RESEARCH / BLOG

Research,
Explained.

Read the research question, follow the authors’ methods and explore the evidence behind the article.

Research Blog

Research methodology · Research insights

How to Identify a Genuine Research Problem

Turn an observation into a researchable problem using rival explanations, a problem tree and FINER. A practical guide for doctoral research and proposal development.

By NPT Research7 min read
The novelty toolkit · S202 / 09
Start with the observation.Test rival explanations before choosing the solution.
NPT Research

The Novelty Toolkit · Series 2 · Part 2 of 9

A researcher notices that a water sensor reports slightly different values in the morning and afternoon. Another researcher notices the same thing and immediately declares: 'We need a smarter AI sensor.' Which one is doing science? Neither conclusion follows yet. A disciplined researcher first asks whether the change reflects the water, the measuring device, temperature, calibration, sampling, or chance.

A good research problem begins with curiosity, not with the solution you already want to sell. Think of a detective: a footprint is a clue, not a verdict.

Keep a field notebook before proposing solutions

Create a two-column observation log. In the left column write only what was seen or measured, including time, conditions, instrument, units and repetition. In the right column write interpretations or questions. Keep them separate. For example, 'sensor A measured 6.8 and sensor B measured 7.2 at 14:00' is an observation; 'heat caused the difference' is a proposed explanation.

This separation makes observations and interpretations easier to inspect; it does not guarantee that confirmation bias is removed. Record inconvenient observations as carefully as confirming ones. If a recurring pattern appears, formulate several explanations before selecting one. Do not discard an explanation simply because it requires a less fashionable methodology.

Use the five whys—but do not mistake the chain for proof

A common technique asks 'Why?' repeatedly. Why is the reading unstable? Perhaps calibration shifts. Why might calibration shift? Temperature, ageing or contamination may contribute. Why does that matter? Operators may incorrectly interpret changes as water-quality events. Each answer creates another question.

The exercise is useful for exposing assumptions, but it is not a validated causal model. It tends to follow one chain and can miss several interacting causes. After the five whys, draw at least three rival explanations and identify the measurement that would separate them.

The novelty toolkit · Series 2Follow the clue before choosing a solution.A sensor disagreement is an observation, not yet a diagnosis.
  1. 01
    RecordNote time, instrument, units and conditions.
  2. 02
    SeparateKeep observations apart from interpretations.
  3. 03
    BranchConsider several plausible explanations.
  4. 04
    BoundChoose an uncertainty that available evidence can address.
  5. 05
    ScreenCheck feasibility, ethics and significance.
A learning framework, not a measured result. Verify evidence before making a research claim.
Figure 1. The observation-to-problem pathway includes deliberate examination of alternatives.

Make a problem tree

Place the observed phenomenon in the trunk. Put possible causes below it and consequences above it. For a field sensor, causes might include temperature sensitivity, sensor ageing, inconsistent sample preparation and network artefacts. Consequences might include false alerts or unnecessary field visits.

Now mark each entry as observed, supported by literature, or hypothesized. This simple annotation prevents an attractive diagram from being mistaken for evidence. The most useful research question often sits on an uncertain connection between a cause and an outcome, rather than on the overall problem itself.

Choose a researchable uncertainty

A problem is researchable when there is a plausible path to generating evidence that can meaningfully reduce the uncertainty. Ask:

  • What exactly is unknown? Can it be expressed in a measurable relationship, explanatory mechanism or interpretive question?
  • What competing explanations might produce the same observation?
  • Which data, measurements, participants or documents would be necessary?
  • What result would make me change my mind?
  • Is the required access, expertise, time, funding and ethical approval realistic?

These questions place the quality of the research design ahead of technological excitement.

Turn a symptom into a bounded study

A weak problem statement is: 'Water sensors are inaccurate, so develop an AI model.' This confuses a concern with a preferred intervention. A stronger starting statement is: 'For sensors deployed outdoors for twelve weeks, the relative contributions of ambient temperature and elapsed time since calibration to disagreement with reference measurements are uncertain.'

That revised statement suggests a testable comparison: readings from multiple instruments against a reference method, measured at pre-specified time intervals across relevant environmental conditions. It does not yet claim the design is new; that requires a literature search.

The novelty toolkit · Series 2Keep the explanation open to challenge.A structured notebook makes assumptions visible.
  1. 01SymptomWhat is repeatedly observed?
  2. 02Possible causeWhich mechanism could explain it?
  3. 03Rival explanationWhat else could produce the same observation?
  4. 04Required evidenceWhich comparison would distinguish the alternatives?
A learning framework, not a measured result. Verify evidence before making a research claim.
Figure 2. Keep symptom, possible cause, alternative and required evidence in separate boxes.

