What Is the TRA Model and How to Apply It in a Quantitative Thesis

What Is the TRA Model

The TRA model, short for Theory of Reasoned Action, explains an individual's behavioral intention through two main groups of factors: attitude toward the behavior and subjective norm. In this model, behavioral intention is usually treated as the construct closest to actual behavior. The core relationship can be presented as a sequence: beliefs influence attitude and subjective norm, attitude and subjective norm influence behavioral intention, and behavioral intention then influences actual behavior.

TRA was developed by Fishbein and Ajzen through research on attitudes, beliefs, and behavior. Among the sources allowed for citation in this article, the foundational work most directly related and commonly used to develop behavioral models is Ajzen's TPB theory. When writing your thesis, check the original source required by your supervisor before entering the publication year for TRA. The illustrative model below focuses on the structure and testing process, so it does not assign an unverified publication year.

You can also read TPB theory to understand the model extended from TRA. TPB adds perceived behavioral control to explain situations in which a person has an intention but still faces constraints when carrying out the behavior.

From a quantitative perspective, TRA fits research questions such as whether customer attitude increases purchase intention, whether the influence of important people increases intention to use, or whether intention is related to actual behavior. Each construct needs to be measured with multiple items in the questionnaire. You then assess reliability, scale validity, and the paths in the model.

Components of the Model

When you open your data file, distinguish clearly between latent variables and observed variables. For example, Attitude is a latent variable, while ATT1, ATT2, and ATT3 are observed variables used to measure it. Consistent coding helps you avoid mistakes when running Cronbach's Alpha, EFA, or building the model in SmartPLS.

In many theses, behavioral beliefs and normative beliefs are measured indirectly through composite variables. Some studies include only Attitude, Subjective Norm, and Behavioral Intention in the structural model to keep the questionnaire shorter. This choice needs to fit the research question and the scale source. Do not add variables simply because the software allows you to run more paths.

Attitude usually expresses the respondent's positive or negative evaluation. Subjective norm reflects perceived social influence, such as the opinions of family, friends, or colleagues. Behavioral intention expresses the intention, plan, or likelihood of performing the behavior within a specific period.

If your study examines a technology-use decision, you can compare TRA with the TPB model, the self-determination model, or technology acceptance models. Base the comparison on the variables your study actually measures, rather than on the model name alone.

The Original Model and Common Extensions

The TRA model has a fairly concise logic, but actual behavior can be limited by money, time, skills, access, or external conditions. A person with a positive attitude and strong intention may still fail to perform the behavior if they lack the ability or opportunity to do so. This is why many studies extend the model by adding other explanatory variables.

The best-known extension is Ajzen's TPB. TPB adds Perceived Behavioral Control, often presented in Vietnamese as perceived behavioral control. According to Ajzen (1991), intention is influenced by attitude, subjective norm, and perceived behavioral control. Perceived behavioral control can also be directly related to behavior when the individual has enough control over carrying out the action.

You can read more about what behavioral intention is if you are unsure whether to use an intention variable or a behavior variable. These two variables should not replace each other in the questionnaire. Intention concerns a plan or readiness, while behavior concerns what the respondent has already done.

Another extension is to add contextual factors to TRA, such as perceived value, perceived risk, convenience, or trust. If you use these variables, explain why they add to TRA's mechanism and find suitable scales. For example, perceived value in consumer behavior can be distinguished from attitude. You can read what consumer value is before deciding whether to place this variable in your model.

When extending TRA, keep the model proportionate to your sample size and the time available for data processing. A model with too many latent variables makes the questionnaire longer, increases the risk that respondents select the same response for many items, and makes the hypotheses harder to explain coherently. Every added variable needs a definition, a scale source, a theoretical reason, and its own hypothesis.

Scales Commonly Used for Each Construct

The table below is a reference map for finding and adapting scales. It is not a validated translation for every Vietnamese context. Compare it with the original article, adapt the wording to your respondents, pilot the questionnaire, and document the adaptation process in your methodology chapter.

For each latent variable, use observed variables that cover the same conceptual range. For example, ATT1 asks whether the behavior is good or bad, while ATT2 asks how useful the behavior is. You need to check whether both items really measure Attitude or whether the second item has moved into Perceived Value.

A 5-point or 7-point Likert scale is commonly used in quantitative questionnaires. Likert (1932) is the foundational source for measuring attitudes through response levels. In an English questionnaire, use one consistent instruction, state clearly what 1 and 5 or 1 and 7 represent, and avoid double negatives.

