Usually no formal H₀/H₁.
A prevalence, distribution or service-description question needs a clear population, measure, timeframe and uncertainty rather than an artificial null hypothesis.
Project planning · a good place to begin
Start with the question.
Turn a broad medical-science idea into an answerable epidemiological question and a brief that can guide the project, analysis and report.
Use this when: you have a topic but are unsure what you can actually ask.
You will make: one focused research question and a short analysis brief.
Then: find the evidence, appraise and save useful sources, and carry the question into the study design.
Your starting point
Begin with a problem, not a statistical test.
Bring a topic, clinical problem or uncertainty. You do not need a dataset, hypothesis or preferred method yet. The question must lead; software and analysis follow.
Before you start: write one sentence about what you want to understand and why it matters. It can be broad.
By the end you can
- distinguish a topic, aim, research question and hypothesis;
- frame a question with population, exposure or intervention, comparator, outcome and timeframe;
- save and trace the prior evidence that shapes the question;
- identify the design, population and variable decisions to take into the study-to-data foundation;
- decide whether a formal hypothesis is appropriate and whether the work is confirmatory or exploratory;
- complete a brief that travels into the project plan and analysis.
1 · Set the sequence
Which comes first: the research question or the hypothesis?
The short answer
Substantive problem or topic → focused research question → hypothesis when appropriate → analysis plan.
A hypothesis is a proposed, testable answer to a defined question. It only becomes testable when the population, exposure or intervention, comparator, outcome, timeframe and target quantity are clear.
- 01Substantive problem
State the medical or public-health uncertainty that matters.
- 02Focused question
Specify who, what is compared, which outcome and over what time.
- 03Hypothesis, if useful
Propose the pattern expected from theory or prior evidence.
- 04Analysis plan
Pre-specify the target quantity, method, assumptions and reporting approach.
Pre-specify before outcome results.
State the question, hypothesis and analysis plan before inspecting the outcome results. Do not rewrite a directional hypothesis after seeing which way the estimate points.
Label the purpose honestly.
Explore signals or assess predictive performance without retrofitting a confirmatory hypothesis. Findings need validation, not a post-hoc claim that they were predicted.
Theory or prior evidence may generate an initial hypothesis first. Treat it as provisional: translate it into a precise research question, operationalise every component and specify the analysis before testing it.
2 · Frame the question
Use PECO or PICO, then check whether the data can answer it.
Use PICO when an intervention is assigned and PECO when an exposure is observed. Add a clear timeframe in either case.
Population
Who is eligible, where are they observed and to whom should the answer apply?
Exposure or intervention
What characteristic, care or treatment is observed or assigned, and when?
Comparator
Compared with whom or what? State the reference group explicitly.
Outcome
Which primary outcome is measured, on what scale and by what definition?
Timeframe
When is time zero, and how long after exposure or intervention is the outcome assessed?
Among [population], is [exposure] compared with [comparator] associated with [outcome] over [timeframe]?
For a randomised intervention, causal wording may be defensible. For an observational exposure, begin with “associated with”.Name one primary question and target quantity. Keep secondary questions distinct, justified and labelled; do not create them because one result was disappointing.
Classify the exposure and outcome as categorical, continuous, count, ordered or time-to-event, then state exactly what will be estimated: for example prevalence, risk difference, risk ratio, mean difference, adjusted odds ratio or prediction error.
Confirm the design, time order, variable definitions, outcome and exposure types, follow-up, missingness and sample size. A precise question is still unusable if the available data cannot answer it.
3 · Worked transformations
See how the purpose changes the question, hypothesis and analysis.
How long are patients waiting?
- Broad topic
- Outpatient waiting times.
- Weak question
- Are waits too long?
- Precise question
- Among first appointments completed by one outpatient service in 2025, what was the median referral-to-appointment time and what proportion exceeded 18 weeks?
- Hypothesis
- No formal null or alternative hypothesis is required: the purpose is estimation and description.
- Variables and target
- Waiting time in weeks and a pre-defined over-18-weeks indicator; median and IQR plus a proportion with 95% confidence interval.
- Analysis family
- Frequencies, percentages, confidence interval and distribution summaries or plots.
Distinction: the question asks for a population quantity; it does not test whether that quantity equals an arbitrary value.
Hearing-aid use and social isolation.
- Broad topic
- Hearing aids and social connection.
- Weak question
- Do hearing aids stop loneliness?
- 12-month data-fit check
- “Is hearing-aid use associated with isolation at 12 months?” is clear, but a dataset with only six-month follow-up cannot answer that timeframe.
- Answerable question
- Among adults with measured hearing loss, is regular hearing-aid use at baseline, compared with non-use, associated with social-isolation status six months later after accounting for pre-specified baseline differences?
- H₀ / H₁
- H₀: the adjusted association is null (odds ratio = 1). H₁: it is non-null (odds ratio ≠ 1). This is deliberately non-directional and non-causal.
- Variables and target
- Exposure: regular hearing-aid use; comparator: non-use; binary six-month isolation; plausible baseline confounders such as hearing severity, prior isolation, age and social support; target: adjusted odds ratio with 95% CI.
- Analysis family
- Descriptive follow-up proportions, crude measures, then pre-specified logistic regression with diagnostics and cautious interpretation.
Distinction: the question defines the association to estimate; H₁ proposes a testable pattern but does not turn an observational comparison into a causal effect.
Walking speed after an exercise programme.
- Broad topic
- Exercise and mobility.
- Weak question
- Are the groups different?
- Precise question
- Among adults randomised to an eight-week exercise programme or usual activity, what is the programme-minus-usual-activity mean difference in walking speed at eight weeks?
