Research Design: How to Build a Thesis Model and Plan

What research design means in a thesis

You open your proposal file and find that you have a fairly broad topic, such as online purchase intention, satisfaction with digital banking, or intention to use technology. The problem at this point is not running SPSS. You need to decide which concepts to measure, which items to use, which relationships to test, and from whom to collect data. That entire chain of decisions forms your research design.

In a quantitative thesis, research design is the blueprint connecting the research question to the data and the analysis. A solid design usually has five parts: the research problem and objectives, the conceptual model, hypotheses, the data collection method, and the testing plan. If one link is missing, you can end up with a questionnaire but no clear explanation of which group of items answers which hypothesis.

You can review the general discussion of scientific research methodology before writing your proposal. This article focuses on turning that general discussion into a quantitative design that can be implemented with a questionnaire, an .sav or .csv file, SPSS, and SmartPLS.

Research design is usually presented mainly in Chapter 3, but it must remain consistent with Chapters 1 and 2. The objectives in Chapter 1 should lead to the research questions. The theoretical basis in Chapter 2 should lead to the model and hypotheses. Chapter 3 should explain exactly how you will measure and test those hypotheses.

Standard structure of a research design

There is no fixed length that applies to every university, but a methodology chapter is often around 10 to 20 pages, depending on the department's requirements and the complexity of the model. You can use the table below to allocate an initial amount of space. This is a writing framework, not a mandatory rule that replaces your supervisor's instructions.

This table helps you avoid writing the analysis section before deciding what data you need. If the model contains only observed variables and the aim is to test effects, regression may be appropriate. If the model contains several latent variables, a mediator, or a moderator, you need to consider PLS-SEM and prepare the data in a compatible structure.

You should also distinguish research design from the research process. Design answers the question, “How is the study organised?”, while the process describes the order of the work. When you need to check that order, you can consult the scientific research process and compare it with your university's template.

A good design should answer four questions when the committee asks: Why did you choose these variables, why did you use this scale, why did you choose this sample size, and why does this analysis answer the hypotheses? If each answer has a clear place in the proposal, Chapter 3 becomes easier to write and defend.

Steps for writing the research design

Step 1: Finalise the problem, objectives, and research participants

Write the research problem as a specific gap. Instead of writing “study online shopping behaviour,” you could limit it to “measuring the factors affecting online cosmetic purchase intention among students in Ho Chi Minh City.” This sentence already identifies the participants, behaviour, and context from which you need to collect data.

Next, write one general objective and three to four specific objectives. Specific objectives should use testable verbs such as identify, measure, evaluate, test, and compare. Avoid writing “explore perceptions” if you plan to use a Likert questionnaire and statistical tests, because that wording does not show which variables you will measure.

Step 2: Identify the concepts and variables in the model

Start from the objectives and make a list of the main concepts. Each concept needs a working definition, a role in the model, and a measurement method. An independent variable explains the dependent variable. A mediator describes the mechanism through which an effect is transmitted. A moderator changes the strength or direction of a relationship.

Assign short codes at the beginning, such as PU for Perceived Usefulness, TR for Trust, and BI for Behavioral Intention. Use these codes consistently in the questionnaire, data column names, SPSS output, and results tables. Changing a code between Chapter 3 and the data file seems minor, but it makes cross-checking much more time-consuming.

If the model is based on TAM, TPB, or UTAUT, state clearly which parts remain unchanged and which parts are adjusted. Do not add a variable simply because it appears in many previous studies. Explain why that variable fits your context, participants, and research question.

Step 3: Draw the research model and write the hypotheses

Draw the concepts using rectangles or ovals, then use arrows to show the direction of the effects. Every arrow in the model must have a corresponding hypothesis. If the model has an arrow from Trust to Behavioral Intention, you need an H1 stating that Trust affects Behavioral Intention in the research context.

A quantitative hypothesis should identify the influencing variable, the affected variable, the expected direction, and the context. For example: “H1: Perceived Usefulness has a positive effect on students' intention to use digital banking applications in Ho Chi Minh City.” The reasoning before the hypothesis should be based on theory and previous studies, not only on the writer's intuition.

After writing the hypotheses, create a cross-reference table linking each hypothesis to its variables and test. If a hypothesis has no measurable variable, you cannot test it. If a variable appears in the questionnaire but not in the model, your supervisor may ask why you included it.