If you have a reverse-coded item, recode it before assessing reliability. A statement such as I usually do not intend to use this product can confuse respondents who read quickly. When entering the data in SPSS, check the new column after reversing the coding. Do not overwrite the original data before saving a copy.

Applying the Model to Your Study

Suppose your topic is the factors influencing students' intention to use e-wallets. You could build an illustrative model consisting of Attitude, Subjective Norm, Perceived Behavioral Control, and Behavioral Intention. This is an illustrative model for practicing the writing process, not the result of a study that has already run its data.

The conceptual model can be presented as follows:

Attitude, Subjective Norm, and Perceived Behavioral Control → Behavioral Intention

The hypotheses could be written as follows:

• H1: Attitude toward using e-wallets has a positive effect on the intention to use e-wallets.
• H2: Subjective norm has a positive effect on the intention to use e-wallets.
• H3: Perceived behavioral control has a positive effect on the intention to use e-wallets.
• H4: The intention to use e-wallets has a positive effect on actual e-wallet use.

Each hypothesis needs to connect three parts: the theoretical argument, the previous research you have read, and the relationship between the two variables in the model. Do not write H1 simply because two variables sound related. If you have no reason for the direction of the effect, frame the hypothesis consistently with the theoretical basis or state clearly that the relationship needs to be tested.

In the questionnaire, you can place the variable groups in the order ATT, SN, PBC, INT, and BEH, but mix the item order when necessary to reduce the chance that respondents guess the research objective. Even when the item order is mixed, the variable codes in the data file must remain clear. One row of data represents one respondent, and one column represents one observed variable.

When writing Chapter 3, describe the operational definition of each variable. For example, Behavioral Intention is measured by the extent to which respondents intend to continue using an e-wallet in the near future. The period must be specific when the behavior can change quickly, such as one month or three months.

You can adapt the following wording for your thesis: The proposed research model includes [number] independent variables, namely [variable names], a dependent variable, namely [variable name], and [mediator or control variable, if any]. Based on [theory name or source], the study proposes [number] hypotheses concerning the effects of the variables in the model.

Once you have data, do not jump straight from the correlation table to a conclusion about the hypotheses. Check the scales first, then run the appropriate model and report the path coefficient, statistical significance, confidence interval if available, and R² or the fit index relevant to the method you selected.

How to Test the Model with SPSS or SmartPLS

SPSS is suitable when your model uses composite scales, linear regression, mediation testing with PROCESS, or data checks before the main analysis. You can run Cronbach's Alpha at Analyze > Scale > Reliability Analysis, check EFA at Analyze > Dimension Reduction > Factor, then create representative variables and run regression at Analyze > Regression > Linear.

In SPSS, the process usually includes checking missing data, reverse-coded items, outliers, and data distribution as needed. Next, assess Reliability Statistics and Item-Total Statistics. A Corrected Item-Total Correlation of 0.3 or above is commonly used as a reference according to (Nunnally and Bernstein, 1994). Cronbach's Alpha of 0.7 or above is commonly considered acceptable according to (Nunnally, 1978), while an exploratory scale may be considered at 0.6 or above according to (Hair et al., 2010).

If you run EFA, read KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix. A KMO of 0.5 or above, a statistically significant Bartlett test with a p-value below 0.05, and a factor loading of 0.5 or above are commonly cited thresholds from (Kaiser, 1974), (Kaiser, 1960), and (Hair et al., 2010). Total variance explained of 50% or above is also commonly used according to (Hair et al., 2010). You still need to consider theoretical meaning. Do not delete an item just because you want a cleaner table.

The following is illustrative output from the software. The sentence is included as an illustrative output label, while the values in your own file must be reported from your actual SPSS or SmartPLS results.

SmartPLS is suitable when your model contains several latent variables, uses PLS-SEM, and needs simultaneous assessment of the measurement model and structural model. In SmartPLS 4, create a project, import the .csv file, drag the constructs onto the canvas, connect indicators to constructs, and run PLS-SEM Algorithm. Variable names in the .csv file should not contain extra characters or duplicate headers.

Under the PLS-SEM procedure, assess the measurement model before interpreting the structural model. This two-step approach is consistent with the guidance in (Anderson and Gerbing, 1988). In the measurement section, read Outer Loadings, Cronbach's Alpha, Composite Reliability, rho_A, AVE, and Discriminant Validity. An outer loading of 0.7 or above can be referenced to (Chin, 1998) or (Hair et al., 2022). CR of 0.7 and AVE of 0.5 or above can be referenced to (Fornell and Larcker, 1981). HTMT below 0.85, or below 0.90 for closely related constructs, is proposed by (Henseler et al., 2015).

