What Is a Research Model in Quantitative Research and How to Build One

What Is a Research Model

You can think of a research model as a map of your topic. It shows which concepts enter the study, which concepts affect one another, the direction of those effects, and the data you will use to test each relationship.

In quantitative research, a model is usually shown as rectangles or ovals connected by arrows. Each shape represents a variable or concept. An arrow represents a proposed relationship in the model, such as service quality affecting satisfaction or perceived usefulness affecting intention to use.

You need to distinguish a model from a theory. A theory is a broader system for explaining a phenomenon. A model is the way you select and arrange certain concepts from a theory to answer a specific research question. If you want a separate explanation of the theoretical foundation, you can read what is a theory.

For example, Davis's TAM includes perceived usefulness, perceived ease of use, attitude, and intention to use technology. If you are studying digital banking applications, you can retain TAM's core variables, add trust or perceived risk, and then test the relationships with a questionnaire and survey data. TAM was proposed by Davis in 1989, with the original source cited as (Davis, 1989).

A good model must answer four questions: what is being studied, how are the variables defined, how are they related, and which indicators will measure each relationship. If a diagram contains many concepts but is not connected to the research question or hypotheses, it is a list of variables rather than a complete model.

Components of a Research Model

A quantitative model commonly includes independent variables, dependent variables, mediators, moderators, and control variables. Your topic does not need every component. Include a component only when you have a theoretical reason for it and a clear way to measure it.

Independent variables are often placed on the left side of the diagram, while the dependent variable is placed on the right. This arrangement makes the diagram easier to read, but it is not a mandatory rule. The important point is that the arrows must reflect the logic of your topic and the direction you intend to test.

A concept such as satisfaction or trust is often a latent variable. You do not observe it directly. Instead, you measure it through several items in the questionnaire. For example, satisfaction may be measured with statements about overall satisfaction, the extent to which expectations are met, and the decision to continue choosing the service.

You also need to distinguish a conceptual variable from an observed variable. “Service quality” is a concept or latent variable. Codes such as CLDV1, CLDV2, and CLDV3 are observed variables used to measure that concept. Using consistent codes from the questionnaire stage will help you avoid errors when entering the .sav or .csv file.

If you want to review terms that are easy to confuse, see hypothesis or assumption. In a thesis, a hypothesis is a statement about a relationship that needs to be tested, while a model is the visual and logical structure containing those relationships.

Original Models and Common Extensions

An original model is the foundational version proposed in the initial work. An extended model retains part of the existing structure while adding new variables or relationships to fit the research context. You need to identify which variables come from the original model, which ones were added, and why the additions are necessary.

Several models commonly appear in business and behavioral research, including TAM, TPB, UTAUT, UTAUT2, and SERVQUAL. Ajzen's TPB focuses on attitude, subjective norm, perceived behavioral control, and behavioral intention, with the original source cited as (Ajzen, 1991). UTAUT explains technology acceptance through factors such as performance expectancy, effort expectancy, social influence, and facilitating conditions, and is cited as (Venkatesh et al., 2003).

UTAUT2 extends UTAUT to the consumer context by adding factors that better fit individual usage behavior, and is cited as (Venkatesh et al., 2012). SERVQUAL is used to assess service quality through dimensions commonly discussed in service research, with the original source cited as (Parasuraman et al., 1988).

When building an extended model, do not add a variable simply because it appears in many earlier studies. A new variable needs to meet three conditions: it has a theoretical basis, it has a scale that can be located, and it fits your research context. Adding too many variables makes the questionnaire longer, increases the required sample size, and makes the results harder to explain.

Suppose you are studying intention to use an electronic wallet. The original model may be based on TAM, with perceived usefulness and perceived ease of use. The extended model adds trust and perceived risk. Your theoretical foundation must then explain why trust may increase intention while perceived risk may reduce it. Every additional arrow needs a corresponding section of reasoning.

You can also read research model to compare how a model is presented in a specific topic. For industry frameworks such as competitive analysis, Porter's Five Forces has a different structure from the model of relationships between variables used in quantitative surveys.

