What Is a Construct in Quantitative Research?
What Is a Construct?
A construct is an abstract research concept that you want to measure in a quantitative model. Satisfaction, perceived usefulness, service quality, repurchase intention, and loyalty are all examples of constructs. You cannot observe these concepts directly in the same way that you observe age or income, so you measure them through multiple items in a questionnaire.
In a research model, a construct is usually shown in a shape representing a latent variable. Each construct can be an independent variable, dependent variable, mediator, or moderator. For example, in a model of intention to use an application, perceived usefulness may affect intention to use. Both are constructs, while the items PU1, PU2, PU3, and IU1, IU2, IU3 are observed variables used to measure them.
You can picture the structure in three layers:
- Construct or latent variable: the abstract concept being studied, such as service quality.
- Observed variable or indicator: an item in the questionnaire, such as “The staff served me quickly.”
- Measurement score: the participant’s response value, usually measured on a 1 to 5 or 1 to 7 Likert scale. The Likert scale was presented in Likert’s original study (Likert, 1932).
When you ask “what is a construct,” the most practical answer is this: a construct is a theoretical concept turned into a variable that can be tested through one or more observed variables. Choosing a construct must go together with a conceptual definition, a scale source, and an appropriate testing procedure.
Why Constructs Matter in Quantitative Research
A construct determines what your model is explaining. If a construct name is too broad or overlaps with another construct, your hypotheses become difficult to test and your questionnaire may contain overlapping items. Before opening SPSS or SmartPLS, you need to know which concept each construct represents and why it belongs in the model.
A construct also connects your theory to your data. For example, hypothesis H1 might state: “Perceived usefulness has a positive effect on intention to use.” In this sentence, perceived usefulness and intention to use are two constructs. The effect and its direction describe the relationship between them.
When you analyze data in SPSS, you usually check the items for each construct with Cronbach's Alpha and EFA. When you analyze data in SmartPLS, you assess the measurement model first and then the structural model. Assessing the measurement part before the structural part follows the two-step logic presented by (Anderson and Gerbing, 1988).
Constructs also help you identify each variable’s role in the model. A construct can be:
- Independent variable: a construct that predicts or explains another construct.
- Dependent variable: a construct affected by one or more other constructs.
- Mediator: a construct that explains the mechanism linking an independent variable to a dependent variable. You can also read what is a mediator.
- Moderator: a construct that changes the strength or direction of the relationship between two variables. This is different from a mediator. See also what is a moderator and what is moderating.
If a construct is a latent variable, you need to state clearly which observed variables measure it. A useful construct description table normally includes the construct code, definition, number of items, reference source, and response scale. This section connects the theoretical foundation chapter, the methods chapter, and the results.
How Many Indicators Are Enough for a Construct?
There is no single threshold that lets you judge a construct just by looking at its name. You assess a construct through the quality of its items and its measurement model. In SPSS, the common indicators are Cronbach's Alpha, Corrected Item-Total Correlation, KMO, Bartlett's Test, factor loading, and Total Variance Explained. In SmartPLS, you usually examine outer loading, Composite Reliability, AVE, HTMT, and VIF.
The table below contains thresholds commonly cited in quantitative research. Interpret each threshold alongside the research context, the type of construct, and the theoretical argument. Do not use a threshold as an automatic delete button.
| Assessment | Reference threshold | Canonical source |
|---|---|---|
| Cronbach's Alpha for a scale | 0.7 or above | (Nunnally, 1978) |
| Alpha for a new scale or exploratory research | May be 0.6 or above | (Hair et al., 2010) |
| Corrected Item-Total Correlation | 0.3 or above | (Nunnally and Bernstein, 1994) |
| KMO in EFA | 0.5 or above | (Kaiser, 1974) |
| Bartlett's Test | p-value below 0.05 | (Kaiser, 1974) |
| Factor loading in EFA | 0.5 or above | (Hair et al., 2010) |
| Total Variance Explained | 50% or above | (Hair et al., 2010) |
| Outer loading in PLS-SEM | 0.7 or above | (Chin, 1998) or (Hair et al., 2022) |
| Composite Reliability | 0.7 or above | (Fornell and Larcker, 1981) |
| AVE | 0.5 or above | (Fornell and Larcker, 1981) |
| HTMT | Below 0.85, or below 0.90 for closely related constructs | (Henseler et al., 2015) |
| VIF in the structural model | Below 5 | (Hair et al., 2019) |
This table gives you a starting point, but it does not replace reading the construct definition. For example, a low Alpha may result from an item that does not match the content, but it may also indicate that the construct is multidimensional. If you delete an item only to increase Alpha without explaining the content decision, the committee may ask why that item was removed.
When you use SmartPLS, do not place CFI, TLI, or RMSEA in your PLS-SEM assessment table. These fit indices belong to CB-SEM and are discussed in (Hu and Bentler, 1999). Read SmartPLS through PLS-SEM measures such as outer loading, CR, AVE, HTMT, VIF, R², and Q².
