numbers or symbols to abstract concepts such as depression or wellbeing according to a fixed rule.
  • Stevens' four levels, nominal, ordinal, interval and ratio, carry progressively more information.
  • The level you choose decides which statistics are valid, so it must be settled at the design stage.
Contents
  1. What is measurement in social work research?
  2. How do the four scales differ?
  3. Which statistics can each scale use?
  4. How are welfare scales built?
  5. How can measurement error be reduced?
  6. FAQ

In social work research, measurement means turning abstract ideas such as depression, wellbeing or self-esteem into numbers or symbols by following a fixed rule, and the level of measurement (nominal, ordinal, interval or ratio) decides which statistics you are allowed to use. If you code gender as 1 and 2, does the average of those numbers mean anything? This guide explains Stevens' four scales with examples from UK and US practice, the statistics each one supports, how welfare scales are built and how to cut measurement error.

What is measurement in social work research?

Social workers constantly deal with things nobody can see directly: how low someone's mood is, how connected they feel to their community, how well they manage daily life. Measurement is the systematic process that makes these concepts workable by attaching numbers or symbols to them according to an agreed rule.

Painterly scene of a young social worker on a footstool gently showing a clipboard questionnaire headed HOW ARE YOU FEELING? to an elderly woman with a cup of tea in a cosy terraced-house living room on a rainy afternoon, illustration by CampusDown
Every score in a case file starts as a conversation: measurement turns a feeling into something you can track (CampusDown)

Accurate measurement is what lets you compare a client before and after an intervention and show whether a programme works. In England, for example, NHS Talking Therapies services routinely record the PHQ-9 depression and GAD-7 anxiety questionnaires at sessions so that recovery can be tracked. Measurement is therefore not a technical extra but part of professional accountability.

Ad

How do the four scales differ?

The psychologist S. S. Stevens set out four levels of measurement in a 1946 paper in Science and argued that each level permits different mathematical operations. Each level adds information to the one before it.

Stevens' four levels of measurement
ScaleKey featureExamples
NominalCategories only, no orderGender, region, religion, census ethnic group
OrdinalOrdered, but gaps are not equalEducation level, satisfaction ranking, social class
IntervalEqual gaps, no true zeroTemperature in Celsius, IQ scores (conventionally)
RatioEqual gaps and a true zeroAge, income, number of service visits

Nominal categories must be mutually exclusive and exhaustive: every case fits exactly one box. The ethnic group question in the census for England and Wales is a familiar nominal example, which is why it includes "other" options so nobody is left out.

How can you tell the scales apart in an exam?

Ask three questions in order. Is there an order? Are the gaps equal? Does zero mean "none at all"? A "no" to the first gives nominal; a "yes" then "no" gives ordinal; equal gaps without a true zero give interval; all three "yes" answers give ratio. This simple routine solves most multiple-choice questions.

IQ is usually listed as interval in textbooks, but some methodologists argue it is closer to ordinal because nobody can prove that a 10-point gap means the same thing everywhere on the scale. Mentioning that debate can lift an essay above a list of definitions.

Skimming the section headings first makes a long guide much easier to map out. Try writing down one question of your own as you read.

Which statistics can each scale use?

Because higher levels carry all the information of the lower ones, the range of permitted statistics widens as you move up the ladder from nominal to ratio.

Permitted operations and typical statistics by scale
ScaleOperationsTypical statistics
NominalEqual or not equalFrequencies, mode, chi-square test
OrdinalGreater or less thanMedian, rank correlation
IntervalAdd and subtractMean, standard deviation, t-test
RatioMultiply and divideAll statistics, including "twice as much"

Income measured in pounds or dollars can later be grouped into low, middle and high bands, but you cannot go back from bands to exact amounts. So it usually pays to measure at the highest sensible level and simplify later.

The gender example from the start is the warning here: software will happily average a column of 1s and 2s, so the researcher, not the program, has to know that those numbers are only labels.

Ad

How are welfare scales built?

Building a scale is a sequence of decisions rather than a single act. The usual steps are to define the concept and its sub-dimensions, write items for each dimension, choose a response format such as a five-point agree–disagree scale, and then test reliability and validity.

