What is the difference between a research question and a hypothesis?

The choice is usually made by the design rather than by preference. If your study will produce a number that could be larger or smaller than expected, you have a prediction whether or not you write it down, and writing it down before you look is what makes the test meaningful. If your study will produce categories, patterns, or an interpretation, a prediction constrains you before you know enough to make one.

Methods guides put the sequence in order. Ratan and colleagues treat the research question as a response to an existing uncertainty and sort questions into those about existence, description, relationships, comparisons, and causality. They then test a candidate question against ten criteria abbreviated as FINERMAPS, meaning feasible, interesting, novel, ethical, relevant, manageable, appropriate, potential value, publishability, and systematic (Ratan et al., 2019, p. 15). The hypothesis comes afterwards as what they call an operational transformation of the question, a statement of the expected nature and direction of a relationship between variables (Ratan et al., 2019, p. 18).

The type of question also picks the design before any prediction is written. A question about disease incidence points to a survey, a question about risk factors points to a case-control or cohort study, and a question about the effect of an intervention points to a clinical trial, with the hypothesis framed once the design is settled (Ratan et al., 2019, p. 19).

The confusion is made worse by fields that use the words differently. In much of psychology and medicine, hypotheses are mandatory and questions are informal framing. In sociology, history, and most of the humanities, hypotheses are rare and would look odd. Neither convention is wrong, and the safe move is to read three recent theses from your own department rather than a generic methods guide.

Where a hypothesis is appropriate, the important discipline is specifying it before you see the data. A prediction written after analysis is not a test, and presenting it as one is a recognized problem with a name. Preregistration exists precisely to make the timing verifiable, and even an informal dated note to your supervisor is better than nothing.

The null hypothesis deserves a mention, since it is where the confusion concentrates. It is a statistical device, not your scientific expectation. You can predict an effect and still test against a null of no effect, and the two statements do different jobs.

Greenland and colleagues define the study hypothesis as a specified effect size targeted for analysis, and the null is only the version that specifies zero. A test can instead concern a nonzero effect, or whether an effect sits above, below, or inside a stated boundary (Greenland et al., 2016, pp. 338-339). They also warn against reading power as a verdict on a finished study. If two studies each have 80% power and the alternative is correct, both reach P at or below 0.05 only 64% of the time, and in 32% of cases one crosses that threshold while the other does not, which is why they ask you to report point estimates and the range of compatible effects rather than an isolated null test (Greenland et al., 2016, p. 345, and p. 340).

Can a qualitative study have a hypothesis?

It can, but usually it should not lead with one. Qualitative designs are built to discover categories rather than test predictions, and a stated hypothesis pushes you towards confirming it. What qualitative work does carry is expectations, and making those explicit as reflexivity is more honest than dressing them as hypotheses.

Grounded theory takes this furthest and asks you to hold the literature at arm's length while coding, precisely so the categories come from the data. Other qualitative traditions are more comfortable with prior theory, and some, such as framework analysis, start from an existing structure openly.

If a supervisor or committee asks for hypotheses in a qualitative study, ask what they want the hypotheses to do. Often the real request is for a clearer statement of what you expect to find and why, which reflexivity handles better.

Do I need both in the same study?

Only when the question is broad and the test is narrow. Then the question frames the chapter and the hypotheses specify what will be measured. Stating both when the hypotheses simply restate the question adds a page and no information, which reviewers notice.

Mixed methods designs are where both genuinely belong. The qualitative strand carries questions, the quantitative strand carries hypotheses, and the integration section explains how the two speak to each other.

Henderson and Chambers give a format for keeping the two aligned. Where hypotheses apply, they should follow directly from the research questions and turn a theoretical prediction into an observable outcome, set out in a design summary table that places each question and hypothesis next to its sampling plan, its analysis, and the result that would confirm or disconfirm it (Henderson and Chambers, 2022, p. 3). They also push back on the pressure to predict a positive result, since a study should be informative whether or not the hypothesis is supported, because the value sits in the question and the quality of the methods rather than in the outcome (Henderson and Chambers, 2022, p. 6).

When you do use both, number them and keep the numbering consistent through methods, results, and discussion. A reader tracking hypothesis three through a hundred pages should never have to search for it.