Designing a research questionnaire

Where should questionnaire design start?

Not with the questions. It starts with your objectives and conceptual model. Every item in the questionnaire should exist because it measures a construct you need, or a background variable you will actually use in analysis.

1. 1

   Map constructs to objectives

   List every construct in your conceptual model and the objective or hypothesis it serves.

2. 2

   Find validated scales

   For each construct, find a published scale with reported reliability, and check its licence.

3. 3

   Adapt and translate

   Adjust wording to your context and translate with a documented forward and back-translation process.

4. 4

   Expert review

   Have a panel judge relevance and clarity; revise and record the changes.

5. 5

   Pilot

   Test with respondents like your sample; check timing, understanding and reliability.

6. 6

   Finalise

   Fix the version, get ethics approval for it, and do not change it during data collection.

Make a simple table before writing anything: construct, definition, source scale, number of items, objective or hypothesis. It becomes the instrument section of your methodology chapter later, and it stops the questionnaire growing. Scholars often add “interesting” questions that never get analysed. Each one costs respondent attention and lowers data quality for the items that matter.

Length matters. A questionnaire that takes more than about 15–20 minutes to complete sees rising drop-out and careless answering, especially online and among busy respondents such as nurses, bank staff or farmers during harvest. Time the pilot and cut if needed.

How do you adapt a validated scale?

A validated scale has been developed and tested in published research, with reported reliability and validity. Using one means your constructs are measured the way the literature measures them, and your results can be compared.

Finding scales

Look at the method sections of the most-cited papers on each construct, and at recent papers in good journals in your field. The appendix or measures section usually lists every item. Prefer scales that have been used in several studies, ideally including one in India or a similar setting.

Check the licence

Many scales are free for research use. Some are commercial: the Multifactor Leadership Questionnaire and the Maslach Burnout Inventory, for example, are sold through Mind Garden. Using a licensed scale without permission is a copyright problem and can surface when the thesis is published on Shodhganga. Check before you commit to a scale.

What you can change

• Context words, such as “my company” to “my school”, are normally fine.
• Dropping items is possible but should be justified, and the reduced scale’s reliability must be checked.
• Changing the response scale, for example from seven to five points, is common but changes comparability; say you did it and why.
• Rewriting items substantially means you are effectively using a new scale, and you will need stronger validity evidence.

Record every change in a table in the appendix: original item, adapted item, reason. Examiners appreciate it, and it answers the question before it is asked.

How do you write good items and choose a Likert scale?

If no suitable scale exists for a construct, you will write items yourself. Keep each item short, about one thing, and in the words your respondents use.

Five points or seven?

Both are widely used and both are defensible. Five points are easier for respondents with less formal education or answering on a phone. Seven points give slightly more spread. The strongest reason for either is consistency with the original scale. Whatever you choose, label every point (strongly disagree, disagree, neither agree nor disagree, agree, strongly agree), not just the ends, especially when the questionnaire is translated.

Should there be a neutral midpoint?

Usually yes. Removing it forces people who genuinely have no view to pick a side, which adds noise. If many respondents may not know about the topic at all, add a separate “don’t know” or “not applicable” option rather than letting them hide in the midpoint.

Reverse-worded items

Many older scales mix positively and negatively worded items to catch careless answering. They often cause trouble: respondents misread them, especially in translation, and they can form a separate factor in analysis. If your scale has them, keep them but check them in the pilot, and remember to recode them before analysis. How Likert data is then analysed is covered in the statistical test guide.

Demographic questions

Ask only what you will use. Put them at the end, where they do not affect how people answer the main scales. Use ranges for income and age if exact values feel intrusive, and use categories that fit India: languages actually spoken, rural, semi-urban and urban locations, and social category only if the study needs it.

How do you translate a questionnaire into an Indian language?

Many PhD surveys in India are administered in Tamil, Hindi, Telugu, Kannada, Malayalam, Marathi, Bengali or another regional language, while the scales were written in English. A careless translation is one of the main reasons a well-known scale produces a low alpha.

The usual process, based on Brislin’s back-translation method and later guidelines such as Beaton and colleagues (2000), is:

1. Forward translation by two people fluent in both languages, working separately. At least one should know the subject; one should not, so that everyday language is kept.
2. Reconciliation of the two versions into one.
3. Back-translation into English by someone who has not seen the original.
4. Expert comparison of the original and back-translated versions, looking for changed meaning, not changed words.
5. Pre-testing with a few respondents, asking them what they understood by each item.

Watch for concepts that do not translate directly. “Work engagement” or “organisational citizenship behaviour” may have no everyday equivalent. It is better to describe the idea in simple words than to use a literal but unfamiliar term. Report who translated, their qualifications in general terms, and what was changed.

How do you pilot a questionnaire?

A pilot tests the questionnaire on people similar to your main sample but not part of it. It answers practical questions: how long it takes, which items confuse people, whether the online form works on a phone, and whether each scale has acceptable reliability.

• Size. There is no single rule. Around 30 respondents is common in theses for a first reliability check; Hertzog (2008) suggested 10 to 40 per group for pilot studies depending on the purpose. Your committee may have a preference.
• Talk to some respondents. Ask five or six people to think aloud as they answer, or to explain afterwards what certain items meant to them. This finds problems that numbers cannot.
• Check reliability and item statistics. Cronbach’s alpha per scale, item-total correlations, and items where nearly everyone gives the same answer.
• Report what changed. A pilot that led to no changes is either unusual or not reported fully.

Pilot respondents should not be included in the main sample, especially if the questionnaire changed afterwards.

How do you reduce common method bias?

When the same person answers every question at the same time on the same kind of scale, some of the correlation between constructs can come from the method itself: a general tendency to agree, a wish to appear consistent, or a mood on that day. This is common method bias, and reviewers in management and social science journals now ask about it routinely. The standard reference is Podsakoff, MacKenzie, Lee and Podsakoff (2003).

Design choices do more than statistics after the fact.

• Separate the measures. Put predictor and outcome scales in different sections, with different instructions or response formats where the scales allow.
• Protect anonymity. Say clearly that there are no right answers and responses are anonymous. People answer more honestly and less consistently with what they think you want.
• Avoid identical anchors everywhere. Thirty items in a row with the same five-point agreement scale invites straight-lining.
• Use a second source where possible. Supervisor ratings of performance, or records such as sales or attendance, instead of self-report for the outcome.
• Separate in time if you can. Collecting predictors and outcomes a few weeks apart reduces shared-method effects, though it adds attrition.
• Add a marker variable. A short scale theoretically unrelated to your model lets you test for method variance statistically later.

Many theses report only Harman’s single-factor test (no single factor explaining more than 50% of variance). Podsakoff and colleagues pointed out that this test is weak and rarely detects a problem. If you use PLS-SEM, the full collinearity test proposed by Kock (2015), with all VIFs at or below 3.3, is commonly reported as well. Report what you did at the design stage first, then the tests.

Ethics and consent

Get approval from your institutional ethics committee before collecting data, for the final version of the questionnaire. Start with a short information sheet and consent statement: who you are, the purpose, that participation is voluntary, how data is stored, and how to contact you. For online forms, a required consent question before the first item is the usual approach. Do not collect names or phone numbers unless you need them, and store them separately from the answers.