Common Scales for Each Concept

Each concept in your model needs to be linked to a suitable scale. Start with the original study or with studies that used the scale in a similar context. The scale source tells you how the concept was defined, how many items were used, and how the authors measured it.

The table above is a guide to sources, not a validated translation for every topic. When using a scale, adjust the wording for your respondents, preserve the meaning of the concept, and document how you used the source. Do not translate each statement mechanically and place it directly into the questionnaire.

In a questionnaire, each item is usually measured with a Likert scale. The Likert scale was introduced in Likert's work, cited as (Likert, 1932). Keep the coding consistent, such as 1 for strongly disagree and 5 for strongly agree. Mark reverse-coded items when you design the data file so that you do not enter them incorrectly when running Cronbach's Alpha or EFA.

After collecting data, you usually assess reliability with Cronbach's Alpha and Corrected Item-Total Correlation before considering EFA or the measurement model. Cronbach's Alpha from 0.7 upward is generally considered acceptable according to (Nunnally, 1978). Corrected Item-Total Correlation from 0.3 upward is used in many studies according to (Nunnally and Bernstein, 1994). For a new or exploratory scale, Alpha from 0.6 may be considered according to (Hair et al., 2010), but you need to explain the context rather than automatically retaining or deleting an item.

Applying the Model to Your Topic

Suppose your topic is “Factors affecting students' intention to continue using a digital banking application”. You could build an illustrative model with four independent variables: perceived usefulness, perceived ease of use, trust, and perceived risk. The dependent variable is intention to continue using.

The model now has four arrows pointing toward the dependent variable. Before drawing the diagram, write the research questions: does perceived usefulness affect intention to continue using, does perceived ease of use affect it, in which direction does trust affect it, and does perceived risk reduce intention.

You can then write hypotheses H1 to H4 as follows:

• H1: Perceived usefulness has a positive effect on intention to continue using the digital banking application.
• H2: Perceived ease of use has a positive effect on intention to continue using the digital banking application.
• H3: Trust has a positive effect on intention to continue using the digital banking application.
• H4: Perceived risk has a negative effect on intention to continue using the digital banking application.

You can read more about what is a hypothesis to check how to name and phrase these relationships. Each hypothesis should come from the reasoning presented earlier. Do not place four hypotheses at the end of the chapter without explaining each arrow.

How to write this in your thesis

You can begin the thesis paragraph as follows: “Based on TAM and previous studies, this study proposes a model comprising perceived usefulness, perceived ease of use, trust, perceived risk, and intention to continue using. The model is tested with survey data from [survey respondents] in [research context], collected from [start time] to [end time].”

When presenting the model, give the figure a clear name, such as “Figure 2.1. Proposed research model”. The text below the figure should explain the symbols, the model source, and the points you adjusted. If you combined the model from several sources, specify which variable was inherited from which source rather than writing only “proposed by the author”.

How to Test the Model with SPSS or SmartPLS

SPSS is suitable when your model uses composite scales and linear regression, such as testing the effect of several independent variables on one dependent variable. The usual process includes cleaning the data, running Cronbach's Alpha, running EFA, creating composite variables, and running regression. In SPSS, the menu paths commonly used are Analyze > Scale > Reliability Analysis, Analyze > Dimension Reduction > Factor, and Analyze > Regression > Linear.

If you use SPSS, check the Reliability Statistics, Item-Total Statistics, KMO and Bartlett's Test, Total Variance Explained, Rotated Component Matrix, and Coefficients tables. KMO from 0.5 upward and a statistically significant Bartlett test below 0.05 are commonly referenced according to (Kaiser, 1974). Factor loading from 0.5 upward and total variance explained from 50% upward can be referenced to (Hair et al., 2010). For regression, the minimum sample size rule of 50 + 8m can be referenced to (Tabachnick and Fidell, 2013), where m is the number of independent variables.

