# Research Question vs. Hypothesis in Biology: Examples

A research question asks what is going on. A hypothesis states what you expect the answer to be and commits you to a test that could prove you wrong. That is the entire difference between a research question and a hypothesis in biology, and it shapes how you design, analyze, and report a study. The rule of thumb is simple. Write a research question when you are defining a problem and cannot yet make a specific, falsifiable prediction. Write a hypothesis when you can. Most strong projects use both in sequence, a question to frame the gap, then a hypothesis to drive the experiment.

## The Short Answer: Research Question vs. Hypothesis in Biology

A research question is an interrogative statement that defines what a study is trying to find out, such as "How does water temperature affect hatching time in freshwater fish embryos?" A hypothesis is a declarative, testable proposed answer, such as "Rising water temperature accelerates hatching by speeding embryonic enzyme activity." The question frames the gap in knowledge. The hypothesis commits you to a mechanism, a direction, and a specific way of being wrong.

Not sure whether to write a research question or a hypothesis? Check how much you can predict right now.

Use a research question when:

- The literature is thin or contradictory and you cannot yet predict an outcome.
- The study is descriptive, such as a survey, a species census, or a tissue atlas.
- You are generating discovery data, for example from genetic screens (systematic surveys that disrupt or measure many genes at once) or [transcriptomics](/knowledge/bioinformatics/modern-transcriptomics-bulk-single-cell-spatial) (measurement of the full set of RNA transcripts in a sample), where the point is to see what is there before claiming what it means.
- A community is setting priorities for an entire field.

Use a hypothesis when:

- You can state a specific, falsifiable prediction with defined variables.
- The design includes controls, replication, and a planned statistical test.
- You are testing a mechanism, not just describing a pattern.
- Competing explanations exist and your data can discriminate between them.

## What a Research Question Is

A research question names the biological system, the variables or processes of interest, and the context in which you will study them. "Biology" is not a question. "Does urban light pollution shift the dawn chorus of songbirds?" is a question, because it names a cause, an effect, and a system.

Research methods textbooks converge on a short list of criteria for a good question, often summarized as FINER: feasible, interesting, novel, ethical, and relevant. Feasible means you can actually access the organisms, samples, instruments, and time the question requires. Novel means the answer is not already established in the literature. In biology, a workable question also needs a defined scale, because the same broad topic looks different at the molecule, cell, organism, and ecosystem levels.

Questions come in recognizable types, and the type determines your design:

- **Descriptive** questions ask what exists. "Which bird species use this restored wetland in winter?"
- **Comparative** questions ask how groups differ. "Do embryos hatch sooner at 24 than at 18 degrees Celsius?"
- **Correlational** questions ask whether two variables move together.
- **Mechanistic** questions ask how or why a process works. Quantitative systems pharmacology, a modeling discipline in drug development, frames its mechanistic work in exactly this form: given specified mechanistic assumptions, how does the system behave under intervention? 

Big datasets routinely generate questions rather than answer them. The genome of the green alga *Spirogyra pratensis* revealed homologous frameworks for hormones and morphogenetic regulation in a lineage with a simple body plan, raising questions about which parts of the land plant developmental toolkit were already present in the last common ancestor of land plants and algae [1].

Research questions also operate at the level of whole fields. A horizon-scanning exercise (a structured method for collecting and ranking forward-looking questions) in subterranean biology, the study of cave and groundwater life, gathered 258 candidate questions from 30 scientists, then used online surveys of 130 researchers to distill 50 top-priority questions across macroecology, microbial ecology, adaptation, evolution, and conservation [2]. Generative AI researchers borrowed the same logic from mathematics, where Hilbert's list of 23 problems focused a field for more than a century, and proposed fifteen grand challenges to concentrate [cell biology](/blog/careers/cell-biology) attention on questions that still lack effective predictive methods [3].

### How to Write a Research Question, Step by Step

![Six numbered steps from a hatchery observation through gap search, system choice, question type, scoping, and feasibility check.](/article-images/illustrations/research-question-vs-hypothesis-biology-examples-question-steps.webp)

Use this running example: you notice that fish embryos in a warm hatchery tank hatch earlier than embryos in a cold one.