How to Read a Construct in the Output
If SPSS is open, the software usually does not print a table named “Construct.” You identify the construct yourself from the group of items specified in your questionnaire and then read the relevant output tables. In SmartPLS, the construct normally appears directly in the diagram, while the measurement results appear in tables such as Outer Loadings, Construct Reliability and Validity, and Discriminant Validity.
The following is illustrative output, not the result of a real study. Suppose the construct “Perceived usefulness” is measured with PU1 through PU4, while the construct “Intention to use” is measured with IU1 through IU3.
| Software table or column | Construct or variable | Illustrative value | How to read it |
|---|---|---|---|
| Reliability Statistics, Cronbach's Alpha | Perceived usefulness | 0.842 | The internal consistency of PU1 through PU4 is at a commonly accepted level according to (Nunnally, 1978) |
| Item-Total Statistics, Corrected Item-Total Correlation | PU3 | 0.418 | PU3 correlates with the total scale above 0.3 according to (Nunnally and Bernstein, 1994) |
| KMO and Bartlett's Test, KMO | Constructs entered into EFA | 0.781 | The data meet the reference KMO threshold from (Kaiser, 1974) |
| Rotated Component Matrix, PU2 | Perceived usefulness | 0.736 | PU2 loads on the expected factor with a loading above 0.5 according to (Hair et al., 2010) |
| Outer Loadings, PU2 | Perceived usefulness | 0.812 | PU2 has an outer loading above 0.7 according to (Chin, 1998) |
| Construct Reliability and Validity, Composite Reliability | Perceived usefulness | 0.889 | Composite reliability is above 0.7 according to (Fornell and Larcker, 1981) |
| Construct Reliability and Validity, AVE | Perceived usefulness | 0.666 | The construct explains sufficient convergent variance under the AVE threshold of 0.5 from (Fornell and Larcker, 1981) |
| Discriminant Validity, HTMT | Perceived usefulness and intention to use | 0.721 | The two constructs have HTMT below 0.85 according to (Henseler et al., 2015) |
In SPSS, the usual starting path is Analyze > Scale > Reliability Analysis. Move PU1 through PU4 into the Items box, select Alpha as the Model, then open Statistics and select Scale if item deleted. Read Reliability Statistics for Alpha and Item-Total Statistics for Corrected Item-Total Correlation. This result describes the reliability of the item group. It does not prove that the entire theoretical model is correct.
To run EFA, go to Analyze > Dimension Reduction > Factor. In the dialog box, select the items for analysis, open Descriptives and select KMO and Bartlett's test, open Extraction to examine eigenvalue, and then open Rotation to choose a rotation method that fits your research design. The tables to check are KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix.
In SmartPLS 4, the construct is the circle or latent variable on the canvas, while the observed variables are rectangles connected to it. After running PLS Algorithm, read Outer Loadings and Construct Reliability and Validity. After running Bootstrapping, move to Path Coefficients to check the relationships between constructs. Read each table according to its role: outer loading assesses the measurement part, while the path coefficient assesses relationships in the structural model.
What to Do When a Construct Does Not Meet the Threshold
Work through the problem in this order: data, item content, and then the model. Do not delete a group of items just because one index is low.
- Check coding and input data. Confirm that reverse-coded items were recoded, missing values were not entered as 0, and column names did not shift when the
.savor.csvfile was imported. A coding error can reduce Alpha or loading without saying anything about construct quality. - Read the item content. Check whether the item really measures the same concept. If PU3 asks about ease of use while PU1 and PU2 ask about benefits, the problem is in the scale content, not only in the output number.
- Check the statistic if the item is deleted. In
Item-Total Statistics, theCronbach's Alpha if Item Deletedcolumn shows how Alpha changes if the item is removed. Use this as evidence for a decision, not as an instruction to delete. - Run the analysis again after each deletion. If both the theory and the statistics support the decision, remove one item, record the reason, run the analysis again, and check the entire group. Save every output version so you can explain the process.
- Review the construct. If the items repeatedly split into several groups, the construct may be multidimensional, or the model may be combining different concepts. Returning to the theoretical model is often more useful than trying to force the output above a threshold.
The procedure differs for a formative construct and a reflective construct. In a reflective construct, the construct is treated as the cause of similar manifestations, so the items usually correlate and are assessed through loading, reliability, and AVE. In a formative construct, the items form the construct, so deleting one may remove an important content dimension. Use theory, check collinearity among the indicators, and examine the significance of the weights instead of applying Cronbach's Alpha mechanically.
AMOS can draw a formative construct, but the representation must match the direction of the arrows in the measurement model. Reflective measurement usually has arrows from the construct to the observed variables, while formative measurement has arrows from the observed variables to the construct. If you are unsure which measurement type you have, determine it from the conceptual definition and the scale source before drawing the diagram in AMOS. Changing software cannot resolve an unclear construct definition.
