03 · What You Need to Know
Find the Hidden Premises Underneath Your Proposed Study
An assumption is not automatically a flaw. Research cannot proceed without some premises about the phenomenon being studied, the evidence that can represent it, and the conditions under which the investigation will take place.
The problem is an assumption that is both consequential and insufficiently examined.
Methodological guidance for comparative effectiveness research recommends identifying major assumptions that affect how the research problem is conceptualized but will not necessarily be examined directly in the study. Making such assumptions explicit helps reviewers assess how they could influence the findings. Research-design guidance likewise emphasizes alignment among the research question, underlying assumptions, methodology, and methods.
A useful starting question is deceptively simple:
“For this study to produce a defensible answer, what needs to be true?”
Do not stop with one answer.
Some Assumptions Concern the Phenomenon Itself
Your research idea may depend on a particular understanding of how the phenomenon works.
Suppose you hypothesize that frequent use of an AI writing assistant influences students' writing development. The study may implicitly assume that AI use is sufficiently stable or meaningful as a phenomenon to measure, that different patterns of use can be represented adequately, and that the proposed mechanism connecting AI use with writing development is plausible.
These are conceptual or theoretical assumptions. They shape what you ask, what variables or experiences you consider relevant, and how you interpret the evidence.
A conceptual framework can help make these premises visible. In causal research, tools such as directed acyclic graphs can be particularly useful because they require researchers to represent assumed causal relationships explicitly rather than leaving them buried in prose.
You May Be Assuming That Your Construct Can Be Represented by Your Measure
Consider a study about “student engagement.” What evidence will represent engagement?
Attendance? Time spent in a learning management system? Number of discussion posts? A self-report scale? Classroom observation?
Each operationalization contains assumptions about how the underlying construct becomes observable. A student can spend considerable time logged into a platform without being cognitively engaged. Another can participate very little in an online discussion while thinking deeply about the material.
This does not make those measures useless. It means the reasoning connecting construct to measurement should be defensible.
Ask:
“What am I claiming this measure represents, and what evidence allows me to make that claim?”
This distinction becomes especially important when researchers quietly substitute an easily measured proxy for a harder construct. Measuring satisfaction is not automatically measuring learning. Publication count is not automatically research quality. Platform activity is not automatically engagement.
You May Be Assuming That the Data You Need Actually Exist in Usable Form
A proposed study using existing data can appear highly feasible because data collection has already occurred. Yet the proposal may depend on several unverified assumptions: that the relevant variables were recorded, that definitions remained consistent, that missingness is manageable, that records can be linked accurately, and that access will actually be granted.
“The institution has the data” is therefore not the same as “the study has usable data.”
Where existing records are central to the project, inspect documentation, data dictionaries, sample records, completeness reports, or other available evidence before treating data availability as established. If substantial uncertainty remains, decide what poor-quality or incomplete data would mean for the proposed study.
You May Be Assuming That the Participants You Need Can Be Recruited
Sample-size calculations can create a reassuringly precise target: perhaps the study requires 240 participants. That number says little about whether 240 eligible people can actually be enrolled and retained.
Recruitment assumptions concern the size of the accessible population, eligibility rates, willingness to participate, gatekeeper cooperation, response rates, attrition, and the time required to reach the target.
Research-planning guidance recommends justifying assumptions used in sample-size estimation and anticipated recruitment rather than treating them as self-evident. Preliminary information may come from previous studies, institutional records, pilot work, local audits, or other relevant evidence.
If your study needs 240 completed cases, ask how many people must realistically be approached to produce them. Then examine what happens if recruitment is substantially lower than expected.
You May Be Assuming That the Study Will Be Implemented as Designed
Interventions and field procedures rarely implement themselves.
A classroom intervention may assume that teachers follow the protocol with reasonable consistency. A longitudinal study may assume that participants complete repeated assessments. A multisite project may assume that data collection procedures are sufficiently comparable across locations. A technology study may assume that participants actually use the system in the way the intervention requires.
When implementation matters to interpretation, it should not remain invisible.
Suppose an intervention study finds no meaningful difference between groups. One explanation is that the intervention does not work. Another is that participants barely used it. Without information about implementation, those explanations may be difficult to distinguish.