1. **Start from a concrete observation.** "Warm-tank embryos hatch first" is an observation you can verify before asking anything else.
2. **Locate the gap.** Search the literature and public databases. If hatching responses to temperature are well documented for trout but unknown for your species, the gap is species-specific.
3. **Choose the system and scale.** One species, one life stage, one response variable. "Freshwater fish embryos" is too broad. "Embryos of the fathead minnow, *Pimephales promelas*, incubated between 18 and 24 degrees Celsius" is workable.
4. **Pick a question type.** Descriptive ("At what temperature do embryos hatch?"), comparative ("Do embryos hatch sooner at 24 than at 18 degrees?"), or mechanistic ("Does temperature accelerate hatching through embryonic enzyme activity?").
5. **Scope it with variables and context.** Name the independent variable (temperature), the dependent variable (hatching time), and the boundaries (one species, controlled oxygen, defined light cycle).
6. **Stress-test feasibility.** Count the tanks, the incubators, the days, and the permits you need. If the answer requires resources you do not have, narrow the question.

A weak question fails on specificity. "How does climate change affect fish?" names no variable, no species, and no measurable outcome. The improved version, "Does a 2 degree Celsius increase in mean incubation temperature shorten hatching time in fathead minnow embryos?", can be answered yes or no with data.

## What a Hypothesis Is

A hypothesis is a proposed answer to a research question, stated specifically enough that evidence could refute it. A statement is falsifiable when some possible observation would count against it. "Temperature affects hatching" is not falsifiable in any useful way, because any result, including no difference, could be described as an effect. "A 2 degree Celsius increase in incubation temperature shortens median hatching time in fathead minnow embryos by at least one day" is falsifiable, because a null result refutes it.

Two distinctions matter here.

**Hypothesis versus prediction.** The hypothesis is the proposed explanation. Predictions are the observable consequences it implies. If the hypothesis is "temperature accelerates hatching by speeding embryonic enzyme activity," then predictions follow: warmed embryos hatch sooner, enzyme inhibitors delay hatching in warmed embryos, and enzyme activity measured in warmed embryos rises before hatching. A test that confirms a prediction supports the hypothesis without proving it, because other mechanisms could produce the same prediction.

**Research hypothesis versus null hypothesis.** In much of biomedical research, the statistical default is the null hypothesis, the assumption that a difference between groups does not exist unless the data demonstrate otherwise [4]. Null [hypothesis testing](/blog/data-analysis/what-is-hypothesis-testing) asks whether an observed difference in a sample is likely to exist in the broader population. This framing has a well-known trap. Studies with low [statistical power](/blog/guides/statistical-power-what-it-is-and-why-it-matters-in-research), meaning a low probability of detecting a true difference, can miss real effects, and the absence of a significant result is then misread as evidence that two treatments are equivalent [4]. The null hypothesis is a testing convention, not a biological claim that nothing is happening.

Hypotheses also vary in breadth. Many are single-experiment claims. Others are frameworks meant to organize whole areas of biology. The core and periphery hypothesis in cell and [developmental biology](/blog/careers/developmental-biology) proposes that many systems decompose into a versatile core with a large behavioral repertoire and a specific periphery that configures it, with examples ranging from Turing patterning to actomyosin dynamics (the contractile actin-myosin machinery that shapes cells) and vertebrate gastrulation [5]. A framework of this kind is tested cumulatively, across many systems, rather than in a single experiment.

### Anatomy of a Testable Hypothesis

A strong hypothesis contains five parts.

1. **Variables.** What you manipulate or compare, and what you measure.
2. **Direction or relationship.** More, less, earlier, later, or a specific association.
3. **Scope.** The organism, tissue, environment, and conditions it applies to.
4. **Mechanism, when known.** The proposed cause. Mechanistic hypotheses generate more predictions and are therefore more valuable, but phenomenological hypotheses ("X tracks Y") are legitimate starting points.
5. **Falsifiability.** A described result that would refute it.

