03 · What You Need to Know
When Research Needs Experiments, Hypotheses, Both, or Neither
An experiment is a particular research design
An experiment deliberately manipulates one or more conditions and examines resulting outcomes, typically using comparison conditions and design features intended to support causal inference. Random assignment, when feasible and properly implemented, can be particularly valuable because it helps make comparison groups similar with respect to both observed and unobserved characteristics before the intervention.
Experiments are therefore exceptionally useful for questions such as, "What is the effect of this intervention compared with that condition?" They can help researchers distinguish an intervention's effect from alternative explanations that would be difficult to separate through simple observation.
But this strength does not make experimentation the definition of research. Research encompasses a much broader family of systematic investigations, as reflected in the essential characteristics of research.
Many important research questions cannot be answered experimentally
Suppose researchers want to investigate whether exposure to a harmful environmental condition is associated with a serious disease. Deliberately assigning people to harmful exposure could be unethical. Researchers may instead use observational evidence and appropriate epidemiological methods.
Historical questions provide an even clearer case. A historian studying a political event from a century ago cannot randomly assign societies to alternative historical conditions. An astronomer cannot randomly reposition galaxies. Researchers studying natural disasters cannot schedule earthquakes for methodological convenience.
These limitations do not make the resulting work non-research. They mean the researcher needs a design appropriate to evidence that can actually be obtained.
Non-experimental does not mean unsystematic
A non-experimental study does not manipulate an exposure or treatment in the way an experiment does. Researchers may instead observe variables as they occur naturally, analyze records, compare groups, conduct surveys, study cases, or use existing datasets.
Such studies still require rigorous decisions about sampling or case selection, measurement, confounding, timing, analytical strategy, uncertainty, and interpretation. The threats to inference may differ from those in experiments, but methodological discipline remains necessary.
A poorly designed experiment does not become rigorous merely because something was manipulated. Conversely, a carefully designed observational study does not become weak merely because nothing was manipulated.
A hypothesis is a testable expectation, not a ceremonial sentence
In research, a hypothesis generally expresses an expectation or prediction that can be evaluated against evidence. It is often derived from theory, previous findings, observations, or earlier exploratory work.
For example, a researcher might hypothesize that students receiving a particular instructional intervention will demonstrate greater improvement on a defined outcome than students in a comparison condition. The study can then be designed specifically to evaluate that prediction.
A useful hypothesis therefore does intellectual work. It specifies what the researcher expects and creates the possibility that evidence could fail to support that expectation.
Writing "There is no significant relationship between X and Y" simply because a thesis template contains a heading called Hypothesis does not automatically strengthen a study. The hypothesis should follow from the logic of the research question and design.
Hypothesis testing does not require an experiment
Experiments and hypotheses are often taught together, which can make them seem inseparable. They are not.
An observational study can test a prespecified hypothesis. For example, researchers analyzing longitudinal data might hypothesize that a particular exposure predicts a later outcome after accounting for specified covariates. Nothing needs to be experimentally manipulated for the study to evaluate that prediction.
Similarly, a study using an existing dataset can test hypotheses that were specified before the relevant analysis. As explained in research using existing data, collecting the observations yourself is not a prerequisite for conducting a research study.
An experiment does not necessarily require a conventional statistical hypothesis test
The reverse distinction is also useful. Conducting an experiment describes how evidence is generated. Statistical hypothesis testing describes one family of approaches for making particular statistical inferences from data.
Experimental results can be analyzed and interpreted using different statistical frameworks and estimation approaches. The existence of experimental manipulation does not logically require that the study culminate in a null-hypothesis significance test or a particular threshold such as p <.05.
It is therefore helpful to keep three ideas separate: experimental design, substantive hypotheses, and statistical hypothesis tests. They frequently occur together, but they are not synonyms.
Exploratory research may begin without a specific hypothesis
Sometimes researchers do not yet know enough to formulate a well-grounded prediction. The purpose of the research may be to discover important patterns, identify relevant concepts, characterize a poorly understood phenomenon, or generate hypotheses for later testing.
The National Academies of Sciences, Engineering, and Medicine distinguishes exploratory research, which can generate hypotheses, from confirmatory research, which begins with a well-defined research question and a priori hypotheses. Both forms of inquiry contribute to science, but their results should be interpreted differently.
Exploratory research is not simply random searching. A strong exploratory study still requires a coherent purpose, appropriate evidence, systematic methods, and transparent reporting. It simply allows greater openness about what patterns or explanations may emerge from the investigation.
Confirmatory research benefits from specifying hypotheses before seeing the relevant results
When researchers claim to test a hypothesis, timing matters. A prediction constructed after seeing the result has not been tested by that same result in the same sense as a prediction specified beforehand.
This is particularly important in statistical inference. The National Academies notes that confirmatory statistical testing requires decisions about the hypothesis test to be made before examining the data on which it will be tested. If researchers explore many possible relationships and then present the most interesting one as though it had been predicted from the outset, conventional error-rate interpretations can become misleading.
