What Is the TPB Model? How to Apply and Test It in a Thesis
What Is the TPB Model
You may encounter the TPB model when studying purchase intention, intention to use an application, entrepreneurial intention, green consumer behavior, or any behavior that can be measured with a questionnaire. TPB stands for Theory of Planned Behavior, also written as Theory of Planned Behaviour in English-language sources. The model explains behavior through two main links: the intention to perform the behavior and perceived behavioral control.
Ajzen developed TPB from the Theory of Reasoned Action by adding the Perceived Behavioral Control variable. The foundational work usually cited is Ajzen (1991). You can see a broader explanation in the article TPB theory, but your thesis still needs to adapt the model to the context, participants, and specific behavior in your study.
The basic logic is that Attitude toward the Behavior, Subjective Norm, and Perceived Behavioral Control affect Behavioral Intention. Behavioral Intention may affect Actual Behavior. Perceived Behavioral Control may also be directly related to Actual Behavior when respondents genuinely have the conditions or resources required to perform the behavior.
You need to distinguish TPB from simply asking whether participants like a product. TPB requires you to define the behavior, target, context, and time period clearly. For example, “intention to buy cosmetics” is broad, while “intention to buy organic cosmetics online within the next three months” gives you a more specific behavior from which to develop indicators.
Components of the Model
TPB has the core constructs below. Keep the English names in your variable table and output so you can match them when running SPSS or SmartPLS.
| Component | Common abbreviation | Meaning in the study |
|---|---|---|
| Attitude toward the Behavior | ATT | The extent to which respondents view the behavior as positive, beneficial, or worth performing |
| Subjective Norm | SN | The pressure or support individuals perceive from family, friends, colleagues, or reference groups |
| Perceived Behavioral Control | PBC | The extent to which individuals feel they have the ability, resources, and conditions to perform the behavior |
| Behavioral Intention | BI | The level of readiness or plan to perform the behavior in the future |
| Actual Behavior | AB | An action that has actually occurred, such as having purchased, used, or registered |
ATT is commonly measured with statements about whether the behavior is useful, reasonable, or worth performing. SN focuses on other people and social norms. PBC concerns perceived control, ease of performance, time, money, or access conditions. BI captures the tendency to act, while AB needs to be measured through frequency, number of times performed, or whether the action has occurred.
If your study examines intention only, the model can end at BI. If you have data on actual behavior, you can include AB in the model. You then need to explain why intention does not always become behavior, especially when respondents face limits involving budget, time, or access.
For a technology study, you can consult the TAM model to compare how Perceived Usefulness and Perceived Ease of Use are measured. In a consumer study, what is consumer value may be an additional variable, but do not add a variable simply because you saw it in another paper.
The Original Model and Common Extensions
The original TPB model contains three predictors of intention: ATT, SN, and PBC. BI predicts AB, while PBC may have a direct effect on AB. This structure fits a research question focused on intention and deliberate behavior.
When extending TPB, you may add variables such as Perceived Risk, Trust, Environmental Concern, Perceived Value, or Past Behavior. The new variable needs a theoretical basis, a suitable scale, and a clear role in the model. A model with too many variables makes the questionnaire longer, increases the risk that respondents select the same response for many indicators, and weakens the explanation of the hypotheses.
For example, a study of intention to use an e-wallet may add Trust to explain transaction risk. A study of green product purchasing behavior may add Environmental Concern. A study of online learning behavior may consider Perceived Usefulness, but you should check its overlap with variables in the self-determination model or a technology model before deciding.
You also need to distinguish a mediator from a moderator. If ATT affects BI and BI affects AB, BI may be tested as a mediator. If age or usage experience changes the strength of the relationship between two variables, it is a moderator. The logic of mediators and moderators is presented in Baron and Kenny (1986), while testing the indirect effect with bootstrap is discussed in Preacher and Hayes (2008).
An extended model is not automatically better than the original model. The committee will usually ask three questions: why is the new variable included, which scale measures it, and does the hypothesis fit the research context? You should be able to answer all three before putting the variable into the questionnaire.
