Research Hypothesis Examples
Six sample hypotheses from real dissertations, with our comments
The 30-second summary
- Example 1 names the direction: "positively related to" rather than "has an impact on". Without a direction there is no way to be wrong.
- Example 2 splits one claim in two when the prior literature genuinely disagrees. Two competing sub-hypotheses, and the experiment decides.
- Example 3 predicts who a relationship holds for. This is usually where a quantitative dissertation makes its actual contribution.
- Example 4 is a set that builds. Three hypotheses ordered so the third only makes sense because the first two came before it.
- Example 5 has no hypothesis at all, and is right not to. Exploratory studies ask research questions instead.
- Example 6 explains the H0 label you will meet in other theses, which often does not mean what you think it does.
Navigate
A research hypothesis is a testable prediction about the relationship between two or more variables. The examples below are not invented for a blog post: every one is quoted from a real dissertation, and you can download each thesis and see how the hypothesis sits inside the whole argument.
That matters more than it sounds. Most hypothesis examples online are written to illustrate a definition, so they are tidy in a way real research never is. Reading a hypothesis in its natural habitat, next to the study that had to test it, teaches you something a clean example cannot.
I have read a few hundred methodology chapters, and weak hypotheses nearly always fail the same way: they name a relationship without saying which direction it runs, or who it applies to. Each example below shows one structural move, and most of them end with a weaker version and the fix so you can see the difference on the page.
Example 1: A directional hypothesis
This is the plainest shape a hypothesis takes, and the one most dissertations need most of the time. It comes from a study of whether Vietnamese shoppers will collect their online orders from a pickup point instead of having them delivered home.
Why it works. Three things are pinned down in one line. The variables are named and abbreviated so they can be tracked through the analysis (price value, adoption intention). The direction is stated: positively. And because the study defines its population early, the reader knows who this claim is about.
The word doing the work is “positively”. Without it you have a topic, not a prediction. With it, the study can be wrong in a specific way, which is the whole point of a hypothesis.
To make that concrete, here is the same claim written the way it usually arrives in a first draft, and the version that survives.
Weak. Price value has an impact on adoption intention.
Better. Higher perceived price value is positively related to adoption intention for collection-and-delivery points.
The first cannot fail. Any result at all, in either direction, could be described as “an impact”.
Example 2: Splitting one claim in two
Sometimes the honest answer is that you do not know which way a relationship runs, because the published research does not agree either. This study of hotel brand partnerships handles that by writing both predictions and letting the experiment settle it.
Why it works. Two hypotheses that contradict each other looks like a mistake, and it is deliberate. Cai’s literature review found genuine disagreement about whether complementary brand personalities help or hurt an alliance, so she wrote both competing predictions rather than picking a side.
This is a legitimate and underused move. If the prior research honestly points both ways, saying so is stronger than pretending half the literature does not exist. Your examiner has read it too.
The lettering matters as well. H1a and H1b signal that these belong to the same question, rather than being two unrelated predictions.
Compare that with the way most drafts handle genuine uncertainty:
Weak. H1: Brand personality congruence affects brand alliance attitudes, though findings in the literature are mixed.
Better. H1a and H1b as written above, with the mixed literature explained in the paragraph that introduces them.
Hedging inside the hypothesis makes it untestable. Put the uncertainty in the framing and keep each hypothesis sharp.
Example 3: A moderation hypothesis
The first nine hypotheses in the Luong study predict relationships. These last two predict that those relationships change depending on who you ask, which is a different and more interesting kind of claim.
Why it works. A moderation hypothesis is usually where a quantitative dissertation earns its contribution. “X predicts Y” may already be established in your field. “X predicts Y for experienced users but not for new ones” is new, useful, and much harder for a reviewer to shrug at.
It also forces a decision you should be making anyway: if you have collected demographic data, you are implicitly claiming it might matter. State that as a hypothesis or explain why you collected it.
Here is the difference between gesturing at that and claiming it:
Weak. Income may play a role in adoption intention.
Better. The relationship between the adoption antecedents and adoption intention differs significantly across income groups.
One treats income as another variable to control for. The other says something specific about the shape of the model, and that is what moderation means.
Example 4: A set that builds
A single hypothesis is rarely the interesting unit. What an examiner reads is the set, and whether it adds up to the question your study actually asks.