This is why assumptions about exposure, adherence, fidelity, participation, and implementation may need to become measured features of the study rather than background expectations.
You May Be Assuming That Your Preferred Method Is Possible
A research idea can become tightly attached to a particular method long before practical constraints are investigated.
You may assume that random assignment will be permitted, that schools will allow classroom observations, that interviews about a sensitive topic can be conducted safely, that a proprietary instrument can be licensed, or that an external platform will provide the required data.
Some assumptions can be verified with a phone call, a policy check, a preliminary data request, or consultation with the relevant institution. It is generally better to discover early that an assumption is false than after the entire proposal has been optimized around it.
For dependencies that remain uncertain, determine whether a defensible alternative exists if the preferred method cannot be used.
Quantitative Analyses Can Introduce Their Own Assumptions
Statistical methods rely on assumptions too, although the relevant assumptions depend on the method and model.
These can concern matters such as independence, distributional form, functional relationships, variance structure, measurement, missing-data mechanisms, or features of the data-generating process. Sample-size calculations also require inputs such as an anticipated effect size, variability, event rate, attrition rate, or other design-specific quantities.
The important point is not to memorize a generic list of statistical assumptions and paste it into every proposal. Determine which assumptions apply to the analysis you actually plan to conduct and which are consequential for the inference.
Where possible, justify planning assumptions using prior evidence or preliminary data. Statistical methodology guidance specifically recommends making assumptions used for sample-size estimation explicit and supporting them appropriately.
Qualitative Research Also Has Assumptions
Assumptions are not a peculiarity of quantitative research.
Qualitative research may involve explicit ontological and epistemological assumptions concerning what can be known about a phenomenon and how that knowledge can be produced. Those assumptions influence the choice of methodology, sampling, data generation, analysis, interpretation, and the role of the researcher.
Research-design guidance therefore emphasizes coherence among the research problem, philosophical assumptions, methodology, and specific methods. A researcher should not select interviews, ethnography, phenomenology, grounded theory, or another approach merely because the technique is familiar. The approach should make sense for the kind of knowledge the research question seeks.
In mixed-methods research, assumptions and methodological commitments also matter when determining how qualitative and quantitative components relate to one another and what claims their integration can support.
Assumptions Differ in How Dangerous They Are
Once assumptions have been identified, do not treat all of them as equally important.
| Type of assumption |
Example |
If it is wrong |
| Low-consequence working assumption |
Recruitment advertisements will receive slightly more responses in the first weeks |
The timeline may need minor adjustment |
| Manageable assumption |
Attrition will remain near the rate observed in similar studies |
Additional recruitment or an adjusted retention strategy may be needed |
| Interpretive assumption |
A measure adequately represents the construct being discussed |
The meaning of the findings may need to be narrowed or reconsidered |
| Design-critical assumption |
The comparison groups provide a defensible basis for the intended inference |
The central conclusion may not be supported |
| Feasibility-critical assumption |
The required records can legally and practically be accessed |
The planned study may not be executable |
| Contribution-critical assumption |
An important uncertainty remains unresolved in the existing evidence |
The study may add too little to justify conducting it |
The assumptions that deserve the most attention are those with both substantial uncertainty and substantial consequences. An assumption that is almost certainly true may require little additional work. An uncertain assumption that would merely inconvenience the project may need a contingency plan. An uncertain assumption whose failure makes the research question unanswerable deserves immediate investigation.
An Assumption Is Not the Same as a Hypothesis
Assumption
A proposition the study relies on sufficiently for its design, reasoning, execution, or interpretation, but which may not itself be the primary proposition being tested.
Hypothesis
A specific, testable prediction or proposition that the study is designed to evaluate with evidence.
The distinction can sometimes be context-dependent. Something treated as an assumption in one study could become an empirical question in another. If the truth of a particular assumption is sufficiently uncertain and consequential, you may decide that it should no longer remain an assumption at all. It may need to be measured, tested preliminarily, or incorporated directly into the research design.
04 · A Practical Example
Unpacking the Assumptions Behind a Simple Research Idea
Hypothetical Example
Does an AI Feedback Tool Improve Academic Writing?
