Questionnaire Design: From Research Objectives to SPSS Data

What is questionnaire design in quantitative research

Questionnaire design is the process of turning your research objectives, model, hypotheses, and scales into a form that respondents can read, understand, and answer consistently. A well-designed questionnaire gives you structured data, with each column in the .sav or .csv file corresponding to a variable with a clear meaning.

Start with the research questions and the variables in your model instead of finding a questionnaire template and copying it word for word. For example, if your study tests the effects of perceived usefulness and perceived ease of use on intention to use, the questionnaire needs indicators for each construct, screening questions for the right respondents, and demographic information that supports your planned analysis.

A Likert scale is commonly used to measure agreement with a statement. A 5-point scale can run from 1, strongly disagree, to 5, strongly agree. You can see the specific coding convention in 5-point Likert scale. If your study needs to distinguish more levels of attitude, a 7-point scale is also an option, but use it consistently throughout the questionnaire.

A quantitative questionnaire usually has four sections: an introduction and confidentiality statement, screening questions, groups of indicators, and demographic information for classifying the sample. This order lets respondents determine whether they belong to the target population before they spend time answering the scale items.

How to determine the sample size for a questionnaire

Determine the sample size before distributing the questionnaire, but do not choose a number simply because another thesis used it. Compare the planned analysis method, the number of indicators, the number of independent variables, and the size of the population you can actually reach.

If the population is finite and you know its size, the formula commonly used in Vietnamese theses is:

n = N / (1 + N × e²)

Here, n is the required sample size, N is the population size, and e is the allowable error. This approach is presented in (Yamane, 1967). If you are estimating a population proportion, you can consider the sampling formula in (Cochran, 1977), provided that it fits your research design.

For regression analysis, another rule sets the minimum sample size at 50 + 8m, where m is the number of independent variables, according to (Tabachnick and Fidell, 2013). For EFA, the number of observations per indicator also needs attention. A range of 5 to 10 observations per indicator is stated in (Hair et al., 2010). This is a planning basis, not a guarantee that your data will automatically meet the requirements for KMO or factor loading.

The illustrative output table below summarizes the sample-size situations and the corresponding basis.

The number of responses you need is not the same as the number of survey invitations you must send. Some responses will be removed because respondents fail the screening questions, leave too many items blank, or answer without sufficient attention. If you need 250 valid responses and expect a validity rate of about 80%, you must send the questionnaire to more than 250 people. This reserve rate is an operational plan, not a fixed statistical threshold.

Designing a questionnaire for this research objective

Turning the research model into a questionnaire structure

Open your research model and create a list containing each latent variable, its definition, variable code, and expected number of indicators. For example, the PU construct can have the codes PU1, PU2, and PU3, while PEOU can have PEOU1, PEOU2, and PEOU3. Keep these codes consistent across the questionnaire, Excel file, SPSS, and SmartPLS.

Link each hypothesis to at least one corresponding independent variable, dependent variable, mediator, or moderator. If a variable appears in the model but has no group of indicators in the questionnaire, you will not be able to test that hypothesis with the data you collect.

Writing statements for indicators

An indicator should express one main idea and use wording that someone in your target sample can understand within a few seconds. Avoid a statement with two independent parts, such as “The application is easy to use and saves time,” because a respondent may agree with one part but not the other.

Distinguish an indicator from a question. In a scale, each indicator is usually a statement used to measure a construct. The complete form is the questionnaire. Do not claim that a translation has been validated simply because the wording sounds natural. State clearly that it is a reference version, then adapt it to the Vietnamese context and pilot-check it.

Reverse-coded indicators should be marked in the codebook immediately. When you enter the data into SPSS, recode them before running Cronbach's Alpha or EFA. If you forget this step, a scale can show negative correlations or a low alpha even when its original content appears reasonable.

Choosing question types and scales

Screening questions determine whether a respondent belongs to the target population. For example: “Have you used service X within the past 6 months?” If the answer is no, the form should end or move to a thank-you page. Do not take respondents outside the target population through experience-evaluation items they cannot answer meaningfully.

For attitude, perception, and intention variables, Likert-format items are often appropriate. State both endpoints of the scale and keep the scoring direction consistent. If 1 means strongly disagree, every Likert group in the questionnaire should follow that convention unless you have a clear methodological reason to do otherwise.

Classification questions about gender, age group, occupation, usage frequency, or spending level should include response options that cover the actual sample. Add an “other” option when you cannot list every group. For income and other sensitive information, value ranges usually make it easier for respondents to choose than asking them to enter an exact number.

You can refer to the structure in survey questionnaire and compare it with survey form template. Use these materials to check the structure. Do not take an arbitrary questionnaire and preserve its scale, context, and terminology unchanged.

Checking the questionnaire before official distribution

Before sending it widely, complete the questionnaire yourself on both a phone and a computer. Check required items, page-branching logic, how response options appear, completion time, and whether the data can be exported correctly. A problem such as the gender column exporting as text or a multiple-choice item being combined into one cell will make data cleaning more difficult.

Ask a small group from the actual target population to read through the form and identify technical terms, long statements, and missing response options. This check is for clarifying the measurement instrument and finding display errors. Do not use a small number of pilot responses to claim that the scale has been officially validated.

Collecting the data

Specify where the responses will come from, the collection period, the validity criteria, and whether each person may respond only once. The data-collection channel must fit the population in your model. If your study concerns people who have used an application, reach that group instead of distributing the link indiscriminately.

The introduction should briefly state the academic purpose, completion time, participation conditions, and data-confidentiality arrangements. Do not make the introduction so long that respondents leave before seeing the first item. Measure completion time through a trial run instead of estimating it casually.