Scales Commonly Used for Each Concept
TPB does not provide one set of indicators that applies to every study. You need to find a scale that fits the behavior, participants, and context. The table below is a map of reference sources, not a questionnaire that has already been translated and validated for your study.
| Concept | What needs to be measured | Suitable reference source |
|---|---|---|
| Attitude toward the Behavior | A positive, useful, worthwhile, or appropriate evaluation of the behavior | (Ajzen, 1991) |
| Subjective Norm | Influence and expectations from people who matter to the respondent | (Ajzen, 1991) |
| Perceived Behavioral Control | Ability, resources, and level of control when performing the behavior | (Ajzen, 1991) |
| Behavioral Intention | Readiness, intention, or plan to perform the behavior | (Ajzen, 1991) |
| Actual Behavior | Behavior that has occurred, frequency, or actual level of use | (Ajzen, 1991) |
| Perceived Usefulness | The extent to which users believe that technology improves work outcomes | (Davis, 1989) |
| Social influence in a technology context | The influence of other people on the decision to use something | (Venkatesh et al., 2003) |
Use a Likert scale with clear instructions, such as 1 for strongly disagree and 5 for strongly agree. Likert introduced the technique for measuring attitudes in Likert (1932). If you use a 7-point scale, keep the same number of points throughout the questionnaire and explain that choice in the methods chapter.
When translating indicators, preserve the target and behavior. An original indicator measuring “using mobile banking” should not become “using financial services” if your research scope is narrower. You also need to check reverse-worded indicators. If you have a reverse-worded indicator, recode it before running Cronbach's Alpha and record the operation in your data-processing log.
In the questionnaire, use “indicator” when referring to an item belonging to a construct. “Question” is convenient during data collection, but in the methods chapter you should state how many indicators measure each construct and identify the source of those indicators.
Applying the Model to Your Study
Suppose you are studying “factors affecting students’ intention to use mobile payment applications in Ho Chi Minh City.” The illustrative model contains ATT, SN, PBC, and BI. This example only demonstrates how to build the model and is not the result of a real study.
The conceptual model can be written as follows: ATT, SN, and PBC are independent variables, while BI is the dependent variable. If you have data on the number of times respondents used the application during the last three months, you can add AB as the next dependent variable. You then need screening questions that distinguish people who have used the application from people who only intend to use it.
Illustrative hypotheses:
- H1: Attitude toward using mobile payment applications has a positive effect on intention to use the application.
- H2: Subjective Norm has a positive effect on intention to use mobile payment applications.
- H3: Perceived Behavioral Control has a positive effect on intention to use mobile payment applications.
- H4: Intention to use mobile payment applications has a positive effect on actual usage behavior.
For a different topic, replace the behavior and target in each hypothesis. For example, if you study what is intended behavior, specify whether “intention” means purchase intention, continuance intention, or intention to recommend. Related concepts should not be used interchangeably in the variable names.
You can adapt the wording in your thesis as follows: “Based on Ajzen’s (1991) TPB, this study proposes that ATT, SN, and PBC affect the BI of [study participants] toward [research behavior]. BI is also assumed to affect [actual behavior, if data are available].”
After finalizing the model, create a variable-coding table, for example ATT1 to ATT4, SN1 to SN3, PBC1 to PBC4, and BI1 to BI3. Variable names in the questionnaire, Excel file, .sav file, and SmartPLS model should be consistent. Changing names casually between steps makes it difficult to trace a variable when you are asked to explain why it was removed.
Testing the Model with SPSS or SmartPLS
SPSS is suitable when you want to check the data, run descriptive statistics, Cronbach's Alpha, EFA, and linear regression. In SPSS, go to Analyze > Scale > Reliability Analysis to check reliability, then use Analyze > Dimension Reduction > Factor to run EFA. If your TPB model uses composite variables after EFA, run the regression through Analyze > Regression > Linear.
SmartPLS is suitable when constructs are measured with multiple indicators and you want to assess the measurement model and structural model at the same time. The .csv file needs the first row to contain variable names, with each following row representing one respondent. In SmartPLS 4, create a project, import the data, draw the ATT, SN, PBC, BI, and AB constructs, and connect the arrows according to the hypotheses.
The procedure should follow two stages. The first stage checks the measurement model, including Outer Loadings, Cronbach's Alpha, rho_A, Composite Reliability, AVE, and HTMT. The second stage checks the structural model, including VIF, Path Coefficients, R², f², Q², and bootstrapping results. Assessing the measurement model before the structural model is consistent with Anderson and Gerbing (1988).