A worked example from sensory biology shows the parts in place. Animals carry many light-sensitive molecules, including opsins and cryptochromes (two large families of light-detecting proteins), even in species that live in environments humans would consider dark. The underlying question is why there are so many photoreceptors. One plausible hypothesis holds that this diversity represents adaptation to the complex light environment of each species' ecological niche, because light conveys information through spectral and intensity changes specific to a given habitat [6]. The hypothesis is testable because it predicts that receptor diversity should track niche complexity, and it can be checked against species from different habitats.

Hypotheses become sharpest when they compete. In epigenetics (the study of heritable changes in [gene regulation](/blog/guides/gene-regulation) that do not alter DNA sequence), the epigenetic switch hypothesis interprets certain cases of [epigenetic inheritance](/knowledge/molecular-biology/epigenetic-inheritance) as an adaptive organismal response to environmental change. An alternative explanation attributes the same observations to coevolutionary dynamics between transposons (mobile DNA elements that copy themselves within a genome) and the host genome. A survey of fifty years of publications across four disciplines found that enthusiasm for epigenetics was often inversely related to interest in transposon dynamics, which shows how easily a favored hypothesis can crowd its rival out of attention [7]. Stating the [alternative hypothesis](/blog/data-analysis/alternative-hypothesis-definition-examples) is part of designing a fair test.

## Side-by-Side Comparison

| Feature | Research Question | Hypothesis |
|--|--|--|
| What it is | An interrogative statement defining what the study will find out | A declarative, testable proposed answer or explanation |
| Form | Ends with a question mark | A falsifiable statement, often with a mechanism |
| Commits you to | A scope, a system, and a gap worth filling | A direction, defined variables, and a way of being wrong |
| Typical timing | Early, during gap analysis and design | After background reading, before data collection |
| Falsifiability | Not applicable, it frames rather than asserts | Required, a refuting result must be describable |
| Role in statistics | Guides the choice of endpoints and comparisons | Pairs with a null hypothesis for formal testing |
| Example (fish embryos) | "Does a 2 degree Celsius increase in incubation temperature shorten hatching time in fathead minnow embryos?" | "Warmer incubation shortens hatching by accelerating embryonic enzyme activity." |
| Typical verbs | What, which, how, does, why | Predicts, drives, mediates, accelerates, impairs |
| Failure mode | Too broad, unanswerable, or already answered | Unfalsifiable, or so vague that any result fits |
| Best suited to | Exploration, description, priority-setting, discovery data | Mechanism testing, controlled experiments, modeling |

## How a Question Becomes a Hypothesis in a Real Study

The standard sequence runs observation, question, background reading, hypothesis, predictions, design, test, analysis, and then new questions. Published work shows how flexible that sequence can be.

**The inverted case.** Human genetics traditionally starts with a phenotype, finds affected individuals, and asks which variants they share. Falling sequencing costs inverted the scientific question. In genotype-first research, investigators gather individuals who carry a specific genomic variant and then test the hypothesis that an associated phenotype exists, using targeted phenotypic evaluations in an approach called reverse phenotyping [8]. Thirteen pilot studies at the National Institutes of Health used this design to facilitate novel genotype-disease associations, expand known phenotypic spectra, and demonstrate ex vivo functional mechanisms of disease [8]. The question and the hypothesis are both present, but their order is reversed.

**The cumulative case.** Some questions take decades. Whether human culture arose through an evolutionary transition in individuality, the process by which cooperating units become new individuals, is a question rather than a single experiment. One analysis examined the steps of biological transitions of this kind, illustrated them with the well-studied evolution of multicellularity in the volvocine algae (a green algal lineage spanning single cells to complex multicellular colonies), and applied the same stages to hominin technology during the early Pleistocene. The analysis supported the hypothesis that human culture has undergone such a transition [9]. The question stayed stable while the hypothesis accumulated support across independent lines of evidence.

