What Is a Research Model and How Do You Build One for a Quantitative Thesis?

What Is a Research Model

You can think of a research model as the map of a quantitative study. It shows which concepts the study examines, how they are related, and which relationships will be tested with questionnaire data. A model usually contains independent variables, dependent variables, mediators, or moderators, together with arrows showing the expected direction of the effects.

When you ask what a research model is, separate the three layers that are often mixed together. A theory explains why a relationship may exist. A research model turns that explanation into specific variables and relationships for your study. A hypothesis is a testable statement about each relationship in the model. You can also read what a theory is, what a model is, and what a hypothesis is to keep these three concepts separate.

A useful model must connect to your research question, measurement scales, and analysis method. If you draw several variables only because they appear in earlier papers, the model will be difficult to defend. A model with fewer variables, where every relationship has a theoretical basis and can be measured with your data, is usually easier to test.

Components of a Research Model

Before drawing the model in Word, PowerPoint, or Visio, create a separate variable list. Abbreviations must remain consistent from your proposal and questionnaire to your .sav or .csv file. Common components of a quantitative model include:

The dependent variable should answer the research objective directly. If your study asks what affects purchase intention, purchase intention is usually the dependent variable. Perceived price, trust, or information quality may be independent variables, depending on your argument.

Use a mediator when you have a reason to explain the mechanism. For example, information quality may increase trust, and trust may then increase purchase intention. Drawing only a direct arrow from information quality to purchase intention is not enough when your research question also concerns the role of trust.

A moderator is interpreted differently. Age, experience, or level of interest may make a relationship stronger for one group and weaker for another. Do not call a variable a moderator simply because you have included it in the model. This role must be tested through an interaction or an appropriate procedure in SPSS or SmartPLS.

The Original Model and Common Extensions

A research model often begins with a theory or original model that has been used in the field. For example, Davis's Technology Acceptance Model focuses on perceived usefulness, perceived ease of use, and technology acceptance (Davis, 1989). Ajzen's Theory of Planned Behavior examines attitude, subjective norm, perceived behavioral control, and intention (Ajzen, 1991). UTAUT expands the explanation of technology acceptance through factors such as performance expectancy, effort expectancy, and social influence (Venkatesh et al., 2003).

When applying an original model, identify what you are retaining and what you are changing. Retaining the model means keeping its core constructs and main relationship logic. An adjustment may involve adding a variable that fits the context, removing a variable for which you have no data, or changing the dependent variable to match the research question.

For example, a study of a banking application might begin with UTAUT and then add trust because the context involves financial transactions. However, every added variable needs an argument and a measurement source. A diagram with five additional variables but no explanation will usually prompt the committee to ask, “Where did this variable come from, and why does this arrow exist?”

You can review the Research Model topic to compare different modelling directions, but when you finalise your study, return to your own population, context, and available data.

A safe rule is to extend the model only when a new variable meets three conditions: it matters to the research question, it has support in verified literature, and it has a suitable measurement scale. Without the third condition, the model may look good in the diagram but fail to produce reliable data.

Measurement Scales Commonly Used for Each Construct

The model tells you which concepts need to be measured. The scale tells you which items will measure each concept. Do not take a few questions from the Internet and attach them to a latent variable. Find the original scale source, read the original context, and adjust the wording for your respondents in Vietnam.

The table above is a starting point, not a validated questionnaire for every context. When translating items, preserve the meaning of the construct, avoid overly technical wording, and check whether respondents understand each item in the same way. Items should be written as single-meaning statements, with each statement focused on one subject.

After you have the questionnaire, a quantitative workflow usually moves from data checking to Cronbach's Alpha, EFA if the study uses EFA, and then regression or PLS-SEM. Cronbach's Alpha at 0.7 or above is generally considered acceptable (Nunnally, 1978). A Corrected Item-Total Correlation of 0.3 or above is a commonly used threshold (Nunnally and Bernstein, 1994). For a new or exploratory scale, Alpha from 0.6 may be considered depending on the context (Hair et al., 2010).

If you use SmartPLS, separate the measurement model from the structural model. The measurement model checks outer loading, reliability, and convergent validity, while the structural model checks the path coefficient, statistical significance, and explanatory power. Do not place CFI, TLI, or RMSEA from CB-SEM in the assessment of PLS-SEM.

Applying the Model to Your Study

Suppose you are studying “factors affecting students' intention to use an online learning application.” You could build an illustrative model with three independent variables, one mediator, and one dependent variable:

Perceived usefulness  ─┐
Perceived ease of use ─┼──> Trust ───> Usage intention
Information quality ─┘          └──────> Usage intention

This is an illustrative diagram, not the result of an actual study. Before placing it in your thesis, replace the variable names, find the measurement sources, and determine which relationships actually fit your study.

From this diagram, you could write the following hypotheses:

• H1: Perceived usefulness has a positive effect on students' trust in the online learning application.
• H2: Perceived ease of use has a positive effect on students' trust in the online learning application.
• H3: Information quality has a positive effect on students' trust in the online learning application.
• H4: Trust has a positive effect on intention to use the online learning application.

If you want to test a direct effect, you can add a hypothesis for the path from perceived usefulness to usage intention. If you want to test mediation, assess the indirect effect of each independent variable through trust. Decide this before running the model. Do not add hypotheses after looking at the p-value.

In your data file, you could use codes such as PU1, PU2, PU3 for perceived usefulness, PEOU1, PEOU2, PEOU3 for perceived ease of use, TR1, TR2, TR3 for trust, and BI1, BI2, BI3 for usage intention. Each column should represent one item, and each row should represent one respondent. Demographic items should be coded separately, with their values documented clearly in the file.