Construct vs. Scale
A construct and a scale are related, but they are not the same thing. A construct is the theoretical concept you want to measure, while a scale is the measurement instrument or group of items used to measure that construct. You can find a more detailed explanation in what is a scale.
For example, “customer satisfaction” is a construct. The group of items SAT1, SAT2, SAT3, and SAT4, answered from 1 to 5, is a scale used to measure that construct. One construct may have several different scales developed by different authors for different contexts.
| Term | Question it answers | Example |
|---|---|---|
| Construct | Which concept do you want to measure? | Customer satisfaction |
| Dimension | Which aspects does the construct contain? | Satisfaction with price, service, and experience |
| Indicator | What is the specific observed data? | SAT1, SAT2, SAT3 |
| Scale | What measurement instrument measures the construct? | A satisfaction scale with 4 items |
| Variable | What is the data column in the file? | SAT1 with values 1, 2, 3, 4, 5 |
A construct is also easy to confuse with a variable. In everyday academic writing, an author may call a construct a latent variable. An observed variable, however, is a specific data column, while a construct is usually represented by a group of variables or a composite score. In Chapter 3, use the terminology precisely so readers know whether you mean a concept or a data column.
Common Errors
The first error is taking a construct name from a paper without reading its definition. Two authors may use similar names for different content. Record the original definition, context, number of items, and translation into English before placing the items in your questionnaire.
The second error is calling every questionnaire item a construct. A statement such as “This application is easy to use” is usually an observed variable, not an independent construct. A construct may contain several observed variables, except when theory and context support a single-item scale.
The third error is using Cronbach's Alpha for every type of construct. Alpha is designed to assess a particular form of internal consistency, while formative constructs require a different assessment approach. If you have not identified whether your measurement model is reflective or formative, the results section can easily lead to difficult questions at the defence.
The fourth error is deleting an item because its loading or Alpha is low without recording the reason. Save the output from every round, record which item was removed, write down the value before and after deletion, and state the content rationale. If an outlier is producing an unusual result, check what is an outlier before modifying the data.
The final error is mixing measurement-model results with structural-model results. A construct that meets reliability criteria does not mean that the hypothesis between constructs has been accepted. You still need to read the path coefficient, p-value or bootstrap confidence interval, R², and the other structural-model indicators.
How to write this in your thesis
Use the following structure when reporting a construct: [construct name], [conceptual definition], [item codes], [scale source], [reflective or formative type], and [assessment results]. Keep the measurement-model findings separate from the structural-model findings.
Frequently asked questions
What is a construct in SPSS?
In SPSS, a construct is the theoretical concept behind a group of observed variables. SPSS does not know what your construct is, so you must select the correct variables, such as PU1 through PU4, before running Cronbach's Alpha or EFA.
The construct name may appear in an analysis table because you assigned it while organizing the results. The columns in the data file remain observed-variable codes such as PU1, PU2, and PU3.
What is a construct in SmartPLS?
In SmartPLS, a construct is a latent variable drawn in the model and connected to indicators. Assess construct quality through Outer Loadings, Construct Reliability and Validity, and Discriminant Validity, then examine relationships between constructs through Path Coefficients.
You need to distinguish a construct from an indicator. The construct is the circle in the diagram, while an indicator is usually a rectangular box representing an observed variable.
Is a construct the same as a scale?
A construct is the concept you want to measure, while a scale is the instrument used to measure that concept. For example, “loyalty” is a construct, while LOY1, LOY2, LOY3, and LOY4 form the scale measuring loyalty.
When choosing a scale, check its definition, source, number of items, and fit with the context of your thesis.
Can AMOS draw a formative construct?
AMOS can represent a formative construct with arrows pointing from the observed variables to the construct. A reflective construct usually has arrows pointing from the construct to the observed variables. The important issue is that the arrow direction reflects the theoretical argument, rather than simply how you draw a line in the software.
For a formative construct, examine collinearity and the significance of the weights. Do not apply Cronbach's Alpha mechanically, because formative indicators may represent different aspects of the same concept.
How many observed variables are enough for a construct?
There is no single number that applies to every construct. The number of observed variables should depend on the original scale, content coverage, and research design. The common sample-size rule is 5 to 10 observations per observed variable (Hair et al., 2010), but this concerns data and sample size, not a mandatory number of items for one construct.
Prioritize a theoretically grounded scale, then adapt the wording and test it with your data. Deleting items only to shorten the questionnaire may remove important content from the construct.
With your data file open, your next step is to create a mapping table containing the construct, definition, item codes, scale source, and reflective or formative type. Compare that table with the output before drawing your conclusion. If you need to run the analysis on your own .sav or .csv file, DoThesis M4 data analysis handles this step.