**The partial-support case.** A hypothesis can be partly supported and still instructive. Reviews of the proposed gut microbiota-nerve growth factor (NGF) axis in postmenopausal bone loss distinguish mechanisms backed by experimental evidence from those supported only indirectly, and note that no study has yet shown that microbiota-driven changes in NGF signaling are necessary or sufficient for abnormal bone remodeling [10]. The hypothesis remains active, but its components mature at different rates, and honest reporting keeps those tiers separate.

## When to Use a Research Question and When to Use a Hypothesis

![Two-column comparison of research questions and hypotheses across form, commitment, falsifiability, timing, and failure mode.](/article-images/illustrations/research-question-vs-hypothesis-biology-examples-question-vs-hypothesis.webp)

The choice depends on what you already know and what kind of data you will collect.

**Choose a research question when knowledge is thin or the data are exploratory.** Discovery approaches such as genetic screens and transcriptomics can provide an unbiased view of organism function, and they are intrinsically hypothesis-generating, meaning their output is a set of new hypotheses rather than a verdict on an old one [11]. Journals have responded to this reality. The [Journal of Experimental Biology](/blog/guides/journal-of-experimental-biology) broadened its scope to accept non-hypothesis-led research, arguing that blue skies work (curiosity-driven experiments not tied to a predefined prediction) can produce genuinely novel insights and that hypothesis-led and hypothesis-free experimentation are not philosophically antagonistic [11].

**Choose a hypothesis when you can define the test.** If you can name the variables, the direction, the controls, and the analysis before collecting data, a hypothesis will sharpen every one of those decisions.

**Modeling blends both.** In neuroscience, hypothesis-driven modeling brings a question into focus and builds a model to test a specific claim about how a system works, while data-driven modeling lets the data shape model structure with less prior commitment. Combining the two, under shared standards that keep models findable, accessible, interoperable, and reusable (the FAIR principles), improves understanding across biological scales [12].

**Data exploration can extend literature search.** Software that overlays neuroimaging and transcriptomic data offers a low-threshold, non-biased approach to hypothesis generation, and reviews of such tools evaluate them by how well they support example research questions [13]. Statistical machinery can formalize the same step. A Bayesian workflow that ranks model parameters by Kullback-Leibler divergence, a measure of how much information each parameter gained from the data, helps researchers form and refine hypotheses systematically instead of by intuition [14].

**Existing data can seed new hypotheses cheaply.** Meta-analyses of public datasets, made findable and reusable through the FAIR principles, offer a lower-cost route to formulating new hypotheses before any new experiment is run [11].

A practical decision list:

- Can I state a specific, falsifiable prediction today? If yes, write the hypothesis.
- Am I describing a pattern for the first time? Write the question and resist inventing a mechanism.
- Do competing explanations exist? State the hypothesis and its main alternative.
- Will the output be a dataset that others will mine? Frame the question, plan the metadata, and let hypotheses come later.

## Worked Examples: From Biology Research Question to Testable Hypothesis

**Example 1, ecology.** Question: "Which environmental variables best predict the species richness of aquatic cave-obligate animals in karst groundwater (terrain dissolved from soluble rock such as limestone, riddled with caves and conduits)?" Hypothesis: "Surface nutrient input, rather than physical habitat volume, limits species richness in these communities." Prediction: richness correlates with measures of surface-derived nutrients across sites with similar geology. Test: survey multiple caves, measure nutrients and habitat volume, and compare competing models. This is the kind of priority question that field-wide horizon scanning was designed to surface [2].

**Example 2, sensory biology.** Question: "Why do animals possess so many photoreceptor molecules, including outside the eyes?" Hypothesis: "Receptor diversity is an adaptation to the spectral and intensity complexity of each species' light niche." Prediction: species whose niches mix solar and lunar light should use biochemically distinct receptors to decode them. Test: comparative work in the lab and field on the same species. The marine annelid *Platynereis dumerilii*, which times its reproduction to lunar phases, illustrates how different photoreceptors can decode solar versus lunar light [6].

**Example 3, human genetics.** Question (inverted): "What phenotypes, if any, are associated with genomic variant V in the people who carry it?" Hypothesis: "Carriers of variant V share an unappreciated clinical feature." Test: reverse phenotyping of a recontactable, already-sequenced cohort, the genotype-first design described above [8].