The following is illustrative output. The thresholds below are reference points, not reasons to delete variables mechanically. Review the indicator content, theoretical fit, and how the model changes after each deletion.
| Indicator | Common benchmark | Source |
|---|---|---|
| Cronbach's Alpha | 0.70 or above | (Nunnally, 1978) |
| Corrected Item-Total Correlation | 0.30 or above | (Nunnally and Bernstein, 1994) |
| Outer Loading | 0.70 or above in PLS-SEM | (Chin, 1998) |
| Composite Reliability | 0.70 or above | (Fornell and Larcker, 1981) |
| AVE | 0.50 or above | (Fornell and Larcker, 1981) |
| HTMT | Below 0.85, or below 0.90 for closely related concepts | (Henseler et al., 2015) |
| VIF | Below 5 | (Hair et al., 2019) |
| R² | 0.25 weak, 0.50 moderate, 0.75 substantial | (Hair et al., 2011) |
| Q² | Greater than 0 | (Stone, 1974) |
| Bootstrap in PLS-SEM | 5,000 subsamples | (Hair et al., 2022) |
When running SmartPLS, do not include CFI, TLI, or RMSEA in the PLS-SEM assessment. These indices belong to CB-SEM, with benchmarks discussed in Hu and Bentler (1999). If you use AMOS for CB-SEM, describe a procedure different from SmartPLS and do not mix the two sets of criteria in the same results table.
With SPSS, you can run EFA before regression. A KMO of 0.50 or above and a statistically significant Bartlett's Test below 0.05 are commonly used to assess whether the data are suitable for factor analysis, according to Kaiser (1974). An eigenvalue greater than 1 is a factor-extraction rule commonly cited from Kaiser (1960), while total variance explained of 50% or above is referenced according to Hair et al. (2010).
Common Mistakes When Using This Model
The first mistake is using the word “behavior” for a variable that only measures intention. If the questionnaire asks “I will buy” or “I intend to use,” that variable measures BI. To measure AB, ask about an action that has occurred, its frequency, or its level of use during a specific period.
The second mistake is adding too many additional variables without clear hypotheses. Each arrow needs to answer which variable affects which variable, through what mechanism, and based on which source. A model with many variables but weak reasoning is usually harder to defend than a concise model.
The third mistake is taking an entire scale from a foreign study and translating it word for word. The context, age group, product, and behavior may differ. Read each indicator again, pilot it with a small group that fits your target participants, and record the adjustments before official data collection.
The fourth mistake is deleting an indicator only because it increases Alpha. Cronbach's Alpha of 0.70 or above is generally accepted according to Nunnally (1978), while exploratory research may refer to a level from 0.60 according to Hair et al. (2010). Even so, the decision to retain or remove an indicator must also consider the Corrected Item-Total Correlation, theoretical content, and factor structure.
The final mistake is looking at the p-value and concluding that all of TPB is right or wrong. Report the effect coefficient, standard error, confidence interval if available, explained variance through R², and practical significance. A statistically significant hypothesis with a small coefficient still needs careful interpretation.
Frequently asked questions
How many variables does the TPB model have?
Core TPB includes ATT, SN, PBC, BI, and may include AB. The first three variables commonly predict BI, while BI predicts AB. A study focused only on intention may not need to include AB if you do not have actual behavior data.
How are the TPB model and the TRA model different?
TRA focuses on Attitude, Subjective Norm, and Behavioral Intention. TPB adds Perceived Behavioral Control to explain situations in which a person has an intention but is limited by resources or the conditions required to act.
Should you run a TPB model in SPSS or SmartPLS?
SPSS is suitable when you want to check reliability, run EFA, and perform regression using composite variables. SmartPLS is suitable when you want to assess latent variables, outer loading, AVE, HTMT, path coefficient, and bootstrapping together. The choice should depend on your research design, model type, and your supervisor’s requirements.
Is actual behavior required in a TPB model?
It is not required. If your research objective is intention, you can build a model that ends at BI. If you include AB, the questionnaire needs items measuring behavior that has already occurred. Avoid using future-intention statements and calling them actual behavior.
What should you do if the TPB model fails when you run EFA?
Check the coding of reverse-worded indicators, missing data, value entry, and the content of the indicators. Then review KMO, Bartlett's Test, communalities, and the Rotated Component Matrix. If an indicator needs to be removed, record the reason at each stage and make sure the decision remains theoretically appropriate instead of deleting items in bulk just to obtain a cleaner output.
Open your data file, create the ATT, SN, PBC, BI, and AB coding table if applicable, and compare each indicator with its hypothesis before running the tests. If you need to run the analysis on your own .sav or .csv file and track every processing step, you can use DoThesis M4 data analysis.