SmartPLS is suitable when your model contains latent variables, multiple items, mediation relationships, or a complex structural model. Prepare the .csv file with each item in one column and each respondent in one row. In SmartPLS 4, create a project, import the data, drag the items into constructs, connect the constructs according to the hypotheses, and then run PLS Algorithm and Bootstrapping.

When reading SmartPLS results, check Outer Loadings, Construct Reliability and Validity, Discriminant Validity, Path Coefficients, R Square, f Square, and VIF. Outer loading from 0.7 upward can be referenced to (Chin, 1998) or (Hair et al., 2022). CR from 0.7 and AVE from 0.5 upward are referenced to (Fornell and Larcker, 1981). HTMT below 0.85, or below 0.90 for closely related concepts, is proposed in (Henseler et al., 2015). VIF below 5 can be referenced to (Hair et al., 2019).

In PLS-SEM, do not include CFI, TLI, or RMSEA in your model assessment as you would in CB-SEM. R² values of 0.75, 0.50, and 0.25 can be interpreted as substantial, moderate, and weak, respectively, according to (Hair et al., 2011). When running bootstrapping, 5,000 bootstrap subsamples can be referenced to (Hair et al., 2022). An indirect effect is considered significant when the bootstrap confidence interval does not contain 0, according to (Preacher and Hayes, 2008).

The illustrative output below shows the form of a table you may see in SmartPLS. This is illustrative output, not the result of an actual study.

When writing the results chapter, replace all illustrative output with the numbers in your own file. Report the path coefficient, t-value or p-value, confidence interval if available, R², and conclusion for each hypothesis. Do not write only “the hypothesis is accepted” without telling the reader which table and which statistic support the conclusion.

Common Errors When Using This Model

The first error is putting too many variables in the diagram because you want the topic to appear complete. Each variable increases the number of items, response time, and testing work. Start with the research question, then select enough variables to answer it.

The second error is naming a theory without using its logic. For example, a topic may claim to be based on TAM while adding many unexplained variables or using a scale unrelated to the original definition. The reader needs to see the connection between theory, model, hypotheses, and scales.

The third error is confusing a mediator with a moderator. A mediator answers the question, “Through which factor does the effect occur?” A moderator answers the question, “When conditions change, does the effect become stronger or weaker?” These two types of variables require different hypothesis development and different testing procedures.

The fourth error is drawing arrows according to the results after running the data. The model and hypotheses should be proposed before testing. If the data do not support a hypothesis, report that result and discuss possible reasons instead of changing the diagram to match the p-value.

The final error is mixing the criteria used by SPSS, SmartPLS, and AMOS. SPSS is commonly used for regression and EFA, SmartPLS uses PLS-SEM, and AMOS belongs to CB-SEM. Each tool has its own approach to model assessment. Choose the tool based on your model, data type, sample size, and supervisor's requirements.

Frequently asked questions

What is a research model in quantitative research?

A model is a diagram and explanatory structure showing the variables in your topic and the relationships between them. It helps you convert the research question and theory into hypotheses that can be tested with survey data.

Does a research model have to be based on a theory?

A quantitative model should have a theoretical foundation or a clear academic rationale. You can combine several theories, but you need to identify which parts were inherited, which parts were adjusted, and why the adjustments were made.

How many variables should a research model have?

There is no fixed number for every topic. The number of variables should be sufficient to answer the research question, have suitable scales, and remain within your ability to collect data. Adding variables without a basis makes the questionnaire longer and the model harder to explain.

What is the difference between a model and a hypothesis?

A model is the overall structure, usually shown through variables and arrows. A hypothesis is a specific statement about one relationship in the model, such as perceived usefulness having a positive effect on intention to use.

Should you use SPSS or SmartPLS to test a model?

SPSS is suitable for Cronbach's Alpha, EFA, and regression using composite variables. SmartPLS is suitable for latent-variable models, multiple items, and mediation relationships or complex structures. You should also check your department's and supervisor's requirements before choosing the software.

Open your proposal and data file, make a list of the variables, scale sources, and each hypothesized arrow before running the tests. If you need to complete the analysis using your .sav or .csv file, you can use module M4 at DoThesis.