Exploratory research
Investigates patterns, possibilities, concepts, or relationships and may generate hypotheses for subsequent testing.
Confirmatory research
Evaluates prespecified hypotheses or predictions using evidence capable of providing an informative test.
The distinction is useful, but it should not be exaggerated into two completely separate species of study. A project can contain both confirmatory and exploratory analyses. Researchers may test their planned hypotheses and then encounter an unexpected pattern worth exploring. The crucial requirement is transparency about which is which.
Exploratory findings do not become inferior findings
Calling an analysis exploratory is sometimes treated as an admission of weakness. That is unfortunate. Exploration is how researchers encounter unexpected relationships, develop concepts, improve theories, and formulate questions that later studies can test more directly.
The problem is not exploration. It is disguising exploration as confirmation.
An unexpected finding can be scientifically interesting precisely because it was unexpected. The appropriate response is to report how it arose, interpret it cautiously, and, when warranted, investigate whether it survives additional testing using independent evidence.
This relationship between discovery and testing is one reason confirmatory and replication studies remain important even after an intriguing initial result has been reported.
Descriptive research may need questions rather than hypotheses
Not every worthwhile research objective concerns whether one variable causes, predicts, or relates to another. Researchers may first need to establish what exists, how common something is, how it is distributed, how it changes over time, or what characteristics a phenomenon has.
A study estimating the prevalence of a condition in a defined population can be valuable without being organized around a prediction that one group will differ from another. A demographic survey may describe a population. An ecological study may document the distribution of a species. A bibliometric investigation may characterize patterns in a body of literature.
Such studies still require precise research questions and appropriate methods. A hypothesis should not be manufactured merely because the word research appears on the title page.
Qualitative research often uses research questions rather than predictive hypotheses
Many qualitative approaches seek to understand experiences, meanings, practices, processes, interactions, or interpretations. An interview study might ask how first-generation university students experience academic belonging. An ethnography might investigate how norms develop within an online community.
Predetermining the expected answer as a hypothesis can sometimes conflict with the purpose of an open-ended qualitative design. Researchers may instead work with research questions, sensitizing concepts, theoretical propositions, or evolving interpretations, depending on the methodology.
This does not mean qualitative research is hypothesis-free in every possible sense or that researchers enter a study without prior knowledge. Different qualitative traditions make different assumptions about theory, prior concepts, and the role of the researcher. The relevant point is narrower: formal predictive hypothesis testing is not a universal requirement for rigorous research.
Nor is the absence of conventional hypothesis testing evidence that the study is somehow less scientific by default. The related question of whether research must use statistics or can be entirely qualitative requires the same attention to methodological fit.
Historical and interpretive research may ask questions that are not naturally hypotheses
Consider a historian examining how educational policy changed during a particular political transition. The project may involve systematic archival research, source criticism, comparison of documents, contextual interpretation, and a carefully argued conclusion.
Forcing that investigation into a prediction such as "Policy X will significantly affect Y" would not make the scholarship more rigorous. It might simply distort the question.
Likewise, some research in the humanities, law, philosophy, and related fields depends heavily on textual interpretation, conceptual analysis, argumentation, archival evidence, or other forms of disciplined inquiry. The standards for evidence and inference differ from those of experimental science, but difference should not be confused with absence of rigor.
The research question should determine the design
A useful methodological principle is to work from the question toward the method rather than from a preferred method toward a convenient question.
| Research Purpose |
Possible Approach |
Experiment or Hypothesis Required? |
| Estimate how common a phenomenon is |
Descriptive survey or population study |
An experiment is unnecessary; a predictive hypothesis may also be unnecessary |
| Estimate the causal effect of an intervention |
Randomized experiment when ethical and feasible |
An experiment may be particularly valuable; prespecified hypotheses may be appropriate |
| Examine an association that occurs naturally |
Observational study |
No experiment is required; hypotheses may be tested |
| Understand participants' experiences or meanings |
Appropriate qualitative design |
Neither an experiment nor a conventional predictive hypothesis is inherently required |
| Explore a poorly understood phenomenon |
Exploratory study |
No experiment or prior hypothesis is inherently required |
| Test an established theoretical prediction |
Confirmatory study |
A prespecified hypothesis is central; an experiment may or may not be necessary |
| Investigate a historical question |
Archival or historical research |
Neither is inherently required |
These are examples rather than rigid rules. Real research questions can require more complicated designs, and methodological traditions differ in how they formulate and evaluate claims.
There is no single scientific method that forces every study into the same sequence
The familiar sequence of observation, hypothesis, experiment, analysis, and conclusion is useful for teaching one form of scientific reasoning. It becomes misleading when presented as the mandatory architecture of every legitimate study.
Scientific inquiry can move between observation, theory, exploration, modeling, prediction, measurement, experimentation, replication, and revision in different ways. The National Academies' public explanation of scientific reasoning likewise presents science as involving questioning, observation and experimentation, confirmation, and revision rather than as a guarantee-producing linear recipe.
The broader issue of whether one scientific method must govern all research therefore cannot be resolved by treating the experimental hypothesis-testing model as universal.