Painterly farmhouse kitchen scene of a grandmother and granddaughter baking, the girl levelling a steel measuring cup beside a brass balance scale, thermometer and measuring spoons, a recipe card reading MEASURE WITH CARE, illustration by CampusDown
Different things need different tools: cups for volume, a balance for weight, a thermometer for heat, and the right scale for each concept (CampusDown)

Summed scales for depression or self-esteem are strictly ordinal, but research often treats the total as interval. Rensis Likert's 1932 agree–disagree format is the best-known response style. Well-known examples include the ten-item Rosenberg Self-Esteem Scale from 1965, the PHQ-9, whose nine items scored 0 to 3 give a total from 0 to 27, and the Warwick-Edinburgh Mental Wellbeing Scale used in public health across the UK.

Using an existing standardised scale makes your results comparable with other studies, so it is efficient to search for one before writing your own. Check that it has been validated for your population and language, and respect any licence terms attached to it.

What extra concepts deepen an assignment?

Five ideas extend the basics. A reliability coefficient, most often Cronbach's alpha, shows on a 0 to 1 scale how consistently items measure the same concept; around 0.7 is often treated as acceptable, but that is a rule of thumb, not a law. A Guttman scale orders items cumulatively, so agreeing with a demanding item implies agreeing with all easier ones, as in questions about social distance from disabled neighbours. A Thurstone scale uses expert judges to assign a value to each statement.

Extended measurement concepts and how to use them
ConceptCore ideaUse in an assignment
Reliability coefficientConsistency of itemsChecking measurement error
Guttman scaleCumulative itemsExample of scale construction
CodebookRules for entering dataLevels of measurement in practice
Missing data handlingDealing with blank answersStep before analysis
Thurstone scaleJudge-rated itemsWhy building scales is hard

A codebook records each variable's name, the number given to each answer, its level of measurement and how non-responses are coded. Missing data need a stated plan: dropping every incomplete case shrinks the sample and can bias results, so look at why answers are missing before choosing a method, and report it.

How can measurement error be reduced?

Every measurement contains error. Ambiguous wording or respondents trying to look good create systematic error; fatigue or a noisy room create random error. Clear wording, a pilot study and consistent survey conditions are the basic defences.

The ideas behind those defences are covered in reliability and validity of measurement: validity asks whether you measure what you intend, reliability asks whether repeated measurement gives similar results. The broader logic of error, bias and uncertainty is set out in measurement error and uncertainty.

Painterly hallway scene of a father kneeling to pencil his small daughter's height on a white doorframe under the words MY HEIGHT, tape measure in hand, golden afternoon light, illustration by CampusDown
Height has a true zero, so a child who grows from 80 cm to 120 cm really is one and a half times as tall: that is a ratio scale (CampusDown)

Good measurement captures what you want accurately and gives similar results when repeated. When both are in place, survey findings can carry real weight as evidence for changing services and, ultimately, people's lives.

Compiled by the CampusDown editorial team from a Korean article on measurement and scales in social welfare research, adapted with public information on NHS Talking Therapies outcome measures, the census for England and Wales and widely used standardised scales. Sources: Stevens, S. S. (1946), "On the theory of scales of measurement", Science 103(2684), 677–680; Likert (1932); Rosenberg (1965); standard social research methods texts. Reviewed November 2026. This is general study information.

Ad

FAQ

Is a Likert scale ordinal?

Strictly yes, but the summed score of several items is often analysed as if it were interval data.

What scale is age?

Exact age is a ratio scale with a true zero; grouped into age bands it becomes ordinal.

Why is Celsius temperature not a ratio scale?

Because 0 degrees does not mean no heat, so 20 degrees is not twice as hot as 10 degrees.

What level of measurement is the PHQ-9?

Each item is ordinal; the 0 to 27 total is commonly treated as interval for analysis and for tracking change.

What Cronbach's alpha is acceptable?

Around 0.7 or above is often treated as acceptable, but it is a rule of thumb and depends on the purpose and number of items.

Ad

Found this helpful? Share it

Send it to a friend who needs it.

X Facebook Threads