DoThesis has fillform.info, our own form tool for creating questionnaires and collecting responses in Vietnamese. Google Forms is a common alternative. Whichever tool you use, check the exported file, column names, response coding, and handling of multiple-choice questions before importing the data into SPSS.

Track valid responses each day instead of looking only at the total number of form opens. If one channel produces many responses but most fail the screening questions, adjust your sampling approach. Record the collection period, distribution channel, and number of excluded responses so you can explain them in Chapters 3 and 4.

Cleaning data before importing it into SPSS

First, create a read-only master copy and a working copy. Do not delete records directly from the original file. Each row should represent one respondent, each column should represent one variable, and the first row should contain short variable names without special characters. Keep columns such as submission time, email address, or contact notes only when they serve the research objective and are handled appropriately.

Check the following four groups of problems:

• Responses that fail the screening questions.
• Questionnaires with too many blank items or that end partway through.
• Multiple submissions from one person, if you have a valid identifier for checking them.
• Straight-line responses across an entire Likert group, unusually fast completion, or contradictory response patterns.

Straight-lining does not automatically prove that a respondent cheated. Review the entire record, completion time, and attention checks before excluding it. Record the exclusion reason in a separate column or log file, such as screenout, incomplete, duplicate, and quality_check. If someone asks, you should be able to report the initial number of responses, the number excluded, and the final number entered into the analysis.

Code responses consistently. For a 5-point Likert scale, you can use numbers from 1 to 5 according to the convention stated in the questionnaire. For categorical questions, set value labels in SPSS so you can see the labels while retaining the numeric codes. Open-ended questions usually do not fit an automated quantitative workflow, so keep them only when they serve a descriptive or information-checking purpose.

After cleaning, check missing values, frequencies, minimum values, and maximum values. If a Likert variable should contain only 1, 2, 3, 4, and 5 but the file contains 0 or 99, determine whether that value is a missing-data code or an entry error. Import the .sav or .csv file for analysis only after the codebook is clear.

How to write this in your thesis

In the methods chapter, present the process in an order that another reader can reproduce: defining the variables and scales, developing the questionnaire, pilot-checking it, collecting the data, specifying the sampling criteria, and cleaning the data. Report the scale sources, initial number of indicators, Likert coding, and collection period.

You can adapt the following paragraph directly: “The questionnaire was developed based on a research model comprising [number] constructs, with [number] indicators adapted from [scale source]. The statements were adjusted to fit the context of [population/market], using a [5/7]-point Likert scale ranging from [lowest level] to [highest level]. After excluding [number] invalid responses according to [criterion], [number] observations were used for the analysis.”

In the research-sample section, state the target sample size, access method, number of questionnaires returned, number of valid responses, and reasons for exclusion. If you use the rule of 5 to 10 observations per indicator, cite (Hair et al., 2010). If you use the finite-population formula, include the formula and (Yamane, 1967) instead of writing only “according to the commonly used formula.”

You can also link the methods section to the survey form so readers can see the instrument structure, but the questionnaire in the appendix must be the actual version you distributed. Do not put a reference template in the thesis and describe it as the instrument used to collect your data.

Common mistakes

The first mistake is starting with a general question without knowing which variable it measures. Fix this by creating a mapping table connecting each hypothesis, construct, variable code, and scale source before writing the statements.

The second mistake is using too many demographic questions while omitting indicators for the dependent variable. Demographic information has value when it supports sample description, group comparisons, or controls included in the analysis plan.

The third mistake is mixing scoring directions, changing “agreement” to “satisfaction” between groups, or using several terms for the same construct. Respondents will interpret the items inconsistently, and you are more likely to enter the codebook incorrectly.

The fourth mistake is reaching the target number without checking response quality. A file with 400 rows can still be unusable if most respondents are outside the target population, leave many items blank, or answer in a straight line. Sample size must be considered together with validity criteria.

The final mistake is deleting responses without saving the reason. When your supervisor asks why 420 responses became 318, you need a step-by-step cleaning log rather than an answer based on memory. If the problem comes from questionnaire design, software cannot repair the data you have already collected.

Frequently asked questions

What is questionnaire design, and where should I start?

It is the process of turning a model, hypotheses, and scales into a form with items, response options, and screening logic. Start with the list of variables in your model, then determine which indicators are needed for each variable.

How many questions should a quantitative questionnaire contain?

There is no fixed number for every topic. The number depends on the constructs, the indicators in each scale, and the sample information you need to collect. Keep enough indicators to measure each construct, but remove items that do not support a hypothesis or describe the sample.

Should I use a 5-point or 7-point Likert scale?

Either can work if it fits your population and research design. A 5-point Likert scale is easy to answer and explain, while a 7-point scale provides more levels of distinction. State both endpoints, keep the coding direction consistent, and report the choice in Chapter 3.

What should I do if the questionnaire has missing data?

First, determine whether the respondent skipped the item, the export failed, or the value is a missing-data code. Do not automatically replace a blank with the mean. Set the exclusion criteria in advance, record the number of excluded cases, and apply the same rule to the entire file.

Should I copy a questionnaire template from the internet?

You can use a template to review structure and presentation, but do not copy it word for word without checking the scale source, context, target population, and coding direction. A questionnaire designed for another study may omit variables or use terminology that does not fit your respondents.

Open your research model and data file now, then create a table with the variable code, statement wording, scale type, and cleaning criterion before collecting more responses. If you need a complete scale and questionnaire built around your topic, use DoThesis M3 for scales and questionnaires.