- H₀ / H₁
- H₀: mean difference = 0. Non-directional H₁: mean difference ≠ 0. Use a directional H₁ only when prior evidence justifies it and it is pre-specified before outcomes are inspected.
- Variables and target
- Randomised group; continuous walking speed; target: baseline-adjusted mean difference with 95% CI.
- Analysis family
- Group summaries, plots and a pre-specified linear model including baseline walking speed.
Distinction: the research question asks how large and uncertain the difference is; the hypothesis only states the null model, so effect size and confidence interval come before the p-value.
Who may have an emergency admission?
- Broad topic
- Predicting hospital admission.
- Weak question
- Which variables are significant predictors?
- Precise question
- Among adults discharged from one hospital, how well can routinely available discharge variables predict emergency admission within six months when performance is assessed with internal validation?
- Hypothesis
- Do not retrofit confirmatory H₀/H₁ after viewing the results. Pre-specify the prediction target, candidate predictors, modelling strategy and performance measures.
- Variables and target
- Binary or time-to-admission outcome; baseline predictors; target: out-of-sample calibration and discrimination, with optimism quantified.
- Analysis family
- Logistic or survival prediction model with resampling or bootstrap internal validation, calibration and discrimination. Do not use predictor p-value screening.
Distinction: prediction asks how accurately outcomes can be forecast for new participants; it does not automatically explain causes or validate a post-hoc hypothesis.
4 · Beginner self-check
Is this a topic, aim, research question or hypothesis?
Decide before opening each answer.
“Hearing aids and social isolation.”
Topic. It names an area, but not a population, comparison, outcome definition or timeframe.
“To estimate the association between baseline hearing-aid use and six-month social isolation.”
Aim. It states the project’s purpose. Objectives would break this into actions such as defining the cohort, estimating crude risks and fitting a pre-specified adjusted model.
“Among adults with measured hearing loss, is regular hearing-aid use, compared with non-use, associated with social isolation at six months?”
Research question. It specifies population, exposure, comparator, outcome and timeframe, while keeping observational language cautious.
“After pre-specified adjustment, the hearing-aid-use odds ratio for six-month social isolation differs from 1.”
Alternative hypothesis. It proposes a testable, non-directional answer to the defined question and names the target scale.
5 · Your reusable output
Complete the research question and analysis brief.
Copy these headings into your project notes. Keep the brief short, date important changes and never quietly rewrite a confirmatory plan after seeing outcome results.
- Topic or substantive problem
- What matters, and why?
- Prior evidence
- The key sources found through a documented search, critically appraised, then saved with that judgement in your reference library.
- Primary research question
- One answerable PECO/PICO-style question.
- Aim and objectives
- The overall purpose, then the concrete steps needed to answer it.
- Question status
- Primary or secondary; descriptive, confirmatory, exploratory or prediction.
- Hypothesis, if appropriate
- H₀ and H₁; directional only when justified and pre-specified.
- Population
- Eligibility, setting and intended population of inference.
- Study design
- How participants, exposure or intervention and outcomes are observed or assigned.
- Exposure or intervention
- Definition, timing, coding and reference level.
- Comparator
- The group or condition that defines the comparison.
- Outcome
- Primary outcome, type, definition and measurement.
- Timeframe
- Time zero, follow-up and outcome assessment point.
- Plausible confounders
- Pre-exposure common causes justified from subject knowledge; not selected by bivariate p-values.
- Target quantity or effect measure
- The exact prevalence, difference, ratio, association or prediction-performance measure.
- Proposed analysis family
- Descriptive summaries, comparison, regression or prediction approach matched to the question and outcome.
- Assumptions and diagnostics
- Design and model conditions that must be assessed.
- Known limitations
- What the data, design, measurement or timeframe cannot establish.
A proposed analysis family is not a final command. It records the reasoning to test in the methods decision log as the project develops. When adjustment is needed, use the choosing covariates methods note to build and justify the set rather than screening variables by bivariate p-values.
6 · Carry it forward
One brief becomes the project’s golden thread.
Question-and-analysis brief
Defines the question, target quantity, design and planned analytical family before outcome interpretation.
Methods decision log
Records dated decisions, reasons, deviations, data checks, assumptions and diagnostics as the analysis develops.
Interpretation record
Captures estimates, confidence intervals, assumptions, diagnostics, alternative explanations and limitations before prose smooths them away.
Use the question and purpose from the brief.
Use PECO/PICO, design, variables, estimand, planned analysis and the decision log.
Use the pre-specified question, estimates, uncertainty and relevant diagnostics from the interpretation record.
Answer the question cautiously, consider alternatives and explain assumptions and limitations.
Optional support · evidence, project planning and statistics
See how a study becomes dataConnect the question to design, population, variables, missingness and potential sources of bias.
Plan a survey from the question onwardChoose the mode, variables, questions, sample, fieldwork and processing as one connected design.
Carry the brief into the project guidanceShape aims, objectives, Methods and the decision trail.
Save the evidence as you readBuild a verified EndNote or Zotero library and leave a useful note with every important source.
Open the statistics deskChoose Stata, SPSS or RStudio when software is unfamiliar.
Go deeper · framing, estimands and reporting standards
Optional authoritative sources for fuller detail. They extend the guide; they are not prerequisites for completing the brief.
Define the review question
PICO components, outcomes and the difference between broad objectives and answerable questions.
Open Cochrane Handbook Chapter 2Define the treatment effect of interest
Formal guidance on aligning population, outcome, treatment conditions and the target quantity in clinical trials.
Open ICH E9(R1) (PDF)Plan transparent reporting
Design-specific expectations for cohort, case-control and cross-sectional studies.
Open STROBE checklistsDevelop and validate transparently
Reporting guidance for clinical prediction-model development and validation.
Open TRIPOD