How to write this in your thesis

In Chapter 3, you can use the following paragraph and replace the bracketed sections: “Based on [name of the theory or original model], this study proposes a model consisting of [number] independent variables, [number] mediators, and [name of the dependent variable]. The hypotheses are developed to test the effect of [variable X] on [variable Y] in the context of [study population and research location].”

In Chapter 4, draw conclusions only from the actual results. A model sentence is: “The analysis shows that the effect of [independent variable] on [dependent variable] has a coefficient of [β] and a p-value of [value]. Therefore, hypothesis [H1] is [accepted or not accepted] at the [significance level] significance level.” Do not place illustrative numbers in your results chapter.

How to Test the Model with SPSS or SmartPLS

SPSS is suitable when your model uses composite variables for regression, tests mediation with PROCESS, or includes EFA. Check missing data, recode reverse-coded items, review descriptive statistics, and assess reliability before creating composite variables. For regression, check multicollinearity, residuals, and the relevant assumptions before interpreting the coefficients.

SmartPLS is suitable when the study includes several latent variables, mediation, or a complex structural model. The .csv file should have variable names in the first row, observations in the following rows, and no duplicate column names. In SmartPLS 4, create a project, import the data, drag latent variables onto the canvas, connect the constructs according to the hypotheses, and then run PLS Algorithm.

At the measurement model assessment stage, an outer loading from 0.7 or above is a commonly referenced threshold (Chin, 1998). CR from 0.7 and AVE from 0.5 or above are used to assess reliability and convergent validity (Fornell and Larcker, 1981). HTMT below 0.85, or below 0.90 for closely related constructs, is a reference threshold for discriminant validity (Henseler et al., 2015).

At the structural model assessment stage, read VIF, path coefficient, t-value, p-value, R², f², and Q² according to the research design. VIF below 5 is a commonly reported threshold in PLS-SEM (Hair et al., 2019). R² values of 0.75, 0.50, and 0.25 can be interpreted as substantial, moderate, and weak, respectively, following (Hair et al., 2011). Q² greater than 0 indicates predictive relevance under this criterion (Stone, 1974).

When running bootstrapping, report the number of subsamples and the confidence interval. Current guidance commonly uses 5,000 bootstrap subsamples (Hair et al., 2022). When testing an indirect effect, a bootstrap confidence interval that does not contain 0 is an important condition for concluding that the effect is significant (Preacher and Hayes, 2008).

SPSS, SmartPLS, AMOS, JASP, and R have different strengths. AMOS is more suitable for CB-SEM and model fit indices, while JASP and R are useful when you want to control the workflow through a graphical interface or code. Paying for an SPSS analysis service may save operating time, but it increases the risk that you cannot explain the output file at your defence. Running the analysis yourself in SPSS or SmartPLS helps you understand each step, but you need to save data versions, result tables, and a log of item removals.

Common Errors When Using This Model

The first error is drawing the model before defining the research question. Start with the problem you need to explain, then choose the theory and variables. If you work in the opposite order, you may create a diagram with many arrows without knowing which relationship is central.

The second error is using inconsistent variable names. In Chapter 2, you write “perceived quality,” in Chapter 3 you change it to “service quality,” and the SPSS file uses “QUAL.” Readers will not know whether these refer to one construct or three different constructs. Create a variable codebook before designing the questionnaire.

The third error is adding variables after looking at the results. If a variable has a large p-value, do not arbitrarily change the direction of the hypothesis or delete the variable just to make the model look better. You can discuss an unsupported result, check for data errors, and report the study's limitations clearly.

The fourth error is taking a scale from an unclear source. A questionnaire copied from a sample paper may not fit your population. Record the source, number of items, translation method, and adjustments in the methods chapter.

The fifth error is mixing criteria between SPSS and SmartPLS. Cronbach's Alpha, EFA, and regression are interpreted differently from outer loading, HTMT, R², and bootstrapping. Identify the tool and method family before choosing assessment thresholds.

Frequently asked questions

What is a research model, and is drawing one mandatory?

A research model is a diagram or structure that describes the variables and relationships to be tested. For a quantitative thesis, you should present the model as a diagram because it lets readers see the scope of the study, the direction of effects, and the position of each hypothesis.

You can draw it in Word, PowerPoint, Visio, or another diagram tool. The tool does not determine the quality of the model. The theoretical logic, measurement approach, and ability to test the relationships are what the committee usually asks about.

How are a research model and a theory different?

A theory is a broader system for explaining a phenomenon. A research model is the way you select part of a theory and apply it to the population, context, and specific variables in your study. One theory can lead to several different models depending on the research question.

Should you add as many variables as possible to the model?

Add a variable only when you have a theoretical basis, a suitable scale, and data that can be collected. The more variables a model contains, the more sample size, questionnaire design time, and explanatory work it requires. A concise model with clear reasoning is usually easier to defend than a complex model with weak support.

Do you need to draw the research model in Visio?

Visio is only a presentation tool. You can use Word or PowerPoint if the diagram is clear, the variable names are readable, and the arrows do not overlap. Save an editable version so you can change a variable or arrow direction later without redrawing the entire model.

Does the research model need to match the original model completely?

Not necessarily. You can retain the core components and adjust the model to fit the context, but every change needs support from the literature, research question, and measurement approach. If you remove or add a variable, explain the inherited and extended parts in the theoretical background chapter.

Open your proposal file and data file, create a variable list with the variable code, scale source, role in the model, and corresponding hypothesis, and then compare it with the diagram before running the analysis. If you need to build the model, scales, and questionnaire for your own study, you can use M3 for the model, scales, and questionnaire.