Concepts, Constructs and Variables

Overview

Health research begins with ideas such as treatment adherence, access to care, symptom burden, social support, quality of life and continuity of care. These ideas are not all observable in the same way. Some can be measured directly, whereas others are broad or abstract and must be represented by one or more observable indicators.

A concept is a general idea used to organise observations. A construct is a concept deliberately defined for a particular theoretical or research purpose, especially when it cannot be observed directly. A variable is a characteristic that can take different values across people, groups, places, occasions or records. An indicator is an observable item used as evidence about a concept or construct. A data item is the actual field recorded in a dataset [1–3].

These terms are related but not interchangeable. Moving too quickly from an abstract idea to a convenient database field can change the research question without the team noticing.

Why this matters

Weak conceptualisation creates problems that statistical analysis cannot repair. A study may claim to investigate access to care but measure only distance to a clinic. It may claim to assess adherence but use prescription issuance rather than medicine-taking behaviour. It may label a single satisfaction question as quality of life. In each case, the recorded variable captures only part of the intended idea.

Clear distinctions also improve communication. Clinicians, patients, laboratory scientists, qualitative researchers and statisticians may attach different meanings to the same word. A shared conceptual map makes assumptions visible before instruments, sample-size calculations and analysis plans are finalised.

The conceptual ladder

Research commonly moves through several levels.

Phenomenon

The phenomenon is the broad real-world occurrence requiring explanation, description or evaluation. Here, the phenomenon is incomplete postpartum glucose follow-up.

Concept

A concept names an aspect of the phenomenon. Examples include access, continuity of care, perceived susceptibility and family support.

Construct

A construct gives a concept a more specific meaning within a theoretical or research framework. For example, perceived access burden could be defined as the practical and psychosocial difficulty a woman expects when arranging and attending postpartum testing.

Indicator

An indicator is observable evidence intended to represent all or part of the construct. Travel time, transport cost, childcare availability and perceived difficulty obtaining an appointment could be indicators of access burden.

Variable

A variable specifies how an indicator or characteristic will vary. Travel time may be recorded in minutes; childcare availability may be recorded as available, uncertain or unavailable; perceived appointment difficulty may be recorded on a multi-item scale.

Data item

The data item is the field actually captured, such as travel_minutes, childcare_available or a response to questionnaire item 4. A construct may require several data items, and one data item may contribute to a derived variable.

Concept-to-data map for postpartum follow-up

illustrative

Table 1. Translating ideas into study elements

Research idea	Construct or concept	Possible indicator	Variable or data item
Access	Perceived access burden	Expected difficulty attending testing	Mean score from four prespecified questionnaire items
Continuity	Coordination between maternity and primary care	Appointment arranged before discharge	yes/no from discharge documentation
Social support	Practical support for attendance	Someone available to provide childcare	available/uncertain/unavailable
Follow-up	Postpartum glucose-testing completion	Eligible test completed within the study window	yes/no from linked records

Note: The entries are illustrative. Their adequacy would require evidence, stakeholder input and measurement evaluation.

Observable variables and latent constructs

Some characteristics can be observed or recorded with relatively little abstraction: age, laboratory concentration, date of admission and treatment allocation. Even these require definitions and measurement procedures, but the characteristic itself is comparatively concrete.

A latent construct cannot be observed directly. Depression severity, health literacy, trust, burnout and perceived barriers are inferred from responses, behaviours or other indicators. A single observed item may be too narrow or unstable to represent a multidimensional construct. Health-measurement guidance therefore emphasises defining the construct and its intended interpretation before selecting or developing a scale [2,3].

A construct can be:

• unidimensional, when indicators are intended to represent one dominant dimension;
• multidimensional, when several related dimensions form the broader construct; or
• context-dependent, when meaning changes by population, setting, language or purpose.

Calling a variable “validated” is insufficient. Evidence supports interpretations and uses of scores in specified contexts, rather than permanently validating an instrument for every population and purpose [4].

Indicators are not the construct itself

An indicator may capture only one expression of a construct. Clinic distance is one component of access, but access can also depend on cost, opening hours, appointment availability, mobility, safety, caregiving responsibilities and perceived acceptability. Using distance alone should therefore be described as measuring geographic proximity, not comprehensive access.

Indicators can also disagree. A participant may live close to a clinic but experience high access burden because appointments are unavailable during working hours. Such disagreement is not necessarily a data error; it may reveal that the construct has several dimensions.

APPRAISAL CHECKLIST

Appraise the link between construct and variable

• Is the construct clearly defined for this study?
• Do the selected indicators cover its important dimensions?
• Are indicators relevant and understandable to the target population?
• Does the measurement method preserve the intended timing and context?
• Is the planned score or category supported by evidence rather than convenience?
• Could another interpretation of the recorded variable be equally plausible?

Variables vary by unit and occasion

Variables do not vary only between individual participants. They may vary between:

• clinics, hospitals, districts or countries;
• clinicians or research staff;
• pregnancies, admissions or treatment episodes;
• repeated visits within a person;
• lesions, specimens or slides within a patient; and
• days, weeks or longer follow-up periods.

The unit of observation is the entity on which a value is recorded. The unit of analysis is the entity represented in the main analysis. These may differ. For example, several appointment attempts may be recorded for each woman, but the primary outcome may be completion per woman. Failing to recognise this hierarchy can lead to duplicated denominators or analyses that treat correlated observations as independent.

A variable has no permanent research role

Age can be an outcome in demographic estimation, an exposure in life-course research, a predictor in a risk model, a confounder in an aetiological study or a stratification variable in reporting. The role comes from the question, causal assumptions and analysis purpose—not from the data type or column name.