Use the FINER screen, not only enthusiasm

FINER asks whether a question is feasible, interesting, novel, ethical and relevant (Stanford’s research-question guidance). The novelty element is a test against literature, not the researcher's personal excitement. Feasibility includes access to a valid reference measurement and ability to estimate variability. Ethical suitability includes relevant institutional approval, even for seemingly benign human-subject surveys or field deployments.

A modest, well-measured question is often more valuable than a sweeping problem statement that requires equipment or populations you cannot obtain.

Practice: the rival-explanation challenge

Write one genuine observation from your field. Generate three explanations, including one that would make your preferred theory wrong. For each, list one observable prediction and one potential discriminating test. Then write a problem statement without using the words 'innovative', 'novel', 'AI-driven' or 'first'. If the problem still seems worth studying, you have a sturdier starting point.

A useful one-paragraph template

In [defined context], [repeatable observation] occurs. Explanations involving [A] and [B] are plausible, but their relative roles remain uncertain under [specified conditions]. Resolving this uncertainty would allow us to [scientific or practical significance]. This motivates an investigation of [bounded question].

When the 'five whys' sends you in the wrong direction

Suppose the first answer is 'because the sensor is defective'. Asking why that defect occurred may lead you into manufacturing quality, maintenance or operator training. Meanwhile, the supposed defect might be a real change in water chemistry. The technique is useful only if you question the first answer and seek independent checks. A branching causal diagram is often safer than a single chain of explanations.

Try this rule: each proposed cause must have one observable implication different from at least one alternative. If temperature affects sensor output directly, a controlled temperature change with stable reference conditions could reveal it. If biofouling is responsible, cleaning or surface inspection may offer a different line of evidence. Which test is best depends on safety, resources, and relevant instrument procedures.

Run a low-cost feasibility pilot

Before writing a full doctoral proposal, perform a small, appropriately authorized pilot when feasible. Its purpose is not to prove the final hypothesis. It is to check whether measurements are stable, reference data are available, the phenomenon occurs often enough and the planned protocol generates interpretable information. Record unexpected obstacles rather than hiding them.

A pilot can also tell you that your original question is not answerable with available instruments. That is useful knowledge. Modify the question rather than presenting an unsuitable device as if it were a valid reference standard. If human participants are involved, confirm required approval before collecting pilot data.

Three problem statements—choose the stronger one

A: 'Development of an innovative AI-integrated monitoring device.' This is a project label, not a research problem.

B: 'Current devices are insufficient.' This is a claim that needs evidence and clear performance criteria.

C: 'Under specified field conditions, it is unclear whether measurement disagreement is primarily associated with temperature, instrument age or both.' This statement identifies uncertainty and candidate explanations, and it suggests what measurements might be needed.

Improve C by specifying the target instruments, relevant conditions, measurement scale and scientific or practical consequence. Then search the literature to determine whether it is truly unresolved. The goal is not to make the problem sentence sound more technical; it is to make the uncertainty more visible.

What to remember

The strongest research questions do not always arise from dramatic discoveries. They arise when a careful observer refuses to confuse an explanation with an observation. The next tutorial shows how to determine whether that uncertainty has already been examined by other researchers.

Methodological context and limits

Farrugia and colleagues’ question–hypothesis–objective guide explains why question and design need to agree. Covvey and colleagues’ question-formulation commentary discusses question frameworks and common pitfalls. These are methodological guides; neither validates a diagnosis made from the fictional sensor observations.

Continue the research novelty series

Series 1: originality, evidence and scientific judgment

Begin with what makes research novel, then find the closest prior research and build a novelty matrix. Before using an assistant, examine the limits of AI novelty judgments and why the researcher owns the question. Develop the conversation with scientific prompting and six-round brainstorming, then challenge the idea against evidence. The complete worked research dialogue connects those decisions.

Series 2: from a problem to a defensible proposal

First separate a research gap from a research problem and identify a genuine problem worth investigating. Next search published literature for the gap and align the question, objectives and hypothesis. For AI-assisted work, clarify the problem without outsourcing judgment, verify the literature search and its references and compare papers in an evidence matrix. Then stress-test the research hypothesis before following the complete observation-to-proposal workflow.

Sources and further reading

  1. Stanford University. Selecting Your Research Question: FINER and study planning. Institutional guidance.
  2. Farrugia et al. Research questions, hypotheses and objectives. Canadian Journal of Surgery 53, 278–281 (2010).
  3. Covvey, McClendon & Gionfriddo. Back to the basics: Guidance for formulating good research questions. Research in Social and Administrative Pharmacy 20, 66–69 (2024; online 2023).