Step 4: Choose the scales and design the questionnaire

Each concept should be measured with a group of items. Record the scale source, the initial number of items, the translation or adjustment process, the Likert scale, and the assessment results after data collection. The Likert scale is used to measure attitudes through the technique proposed by Likert (Likert, 1932).

The questionnaire should include an introduction, screening questions, demographic information, and groups of items. Screening questions should confirm that the respondent belongs to the correct target group, such as having used the service within the past six months. Do not collect personally identifying information unless it serves the research objective.

Before distributing the questionnaire, check that each item expresses one clear meaning, uses neutral wording, and can be understood within a few seconds. If you translate a scale from English, read it again in the Vietnamese context and document the adjustment process in Chapter 3. A referenced scale has not automatically been validated for your sample.

Step 5: Define the sample and data collection plan

Describe the target population, respondent criteria, sampling approach, collection period, and expected number of valid responses. For research with many items, the rule of 5 to 10 observations per item is often used as a reference point (Hair et al., 2010). For regression, another reference point is a sample size of at least 50 + 8m, where m is the number of independent variables (Tabachnick and Fidell, 2013).

Do not choose the sample size only because you found a familiar number online. Calculate it based on the actual model, number of items, number of groups to compare if applicable, and your ability to collect data within the remaining time. After collection, remove responses with substantial missing information, unusually repetitive response patterns, or failed screening questions. Record the initial number of questionnaires, the number removed, and the number used for analysis.

Step 6: Choose the analysis plan and reporting criteria

A typical plan follows this order: descriptive statistics, reliability testing, scale validity assessment, and then model and hypothesis testing. With SPSS, you can run Cronbach's Alpha, EFA, correlation, and regression. With SmartPLS, you assess the measurement model first and then the structural model, following the two-step logic presented in (Anderson and Gerbing, 1988).

Record in advance which output tables you will use. Cronbach's Alpha is read from Reliability Statistics, corrected item-total correlation is read from Item-Total Statistics, and EFA usually involves KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix.

When assessing EFA, a KMO of at least 0.5 and a statistically significant Bartlett's test are considered suitable conditions according to (Kaiser, 1974). An eigenvalue above 1 is an extraction rule described in (Kaiser, 1960). A factor loading of at least 0.5 and total variance explained of at least 50% are commonly used reference points according to (Hair et al., 2010).

If you use PLS-SEM, state which outer loadings, CR, AVE, HTMT, VIF, R², f², and Q² values will be reported. CR of at least 0.7 and AVE of at least 0.5 are used to assess composite reliability and convergent validity (Fornell and Larcker, 1981). HTMT is commonly assessed below 0.85, or below 0.90 when the concepts are closely related (Henseler et al., 2015). Do not include CFI, TLI, or RMSEA in the SmartPLS assessment, because those indices belong to the CB-SEM logic described in (Hu and Bentler, 1999).

How to write this in your thesis

You can replace the square-bracketed sections with the actual information from your study. These are sentences to adapt, not verified results from your research.

Complete the sentence about results only after you have the output. Do not create a coefficient, p-value, or confidence interval to fill Chapter 4. When writing, copy the exact variable name and value from the output, then interpret it against the corresponding hypothesis.

A shortened research design example

Suppose you choose the topic “Factors affecting students' intention to use e-wallets in Hanoi.” After reviewing the theoretical basis, you propose three independent variables, Perceived Usefulness, Perceived Ease of Use, and Trust, together with Behavioral Intention as the dependent variable.

The general objective is to assess the effects of the three factors on e-wallet use intention. The three corresponding hypotheses are H1, Perceived Usefulness has a positive effect on Behavioral Intention; H2, Perceived Ease of Use has a positive effect on Behavioral Intention; and H3, Trust has a positive effect on Behavioral Intention.

The questionnaire uses a five-point Likert scale, from 1 for strongly disagree to 5 for strongly agree. Respondents must be students who know about or have used an e-wallet. The data are cleaned, coded, and saved as .sav before SPSS is run.

The analysis plan has four layers. First, descriptive statistics describe the sample. Next, Cronbach's Alpha and corrected item-total correlation assess reliability. EFA then examines the factor structure. Finally, multiple regression tests H1 through H3. If Trust has a positive coefficient and a p-value below the selected significance level, you have statistical grounds to conclude that the relationship follows the hypothesised direction within the research sample.

