Original Research
Part of Original research for digital PR
Designing a research question before collecting data
Turn a broad research topic into an answerable question with a defined population, measure, output and claim boundary.
A useful research question names what you want to learn, the people or records concerned, and the answer the method could produce. Write it before collecting data. Otherwise, an interesting result may tempt the team to redefine the study after seeing the numbers.
Move from a topic to an answerable question
“Customer behaviour” is a topic. “How did recorded requests for a specified service change among this business’s Australian customers across two defined periods?” is a research question. It identifies a measure, a group and a comparison. The business must still check whether its records are consistent enough to answer it.
Define five parts of the question:
- Unit:
- Is each case a person, account, request, transaction or response?
- Population:
- Which units should the result describe?
- Measure:
- What observation would answer the question?
- Period:
- When will it be measured, and are compared periods alike?
- Purpose:
- Is the study describing a pattern, comparing groups or testing an explanation?
The purpose determines what can be claimed. A comparison of two observed periods does not establish the cause of a change unless the design can address other explanations.
Key elements of a strong research question
- Unit
- Individual, account, request, transaction or response
- Population
- Specific group being studied (e.g., Australian customers)
- Measure
- Observable outcome (e.g., frequency of requests)
- Period
- Time frame(s) for measurement
- Purpose
- Describe, compare or test an explanation
Work backwards from the intended output
Sketch the table needed to answer the question, with row labels, units and denominators. If the intended output is a share, identify its eligible base. If it is a change, check whether the groups, time periods and recording rules are comparable. If an essential field is absent or unreliable, revise the question or collection plan.
For a survey, turn the research question into wording participants can understand. “Are customers satisfied?” may combine several experiences. Decide which experience matters and what answer options would capture it. Test the wording before collection: a precisely recorded answer can still measure a different idea from the one intended.
Check whether existing information already answers the question. If it nearly does, name the gap, such as a missing group, a changed definition or a question that was never asked. Do not infer the missing answer from the available data.
Set the boundary before looking at results
Write down a claim the planned method might support and a broader claim it could not support without further evidence.
A voluntary customer survey can describe the answers of participating customers. Its size alone does not make those answers a design-based estimate of all Australian customers’ views. Keep recruitment and likely coverage in mind when choosing words such as “customers”, “respondents” or “Australians”.
Record the primary measure and planned comparisons before analysis. Also set a reason to hold the claim, such as too few usable cases for a planned subgroup or definitions that changed between periods. Unexpected patterns can be investigated, but identify them as exploratory.
The question is ready for collection when another person can identify its population, measure, intended output and claim boundary without guessing. If they cannot, sharpen it before gathering data.