Read them in order and you can see the argument being assembled. Each one adds something the next needs.
| Hypothesis | What it adds |
|---|---|
| H1 | Symptoms relate to getting help. Establishes that the outcome is worth predicting at all. |
| H2 | Family strengths push symptoms down. Introduces the protective factor. |
| H3 | Family strengths also push help-seeking up, directly and through symptoms. Ties the two together and names both routes. |
No single hypothesis states the question the study actually cares about, which is whether family strengths work directly, indirectly, or both. The set states it.
Notice too that Reyes writes H2 and H3 as one sentence joined by “and”. They are two predictions sharing a subject and a figure, so keeping them together tells the reader they are one idea tested two ways.
There is no weak version to show here, because the failure is subtractive rather than a matter of wording. It looks like three hypotheses that each test a separate correlation in an order you could shuffle without losing anything. So the test is a deletion test: take any one of your hypotheses out, and if the argument survives unchanged, it was decoration.
Example 5: When you should have none
Not every dissertation needs a hypothesis, and forcing one in is the most expensive mistake in this post. Here is how a qualitative thesis opens its inquiry instead.
Why there is no hypothesis here. A hypothesis predicts a relationship you can measure and be wrong about. Carlson is not measuring a relationship. She is asking what emerges, which means she does not know the categories yet, and inventing a prediction would commit her to categories before she had met her participants.
That is not a softer form of research. It is a different logic, and it runs all the way down to your research philosophy: what you believe counts as knowledge decides whether prediction is even the right move. Interpretivist work generates the concepts; positivist work tests predictions about concepts we already have. If your philosophy chapter says one thing and your hypotheses say the other, an examiner will find the seam.
The rule of thumb I give coaching clients: if you can write your prediction as “more X, less Y” and imagine the table that would test it, write a hypothesis. If your central question starts with “how”, “what” or “in what ways”, and your answer will be themes rather than numbers, write research questions and stop apologizing for it.
Where students get into real trouble is the middle case: a qualitative study with a hypothesis bolted on because a template had a heading for it. Examiners notice immediately, because the hypothesis is never mentioned again after the methodology chapter.
Example 6: Why H0 is rarely written out
One last thing, because it confuses people reading other dissertations for guidance. You will occasionally meet a hypothesis labeled H0, and the label does not always mean what the statistics textbook says it means.
Read that label carefully. It says H0, but it is not a null hypothesis. A null hypothesis predicts no relationship. This one predicts a bidirectional relationship and even predicts which direction is stronger. Reyes simply numbered her hypotheses starting at zero, and her Figure 14 is captioned “Bidirectional Association of Parent-Child Depression (H0)”.
I point this out because students lose marks on it in both directions. Some copy the label into their own work for a prediction that is not a null. Others assume they must write out a formal null for every hypothesis, and pad their methodology chapter with a page of “there is no significant difference between…” statements nobody asked for.
What actually happens in a dissertation. You state the alternative hypothesis, the thing you expect. The null is its logical opposite, and your statistical test evaluates it whether you write it down or not. That is why, reading through real theses, you will find the alternative stated and the null mostly implied.
When you should write it out anyway: if your program’s template asks for it, if your supervisor asks for it, or if the null is genuinely interesting because prior research expected an effect and you suspect there is none. Otherwise, state what you predict and let the statistics do their job.
Still have questions?
How many hypotheses should a dissertation have?
There is no fixed number, and the right answer follows from your research questions rather than a quota. Small experimental studies often have two or three; the Luong thesis above has eleven because it tests a full model with moderators. What matters is that each one is testable and earns its place in the argument.
What is the difference between a hypothesis and a research question?
A research question asks what is going on; a hypothesis predicts what you will find. Quantitative studies usually have both, with the questions framing the study and the hypotheses stating the specific testable predictions underneath them. Purely qualitative studies normally have questions only.
Do I need to write out the null hypothesis?
Usually not, unless your program’s template or your supervisor asks for it. You state the alternative hypothesis, the effect you expect, and your statistical test evaluates the null whether or not you have written it down. Check your department’s guidelines, because a few still require both to be stated explicitly.
Can a hypothesis be non-directional?
Yes. If the literature genuinely gives you no basis to predict which way a relationship runs, a non-directional hypothesis (“there will be a difference between the groups”) is legitimate. Use it because the evidence is genuinely mixed, though, not to avoid committing to a prediction you could get wrong.
Where can I find more full dissertation examples?
We host a collection of complete, openly licensed dissertations across twelve disciplines, all free to download and read in full. Every thesis quoted in this post is in there. Browse the dissertation and thesis examples collection.
Can’t find your answer here? Ask a Grad Coach directly – the initial chat is free.
Speak with a friendly coach →