Every part of the example connects to the others. The screening question produces the correct target group. The variables in the model appear in the questionnaire. Each group of items has a clear code. The final analysis directly answers the arrows in the model. If you add more variables simply to make the model look more complete, the design becomes harder to explain rather than stronger.

You can also compare the methodology presentation in research methodology or review the scientific research methodology textbook PDF to keep your terminology consistent with your department's requirements. If you are still choosing a topic, the graduation thesis topic list can help you check whether the expected data are feasible.

Common problems that make the committee ask follow-up questions

The first problem is a model that does not come from the research question. The writer includes several variables because previous studies often use them, but does not explain their roles. When asked, “Why did you choose this variable?”, answering “because many studies have used it” is not enough if you have not shown that it fits the context.

The second problem is a hypothesis that uses wording that is too broad. The sentence “factors affecting behaviour” does not identify which factors, which behaviour, or the direction of the effect. Separate it into individual hypotheses, each containing one specific relationship.

The third problem is mixing the measurement model with the structural model. Cronbach's Alpha, outer loading, CR, and AVE concern the scale. Path coefficient, p-value, R², and f² concern relationships in the model. Assess the scale before interpreting effects between variables.

The fourth problem is choosing the analysis after looking at the results. Removing an item only because it makes the hypothesis insignificant will raise questions about your methodological basis. If you need to remove an item, record the reason, criterion, analysis round, and effect on the model.

The fifth problem is stating a sample size without describing the actual sample. Chapter 3 says that 300 questionnaires were expected, while Chapter 4 reports only 217 valid responses without explaining where the others went. Create a tracking table showing the number distributed, received, removed, and analysed.

The final problem is confusing sample results with conclusions about the entire population. Survey data reflect the sample you collected. When writing the conclusion, limit it to the participants, location, and period of the study, especially if you used convenience sampling.

Frequently asked questions

Is research design the same as research methodology?

The two are connected without being identical. Research methodology is how you collect and analyse data, such as surveying with a questionnaire, running regression in SPSS, or using PLS-SEM in SmartPLS. Research design is broader because it includes the research question, model, hypotheses, sample, scales, and analysis plan.

In Chapter 3, you can present the design first and then describe the specific methods and implementation process. This helps the reader understand why you selected a method instead of seeing only a list of software procedures.

Do I need to build the research model before creating the questionnaire?

Yes. The model shows which concepts you need to measure and which group of items supports each hypothesis. If you create the questionnaire first, you may collect many variables that you never use or miss variables that are necessary.

You can still adjust the model after reviewing the literature and checking its fit with the context. However, explain the adjustment before the formal analysis. Do not revise the model only to make the results look better.

Should a quantitative research design use SPSS or SmartPLS?

SPSS is suitable for descriptive statistics, Cronbach's Alpha, EFA, correlation, regression, and some group tests. SmartPLS is suitable when the model contains latent variables, several simultaneous relationships, a mediator, or a moderator, and you want to assess the measurement model together with the structural model.

Choose based on the model, scale type, research objective, and your supervisor's requirements. Software does not replace the design. An unclear model remains difficult to explain whether you run it in SPSS or SmartPLS.

What sample size is sufficient for a research design?

There is no single correct number for every study. You need to consider the number of items, number of independent variables, model structure, groups to compare, and sampling method. The rule of 5 to 10 observations per item (Hair et al., 2010) and the regression formula 50 + 8m (Tabachnick and Fidell, 2013) should be used together with an explanation of the actual sample.

In the thesis, state how you calculated the sample size, how many responses you obtained, the criteria for removing questionnaires, and the final number of observations. Supervisors usually care more about consistency among these numbers than about a large number with no clear basis.

If the results do not support the hypothesis, does that mean the design is wrong?

Not necessarily. An unsupported hypothesis may occur because the relationship is genuinely weak in the research context, the sample is not representative, the scale is unsuitable, or the data have limitations. Check the cleaning process, reliability, scale validity, and assumptions of the analysis before interpreting the result.

Do not change the direction of the hypothesis or remove data simply because the p-value does not meet the criterion. Report the result exactly as shown in the output, explain possible reasons based on the context, and state the study's limitations.

When your proposal file is still blank, open a table with columns for the objective, variable, hypothesis, item, scale source, and analysis method. Fill each row with the actual information from your study. If you need to run the analysis on your .sav or .csv file and connect the output to the results chapter, you can use M4 Data Analysis from DoThesis.