How Do You Measure Something You Can't Observe? Understanding Constructs, Indicators, and Proxies

How Can You Measure Something That Cannot Be Observed Directly?

You can observe a person's height. You can record the number of classes a student attended. You can count how many papers a researcher published.

But what if your study is about motivation, trust, anxiety, self-efficacy, organizational culture, or socioeconomic status?

You cannot observe these ideas in the same direct sense. Instead, researchers infer something about them from observable evidence. That evidence might consist of questionnaire responses, behaviors, records, physiological measurements, multiple indicators, or a proxy chosen because the target phenomenon itself is difficult or impossible to measure directly.

The challenge is not merely finding something you can count. It is establishing why the thing you can observe provides defensible evidence about the thing you actually care about.

Researchers Measure Unobservable Constructs Through Observable Evidence

How Constructs Become Measurable Without Becoming the Measure

What Is a Construct?

A construct is an abstract concept created or used by researchers to describe or explain a phenomenon of interest.

Some constructs refer to psychological attributes, such as anxiety, motivation, intelligence, or self-efficacy. Others can concern social or organizational phenomena, such as social capital, institutional trust, organizational climate, or socioeconomic status.

These constructs are not directly observable in the same way that body mass, temperature, or the number of published papers is observable.

In measurement models, such unobservable concepts are often called latent variables. Researchers observe variables thought to provide information about the latent construct and use those observations to make inferences about it. SAGE's treatment of latent variables makes this distinction explicit: researchers are typically interested in the underlying constructs rather than questionnaire items or scales themselves, with measures serving as the empirical means through which constructs are assessed.

This is why understanding the broader distinction among variables, constructs, and operational definitions matters before tackling measurement.

What Is an Indicator?

An indicator is an observable measure used to provide information about a construct.

Suppose your construct is academic self-efficacy. You cannot observe “self-efficacy” directly. Instead, you might ask students to respond to carefully designed items about their perceived ability to complete relevant academic tasks.

Individual responses are observable. The underlying self-efficacy construct is inferred.

Likewise, if the construct is institutional trust, researchers might use responses to several questions concerning confidence in, expectations of, or perceptions of an institution. The exact indicators should follow from a clear definition of the construct rather than from whatever questions happen to be available.

Why Researchers Often Use Multiple Indicators

Complex constructs are rarely exhausted by one observation.

Imagine trying to measure academic engagement with one question: “Do you pay attention in class?” Even if the response is useful, engagement may also involve participation, persistence, effort, emotional involvement, or cognitive investment, depending on how the construct has been defined.

Multiple indicators can provide broader evidence about an underlying construct and allow researchers to examine how the observed variables relate to one another. Factor-analytic approaches, for example, use patterns of covariation among measured variables to investigate hypothetical underlying dimensions.

But more indicators are not automatically better. Five poorly chosen questions do not become a good measure simply because there are five of them.

What Is a Proxy?

A proxy is a substitute used to represent something that cannot be measured directly or for which the preferred measurement is unavailable or impractical.

The word is used somewhat differently across disciplines, so researchers should explain precisely what they mean by it.

Consider socioeconomic circumstances. A researcher may ideally want a rich representation involving income, wealth, occupation, education, housing, and other resources. If those data are unavailable, the researcher might use a more limited available measure as a proxy for the broader concept.

The proxy can be useful without being equivalent to the target concept.

That qualification is essential. A proxy inherits the limitations of the relationship between what you can observe and what you actually want to represent.

Indicator and Proxy Are Not Always Interchangeable Terms

Both indicators and proxies connect observable data to something researchers care about, but the reasoning is often different.

An item in a validated multi-item scale might function as an indicator of a latent construct. A readily available administrative variable used because the desired construct was not measured may be better described as a proxy.

These distinctions are conceptual rather than merely terminological. State how the measure functions in your particular research instead of assuming the label settles the issue.

Measurement Is an Inference

This is the central idea to remember.

If someone answers “strongly agree” to a questionnaire statement, you directly observed the response. You did not directly observe motivation, trust, anxiety, or another latent construct.

The researcher makes an inference from the observed response, usually together with other evidence, to the construct.

Modern validity thinking therefore focuses on the evidence supporting interpretations of measurements rather than treating validity as a permanent sticker attached to an instrument. APA PsycTests, for example, defines test validity in terms of the extent to which evidence and theory support specific interpretations of test scores for their proposed use.

This means the question is not simply:

“Is this a valid scale?”

A more useful question is:

“What evidence supports interpreting these scores as evidence about this construct for this purpose and context?”

Start by Defining the Construct Precisely

Measurement problems often begin before measurement.

Suppose three researchers say they are measuring “research success.” One means publication output. Another means citation impact. The third means whether research influences policy or practice.

No measurement procedure can resolve that disagreement until the construct itself is clarified.

Scale-development research likewise emphasizes clear conceptualization of the target construct as a fundamental step in developing valid measures.

Before selecting indicators, write down:

• what the construct means;
• what it does not mean;
• whether it has distinct dimensions;
• which population and context the definition concerns; and
• what interpretation you eventually want the measure to support.

Then Ask What Observable Evidence Should Follow From That Construct

Once the construct is defined, ask what you would reasonably expect to observe if the construct differed across cases.

If the construct is confidence in performing research tasks, relevant evidence might involve people's reported confidence in performing clearly specified research activities. If the construct concerns actual research performance, self-confidence alone would be inadequate because confidence and performance are not the same concept.

The indicator should follow from the construct rather than redefine the construct simply because the indicator is easy to obtain.