**Example 4, collective behavior.** Hypothesis: "Schools of prey evade predators best when operating near a critical point (the parameter region where a system's behavior shifts qualitatively between regimes), because criticality maximizes responsiveness to attack." Test: a spatially explicit model of schooling prey under predator attack. Outcome: schools at criticality did perform best, but for a different reason than the hypothesis proposed, because of the spatial structure of the school rather than an optimal response of individual prey to the predator [15]. The test kept the pattern and replaced the mechanism, which is a normal and healthy outcome.

Each example pairs one question with one primary hypothesis and either a stated alternative or a described refuting result. That pairing is the transferable skill.

## Common Mistakes and Limitations

**Mistaking a topic for a question.** "Antibiotic resistance" is a topic. "Does sub-inhibitory antibiotic exposure increase mutation rate in biofilm-dwelling bacteria?" is a question.

**Confusing the prediction with the hypothesis.** "Warmed embryos hatch sooner" is a prediction. The hypothesis is the explanation that makes the prediction expected.

**Writing an unfalsifiable hypothesis.** If every conceivable result could be described as consistent with your statement, it cannot be tested. Add a direction, a magnitude, or a mechanism.

**Reading non-significance as equivalence.** An underpowered study has a low probability of detecting a true difference, so a non-significant result can coexist with a real, practically meaningful effect [4]. Plan statistical power before data collection and interpret the absence of significance with the study's power in mind.

**Marrying one hypothesis.** Researchers who favor a framework can under-invest in its rivals. The epigenetics and transposon survey found enthusiasm for one framework tracking inversely with attention to its alternative across disciplines [7]. State the leading alternative and design the study so the data can discriminate between them.

**Presenting exploratory results as confirmatory.** Hypothesis-free screens and omics surveys generate candidate findings that require dedicated follow-up tests. Treating a screen hit as a conclusion repeats the equivalence error in a new form.

**Overstating hypothesis maturity.** A hypothesis can rest on direct experimental support, indirect support, or speculation. Reporting should say which, because components of a popular hypothesis can be well supported while the full causal chain remains undemonstrated [10].

**Limitations.** The worked examples here are simplified for teaching. Real projects add power analyses, ethical review, preregistration where appropriate, and pilot data, and the right balance of question-first versus hypothesis-first framing varies by field, journal, and career stage. When a specific project carries real stakes, for example a thesis committee or a funding panel, get feedback on the question and hypothesis before committing resources.

## Frequently Asked Questions

### Can a study have both a research question and a hypothesis?

Yes, and most well-designed studies do. The question frames the gap in knowledge, and the hypothesis commits the study to a testable answer within that frame. Papers typically state the question early in the introduction and the hypothesis at the end of it, immediately before the methods.

### Does a hypothesis always take an if-then form?

No. The if-then format is a teaching shortcut, not a rule. Many hypotheses read as direct claims, such as "receptor diversity tracks light-niche complexity," with the if-then structure living in the predictions instead.

### What makes a good biology research question?

A good biology research question names a specific system, a measurable variable or process, and a context, and it can be answered with data you can realistically obtain. Methods textbooks summarize these properties as feasible, interesting, novel, ethical, and relevant.

### Do exploratory studies like omics screens need a hypothesis?

No. Discovery approaches such as genetic screens and transcriptomics are often deliberately hypothesis-free, and they are intrinsically hypothesis-generating, producing candidate hypotheses that later, dedicated experiments can test.

### What is the difference between a prediction and a hypothesis?

A hypothesis is the proposed explanation. A prediction is the specific, observable outcome that explanation implies. Confirming a prediction supports the hypothesis but does not prove it, because other mechanisms can produce the same prediction.