For the structural model, run Bootstrapping in SmartPLS 4. You can use 5,000 bootstrap subsamples according to (Hair et al., 2022). Then read Path Coefficients, T Statistics, p-values, and Confidence Intervals. VIF below 5 is a reference point for checking multicollinearity according to (Hair et al., 2019). R² can be interpreted using 0.75, 0.50, and 0.25 as substantial, moderate, and weak levels, respectively, according to (Hair et al., 2011), but you still need to place the result in the context of the field.

If the model has a mediator, read the indirect effect and the bootstrap confidence interval. A confidence interval that does not contain 0 is commonly used to conclude that the indirect effect is significant according to (Preacher and Hayes, 2008). Do not use CFI, TLI, or RMSEA to assess a SmartPLS model. These indices belong to CB-SEM, and their reference thresholds are presented in (Hu and Bentler, 1999).

Common Mistakes When Using This Model

The first mistake is calling every model with an intention variable TRA. TRA has a specific logic involving attitude, subjective norm, and behavioral intention. If you add perceived behavioral control, consider calling it a TPB model or a model extended from TRA, and explain the role of the new variable.

The second mistake is using behavioral intention and actual behavior as two names for the same variable. If the questionnaire only asks I will use it in the near future, that is intention. If it asks How many times have you used it in the past 30 days, that is closer to actual behavior.

The third mistake is taking a scale from one article without reading the original definition. A variable called Attitude in an earlier article may not cover the same range as Attitude in your study. Record the source, context, number of observed variables, response format, and any adaptations you made.

The fourth mistake is running SmartPLS before cleaning the .csv file. Blank columns, text codes in numeric columns, extra header rows, or cells containing unusual symbols can prevent the software from recognizing indicators correctly. Open the file in a spreadsheet, check the column names, missing values, and data types before importing it.

The fifth mistake is continuously deleting items until every index reaches a threshold without recording the reason. The committee may ask why ATT3 was deleted, why two observed variables measure the same content, or why the final model differs from the proposed model. Save each data version and keep a processing log.

The final mistake is presenting the results as conclusive evidence of causation. Cross-sectional survey data usually show the level of association or effect specified by the statistical model, but the interpretation still needs to follow the research design. Report the coefficient, p-value, confidence interval, and sample limitations instead of using claims broader than the data support.

Frequently asked questions

What is the TRA model and what is it used for?

TRA explains behavioral intention through Attitude and Subjective Norm. It is suitable when your study needs to examine how personal evaluation and social influence relate to the intention to perform a behavior.

How are the TRA and TPB models different?

TPB extends TRA by adding Perceived Behavioral Control. If your study includes skills, resources, opportunities, or perceived ease of performing the behavior, TPB may provide a more suitable explanatory structure. You still need to justify that choice with a theoretical basis.

Should I run the TRA model in SPSS or SmartPLS?

SPSS is suitable for regression and scale tests using a composite-variable procedure. SmartPLS is suitable when you want to assess the measurement model and structural model using PLS-SEM. Base the choice on the research design, sample size, scale type, and your supervisor's requirements, rather than only on the software available to you.

How many observed variables are enough for a TRA scale?

There is no single number for every study. Each construct needs enough observed variables to cover its definition and must have a suitable scale source. Sample size also needs to be considered in relation to the number of observed variables and the analysis method. The benchmark of 5 to 10 observations per observed variable is commonly cited according to (Hair et al., 2010), but it is a reference basis, not a substitute for a methodological argument.

What should I do if the Behavioral Intention variable does not reach Cronbach's Alpha?

First check Item-Total Statistics, reverse-coded items, missing data, and incorrect coding. You may consider deleting an observed variable if there is a clear statistical and theoretical reason, then rerun the analysis with the new data version. Do not keep deleting items simply to reach a threshold if doing so removes important content from the construct.

Can I add Perceived Value to the TRA model?

You can, if you can explain the mechanism through which it adds to Attitude or Behavioral Intention, find a suitable scale, and state a separate hypothesis. You also need to check for overlapping content between Perceived Value and Attitude, because closely related constructs can weaken discriminant validity.

Open your data file and begin by redrawing the variables, arrows, and observed-variable codes in the TRA model. Then compare them with the questionnaire before running the tests. If you need to run the analysis on your own .sav or .csv file, see M4 quantitative data analysis.