Indicators Can Come From Different Measurement Methods

Observable evidence is not limited to questionnaires.

Depending on the construct and research design, researchers may use:

• self-report items;
• observer ratings;
• behavioral measures;
• performance tasks;
• administrative records;
• physiological measurements;
• digital traces; or
• combinations of methods.

Different methods introduce different sources of error and may capture different aspects of a construct. Research examining latent variables built from self-report and objective measures has shown that the particular indicators and their biases can substantially affect what the resulting latent variable represents.

Not All Indicators Relate to Constructs in the Same Way

There is an important distinction in measurement theory between indicators treated as consequences or manifestations of an underlying construct and indicators treated as contributing to or forming a construct.

In a familiar reflective model, the underlying latent construct is theorized to generate variation in its indicators. For example, an underlying attribute may be proposed to influence how people respond to several related questionnaire items.

Other models treat observed components as contributing to a composite or construct. Methodological literature warns that causal, formative, composite, and reflective indicators should not be treated as interchangeable because different measurement assumptions follow from each formulation.

You do not need structural equation modeling for every research project. But you do need to know what relationship you are claiming between the construct and the observations used to represent it.

Do Not Choose Indicators Solely Because They Correlate

Several variables can correlate strongly without measuring the same construct.

Students' attendance, grades, study time, and academic confidence may all be related. That does not make them four interchangeable indicators of “academic success.”

Theoretical reasoning should come before statistical convenience.

Content-validity work similarly emphasizes defining the target construct and evaluating whether proposed items actually represent its content, rather than allowing statistical patterns alone to determine what the construct supposedly means.

A Proxy Needs Its Own Justification

Researchers sometimes write “X was used as a proxy for Y” as though the word proxy itself justifies the substitution.

It does not.

Whenever you use a proxy, ask:

• Why should this variable provide information about the target?
• What dimensions of the target does it capture?
• What dimensions does it miss?
• What other processes can affect the proxy?
• Could two cases have the same proxy value while differing substantially on the target?
• Is there evidence that the proxy behaves appropriately in the population and context being studied?

The weaker the connection between proxy and target, the narrower your interpretation should become.

A Measure Can Be Reliable Without Measuring the Right Construct

Reliability concerns consistency or precision. Validity concerns whether the evidence supports the intended interpretation.

A bathroom scale that consistently adds five kilograms can produce highly consistent readings while being systematically inaccurate. In construct measurement, a questionnaire can likewise produce internally consistent scores while failing to represent the construct as intended.

APA's research-methods handbook treats reliability and construct validity as distinct areas of psychometric evaluation, reflecting the fact that one does not substitute for the other.

Do not defend a measure solely by reporting a reliability coefficient.

Construct Validity Requires More Than One Convenient Correlation

Evidence concerning construct validity can involve several questions.

Does the measure adequately represent the intended content? Does it relate to measures of similar constructs in theoretically expected ways? Is it distinguishable from measures of different constructs? Does its internal structure correspond to the proposed dimensional structure?

APA PsycTests distinguishes content, convergent, discriminant, criterion, and other forms of validity evidence, while methodological guidance on scale development similarly emphasizes multiple sources of evidence when evaluating measures.

No single statistic proves that a construct has been “measured correctly” for all purposes.

Context Matters

A measure that works well in one population, language, setting, or period may not behave identically in another.

For example, translating an instrument can alter item meaning. A behavioral indicator may have different social meanings across cultures. An administrative proxy may be recorded differently across institutions.

Methodological work on measurement emphasizes that validity concerns a measure's usefulness for a particular purpose and context and that equivalence across contexts should be investigated rather than assumed.

This is why “validated scale” should not be interpreted as “universally valid everywhere.”

Connect the Construct to an Explicit Operational Definition

Once you decide which evidence will represent the construct, specify exactly how that evidence becomes data.

Which items will be administered? How are they scored? Which records will be used? What period is covered? How are multiple indicators combined? What does a higher value mean?

Those decisions turn measurement reasoning into an operational definition another researcher can understand.

Keep the Construct and the Measure Separate in Your Interpretation

Suppose you operationalize “research productivity” as the number of peer-reviewed journal articles published during the previous three years.

Your dataset contains publication counts. It does not contain “productivity itself” in every possible sense.

The measure omits books, datasets, software, patents, conference outputs, quality, contribution, disciplinary publication norms, and potentially other aspects someone might reasonably associate with research productivity.

You may still have a useful operational measure. The important discipline is to describe your findings at the level the measure supports.

How Would You Measure “Academic Engagement”?

Now suppose attendance records are the only data available.

The researcher might decide to use attendance as a proxy for one behavioral aspect of engagement. That may be useful, but it requires a more cautious claim than saying “attendance measures academic engagement.”

The difference is not pedantic. A student may attend because attendance is mandatory, while another may miss class but engage deeply with course materials. The proxy contains information, but it does not become the whole construct.

Common Mistakes When Measuring Unobservable Constructs

How to Decide Whether an Indicator or Proxy Is Good Enough

When you cannot observe the phenomenon directly, make the inferential chain explicit rather than hiding it behind a variable name.

The final test is alignment. The construct in your conceptual framework should be recognizably connected to the indicators in your data and to the interpretation you eventually make.

If those connections break, even sophisticated statistical analysis cannot restore alignment across the research question, framework, measurement, and analysis.

Construct Measurement Checklist

Frequently Asked Questions About Constructs, Indicators, and Proxies

You Observe the Indicator and Infer the Construct

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.