### How many hypotheses should one experiment test?

One primary hypothesis, stated before data collection, with the main alternative made explicit. Secondary hypotheses are acceptable if the design and statistical plan account for them, but every added hypothesis dilutes attention and increases the risk of spurious findings.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Can a study have both a research question and a hypothesis?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes, and most well-designed studies do. The question frames the gap in knowledge, and the hypothesis commits the study to a testable answer within that frame. Papers typically state the question early in the introduction and the hypothesis at the end of it, immediately before the methods."
      }
    },
    {
      "@type": "Question",
      "name": "Does a hypothesis always take an if-then form?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No. The if-then format is a teaching shortcut, not a rule. Many hypotheses read as direct claims, such as receptor diversity tracks light-niche complexity, with the if-then structure living in the predictions instead."
      }
    },
    {
      "@type": "Question",
      "name": "What makes a good biology research question?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "A good biology research question names a specific system, a measurable variable or process, and a context, and it can be answered with data you can realistically obtain. Methods textbooks summarize these properties as feasible, interesting, novel, ethical, and relevant."
      }
    },
    {
      "@type": "Question",
      "name": "Do exploratory studies like omics screens need a hypothesis?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No. Discovery approaches such as genetic screens and transcriptomics are often deliberately hypothesis-free, and they are intrinsically hypothesis-generating, producing candidate hypotheses that later, dedicated experiments can test."
      }
    },
    {
      "@type": "Question",
      "name": "What is the difference between a prediction and a hypothesis?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "A hypothesis is the proposed explanation. A prediction is the specific, observable outcome that explanation implies. Confirming a prediction supports the hypothesis but does not prove it, because other mechanisms can produce the same prediction."
      }
    },
    {
      "@type": "Question",
      "name": "How many hypotheses should one experiment test?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "One primary hypothesis, stated before data collection, with the main alternative made explicit. Secondary hypotheses are acceptable if the design and statistical plan account for them, but every added hypothesis dilutes attention and increases the risk of spurious findings."
      }
    }
  ]
}
</script>

## Sources

1. [The Spirogyra pratensis genome illuminates the evolution of developmental programs and spiral chloroplast biology.](https://pubmed.ncbi.nlm.nih.gov/42520407/)
2. [Fundamental research questions in subterranean biology.](https://pubmed.ncbi.nlm.nih.gov/32841483/)
3. [Fifteen challenges for generative AI applications to cell biology.](https://pubmed.ncbi.nlm.nih.gov/42607668/)
4. [Underpowered studies in muscle metabolism research: Determinants and considerations.](https://pubmed.ncbi.nlm.nih.gov/39461591/)
5. [Versatile system cores as a conceptual basis for generality in cell and developmental biology.](https://pubmed.ncbi.nlm.nih.gov/39236709/)
6. [Guiding Light: Mechanisms and Adjustments of Environmental Light Interpretation with Insights from Platynereis dumerilii and Other Selected Examples.](https://pubmed.ncbi.nlm.nih.gov/39932759/)
7. [Transposon dynamics and the epigenetic switch hypothesis.](https://pubmed.ncbi.nlm.nih.gov/34919173/)
8. [Genotype first: Clinical genomics research through a reverse phenotyping approach.](https://pubmed.ncbi.nlm.nih.gov/36608682/)
9. [Steps to individuality in biology and culture.](https://pubmed.ncbi.nlm.nih.gov/36688387/)
10. [The potential role and research progress of the gut microbiota-NGF axis in postmenopausal bone homeostasis imbalance.](https://pubmed.ncbi.nlm.nih.gov/42819214/)
11. [Big data and experimental biology: the complementary roles of hypothesis-led and blue skies research.](https://pubmed.ncbi.nlm.nih.gov/39287119/)
12. [Combining hypothesis- and data-driven neuroscience modeling in FAIR workflows.](https://pubmed.ncbi.nlm.nih.gov/35792600/)
13. [Multi-modal brain data exploration software for imaging transcriptomics: A comparative evaluation and future directions.](https://pubmed.ncbi.nlm.nih.gov/41614049/)
14. [Systematic Bayesian posterior analysis guided by Kullback-Leibler divergence facilitates hypothesis formation.](https://pubmed.ncbi.nlm.nih.gov/36335999/)
15. [Collective predator evasion: Putting the criticality hypothesis to the test.](https://pubmed.ncbi.nlm.nih.gov/33720926/)