Suppose a researcher proposes a study comparing university students who receive AI-generated formative feedback on draft essays with students who receive the usual feedback process. The primary outcome is the quality of students' final academic writing.
The proposal sounds straightforward. Now remove the assumptions one by one.
Conceptual assumption The researcher assumes that the intervention provides a form of feedback that could plausibly influence the writing processes relevant to the intended outcome. If the system mostly corrects surface-level language while the outcome emphasizes argumentation and evidence, the proposed mechanism may be weaker than assumed.
Measurement assumption The researcher assumes that the scoring procedure can represent differences in academic writing quality with sufficient validity and consistency. If scoring largely captures grammar while the study makes claims about academic writing more broadly, the inference becomes questionable.
Implementation assumption The design assumes that students actually read, understand, and use the AI-generated feedback. Merely giving students access does not establish meaningful exposure to the intervention.
Comparison assumption The researcher assumes that differences between groups can reasonably be attributed to the intervention under the chosen design. If the groups differ systematically in prior writing ability, instructor support, or motivation, the intended interpretation may be weakened.
Recruitment assumption The sample-size plan assumes that enough eligible students will enroll and complete the study. Institutional enrollment figures alone do not establish participation rates.
Analytical assumption The planned statistical analysis has assumptions appropriate to the model being used, while the sample-size calculation relies on anticipated values that need defensible justification.
Decision The researcher does not abandon the idea. Instead, the assumptions are converted into design work: examine prior evidence, refine the outcome measure, collect information about actual tool use, strengthen the comparison, obtain realistic recruitment estimates, and plan appropriate analytical checks.
The exercise changes the question from “Is this a good idea?” to something more useful: “What conditions must hold for the evidence from this study to support the conclusion I want to draw?”
06 · What This Means for You
Turn Hidden Assumptions Into Things You Can Examine
Take your research idea and write one sentence beginning with:
“For this study to work as intended, I am relying on...”
Complete it repeatedly. Think about the theory, population, access, measures, data, intervention or procedure, comparison, analysis, timeline, and contribution. Then examine each statement using two questions: How uncertain is this assumption? What happens if it is wrong?
A simple decision framework
If an assumption is well supported and failure would have little consequence
Document it where relevant and proceed without allowing it to consume disproportionate attention.
If an assumption is uncertain but its failure is manageable
Create a contingency plan and identify what evidence would signal that the plan is needed.
If an assumption can be verified before the study
Check it now through literature, preliminary data, institutional records, consultation, feasibility work, or another appropriate source.
If an assumption can be measured during the study
Consider making it observable rather than simply taking it for granted.
If an assumption is required only because of your preferred design
Ask whether another design could answer the research question with less dependence on that uncertain condition.
If an assumption is highly uncertain and the study fails without it
Treat it as a critical vulnerability that should be resolved before substantial investment.
Look for Assumption Chains
Assumptions rarely operate independently.
For example, a study may assume that a digital trace represents engagement. That assumption supports the use of the trace as an outcome. The outcome then supports a comparison between groups. That comparison supports a conclusion about whether an intervention improves engagement.
If the first link is weak, later analytical sophistication cannot restore the missing conceptual connection.
Tracing these chains can help you identify the assumption with the greatest leverage over the rest of the study.
Ask What Evidence Would Make You Change the Study
A useful assumption audit should lead to decisions, not merely a longer proposal.
If preliminary inspection reveals that the primary outcome is missing for 40% of records, what will you do? If recruitment estimates fall far below what the design requires, what changes? If your chosen instrument performs poorly in the target population, will you replace it? If the theoretical premise turns out to have much weaker empirical support than expected, does the research question change?
This connects assumption testing to the broader question of what the strongest argument against conducting the proposed study might be. Sometimes the strongest objection is simply an assumption that everyone has been treating as a fact.
Watch Out
Do not respond to an uncertain assumption by quietly making your claim broader than the evidence permits. If the study can only support conclusions under particular conditions, state those conditions. If an assumption cannot be adequately defended, redesigning or narrowing the study is generally more defensible than hiding the dependency in a limitations paragraph at the end.
When a critical assumption cannot be supported, measured, relaxed, or designed around, it may become part of the decision about whether the research idea should be abandoned before further investment.
