How to Write a Research Question Step by Step

Learning how to write a research question the hard way is a common way to be handed back a proposal with 'this is a topic, not a question' in the margin by your supervisor. The line above the comment usually reads something like 'this paper examines remote work and employee wellbeing,' which names a field, not something anyone could go and answer. Turning an area of interest into a real question happens in stages, and skipping one is exactly what produces a methods section that can't say what it's testing.

The reliable version of that process has four moves. Narrow the topic until it names a population, a variable, and an outcome. Check the result against a five-part test called FINER. Tighten the wording until nothing in it could be misread. This piece runs one example, whether retrieval practice helps students retain material longer than rereading does, through all four moves, so each one is visible rather than just asserted.

Start With the Topic, Not the Question

A topic names a field. 'Study techniques and exam performance' is a topic: it tells a reader what area you're working in and nothing about what you actually want to find out. A question names a relationship you intend to investigate. Until a sentence does that, there is nothing yet for a methods section to test.

A first attempt at narrowing it might read 'does retrieval practice help students learn better than rereading.' That is progress, but the sentence still cannot be studied as written. 'Help' and 'learn better' name no measurement, and 'students' could mean a primary school class or a doctoral cohort. A question this loose forces every reader, supervisor included, to guess at what you actually meant.

How to Write a Research Question That Names Its Parts

A workable research question names three things: a population, a variable you can measure or manipulate, and an outcome. Population is who or what you are studying. Variable is the thing that differs from case to case, whether that is a behavior, a condition, or a treatment someone receives. Outcome is what you would actually measure to see whether the variable made a difference.

Applied to the retrieval-practice topic, the population becomes first-year undergraduates in an introductory course, not 'students' in general. The variable becomes weekly low-stakes retrieval quizzes compared with rereading the same material for the same amount of time, not the vague idea of 'retrieval practice' on its own. The outcome becomes a score on a delayed test given two weeks after the material was covered, not 'learning' left as an unmeasured idea. Put together: among first-year undergraduates in an introductory course, do students who complete weekly low-stakes retrieval quizzes score higher on a delayed test two weeks later than students who spend the same amount of time rereading the material?

Run the Question Through the FINER Test

It's a five-part check called FINER proposed by Hulley and colleagues. FINER stands for feasible, interesting, novel, ethical, and applicable. It's used for clinical research but we teach it really well outside of medicine because those same five failure modes appear in every field. A question has to clear all five, not just the ones that are easy to answer yes to.

Letter	What it asks	Applied to the retrieval-practice question
F, feasible	Can you actually get the participants, time, and access this needs	One course section, two weeks, and a standard quiz platform make this realistic to run in a single term
I, interesting	Would the answer matter to you, a supervisor, or the field	Instructors deciding how to structure review sessions have a direct stake in the answer
N, novel	Does it extend or challenge something already published, rather than repeat it	Most published comparisons pit quizzing against no additional review at all, rather than against rereading matched for time
E, ethical	Would an ethics board approve how you would study it	A classroom quiz and a delayed test carry no more than minimal risk
R, relevant	Does it connect to a real theoretical or practical question	Connects to how courses are actually designed, beyond the psychology of memory in the abstract

Running a draft question through a research question generator before committing to it applies roughly this same check in seconds rather than across a week of second-guessing. It is most useful for catching the failure that is hardest to see in your own writing, a question that sounds specific but still names no real variable.

Tighten the Wording Until Nothing Can Be Misread

Once the three parts are in place, the words connecting them still do a lot of damage if left loose. Verbs like 'affect,' 'impact,' and 'improve' sound precise and measure nothing on their own; a reader cannot tell from 'affect' whether you mean increase, decrease, or simply correlate with. The fix is the same across every field: replace the vague verb with the actual comparison or measurement you plan to make.

Weak version	What's missing	Strong version
Does remote work affect productivity?	No population, no defined variable, no measurable outcome	Among software engineers at mid-size companies, does working from home more than three days a week change the number of code reviews completed per sprint, compared with working from home one day a week or less?
Does nurse fatigue affect patient safety?	'Fatigue' and 'safety' are both unmeasured, and there is no comparison built in	Among registered nurses working three consecutive night shifts, is self-reported fatigue associated with a higher rate of medication administration errors than among nurses working a single night shift?
Is urban green space good for mental health?	'Good for' assumes the direction of the answer before the study runs	Among adults living within two kilometers of a public park, is weekly park-visit frequency associated with lower scores on a standard anxiety scale?

Every strong version on the right does the same three things: it names who is being studied, it replaces a vague verb with a measurable comparison, and it drops any word that assumes the answer in advance. None of that makes the question longer for its own sake. It makes the question possible to actually go and answer.

Common Ways a Draft Question Fails

Four patterns account for most unusable research questions, and all four are easy to spot once you know to look for them. Reading a draft question aloud and asking whether two different researchers could design two different studies from it catches most of them before a supervisor has to.

• A yes-or-no dead end, such as 'is homework helpful,' which produces a single bit of information and cannot be developed into a real study design
• Two questions stacked into one, such as 'how does stress affect academic performance and social relationships,' which forces a choice a writer avoided making
• A value-laden frame, such as 'why is remote work better than office work,' which assumes the answer the study is supposed to find out
• A missing outcome, such as 'how does classroom design impact learning,' where neither 'impact' nor 'learning' names anything a researcher could measure

What Comes After the Question Is Settled

A question that survives the FINER test still forks into two different next moves. An empirical version needs a hypothesis, a single testable prediction plus its null counterpart, while a version built for a literature review or an argumentative paper needs a thesis statement instead. The fuller sequence, from this exact step through that fork to a finished outline, is worked through in our guide to how to plan a research paper.

The types of research questions are catalogued separately, and the distinction between a research question and a hypothesis decides which of them you actually need.

A tightened question is worth more than a polished paragraph at this stage, since everything downstream, the hypothesis, the outline, the actual writing, takes its shape from it. An AI humanizer built for academic writing is worth knowing about for later, once a draft exists that needs to sound like the person who asked the question rather than a template.