Terms such as independent variable and dependent variable can be useful in simple settings, but they may hide important distinctions among exposure, intervention, predictor, confounder, mediator and effect modifier [1]. Later lectures in this part make those roles explicit.

Building a defensible construct map

A practical sequence is:

1. State the research question and intended inference.
2. Define each central concept in words.
3. Identify whether it is directly observable or latent.
4. Describe its dimensions and boundaries.
5. Review theory, previous studies and stakeholder perspectives.
6. Select indicators that represent the important dimensions.
7. Specify variables, instruments, sources and time points.
8. Anticipate measurement error and missingness.
9. Confirm that the planned analysis uses the same interpretation.
10. Revise the map after pilot testing, without hiding substantive changes.

PROTOCOL CHECKLIST

Document the conceptualisation before data collection

• central phenomenon and research purpose;
• definitions of key concepts and constructs;
• assumed dimensions and boundaries;
• proposed indicators and rationale;
• observable variables and data sources;
• unit and timing of measurement;
• score or derivation rules;
• evidence supporting the intended interpretation; and
• limitations of what the variables do not capture.

Ethical and equity considerations

Construct labels can carry judgement. Terms such as non-compliant, poorly motivated or hard to reach may place responsibility on participants while ignoring service barriers. Researchers should use respectful language, involve relevant communities when defining sensitive constructs and examine whether indicators function differently across language, disability, socioeconomic or cultural groups.

Interface with statistical analysis

Conceptualisation precedes decisions about coding and modelling. The analysis team needs to know:

• whether a variable is directly observed or a derived score;
• whether categories are natural, clinically defined or imposed by the researcher;
• whether repeated indicators are intended to form one construct;
• which level generates the variation;
• what missing values mean; and
• which interpretation the estimate is intended to support.

Detailed data-type classification belongs in the related Biostatistics lecture. The methodological priority here is preserving meaning from the question to the recorded value.

Reflective and formative measurement modelsMMed extension

In a reflective model, the latent construct is assumed to give rise to the indicators. Indicators of depressive symptom severity, for example, may covary because they reflect an underlying condition. In a formative model, indicators jointly define or form the construct; transport cost, travel time and appointment inflexibility may together form access burden. Removing a formative indicator can change the construct’s meaning. This distinction affects item selection, model evaluation and interpretation, and should not be decided solely from correlations among items [2].

Worked example

WORKED EXAMPLE

Defining access burden without losing the research question

Initial idea: Women fail to complete testing because access is poor.

Problem: “Access” is too broad, and failure may also arise from referral breakdown, low perceived need or record-linkage problems.

Refined construct: Perceived access burden is the anticipated practical difficulty of arranging and attending a postpartum glucose test within the local care pathway.

Dimensions: appointment availability, travel, cost, caregiving constraints and time away from work.

Indicators: travel time; direct cost; childcare availability; ability to obtain a suitable appointment; perceived difficulty attending.

Variables: continuous travel minutes and cost; categorical childcare availability; item responses using prespecified response options; a composite score only if its construction and interpretation are defensible.

Boundary: The construct does not include awareness of long-term diabetes risk or whether a referral was issued. These are represented separately.

Resulting benefit: The study can distinguish practical access burden from knowledge, motivation and service coordination rather than collapsing all non-completion into one vague explanation.

Suggested protocol or reporting wording

Common mistakes

Knowledge check

Is travel time to a clinic a complete measure of access to care?

Answer and reasoning

No. Travel time is an indicator of geographic or transport burden. Access may also include affordability, appointment availability, opening hours, physical accessibility, acceptability and competing responsibilities. The variable should be labelled according to what it measures unless additional indicators adequately represent broader access.

A questionnaire has six items intended to measure trust in healthcare. What is the construct and what are the variables?

Answer and reasoning

Trust in healthcare is the intended construct. Each item response is an observed variable or data item. A total or mean score is a derived variable whose interpretation requires a justified scoring model and supporting measurement evidence.

Can age be an outcome in one study and a covariate in another?

Answer and reasoning

Yes. A variable’s role is defined by the research question and analysis purpose. Age could be the quantity being described, an exposure, a predictor, an adjustment variable or a stratification variable.

Why is a statistically reliable score not automatically a valid representation of a construct?

Answer and reasoning

Reliability concerns consistency or precision. A score can be consistent while measuring only a narrow or unintended attribute. Validity requires evidence that the interpretation and use of the score are appropriate for the construct, population and purpose.

Key take-home messages

SUMMARY

Key take-home messages

• Concepts organise ideas; constructs specify abstract ideas for research; variables and indicators make them observable.
• The study records representations of a construct, not the construct itself.
• A variable’s research role depends on the question, causal assumptions and intended inference.
• Good conceptualisation states what each measure captures and what it leaves out.
• Ethical conceptualisation avoids blaming participants for structural or service-level barriers.

References

1. Andrade C. A student’s guide to the classification and operationalization of variables in the conceptualization and design of a clinical study: part 1. Indian J Psychol Med. 2021;43(2):177–179. https://doi.org/10.1177/0253717621994334
2. DeVellis RF, Thorpe CT. Scale development: theory and applications. 5th ed. Thousand Oaks (CA): Sage; 2021.
3. Streiner DL, Norman GR, Cairney J. Health measurement scales: a practical guide to their development and use. 5th ed. Oxford: Oxford University Press; 2015.
4. Messick S. Validity of psychological assessment: validation of inferences from persons’ responses and performances as scientific inquiry into score meaning. Am Psychol. 1995;50(9):741–749. https://doi.org/10.1037/0003-066X.50.9.741