Hypotheses, Questions and Investigation Constraints
From Curiosity to a Testable Question
Every statistical enquiry starts with someone noticing something interesting. A teacher wonders whether students who sleep longer do better in tests. A council wants to know whether recycling rates differ between flats and houses. A shop owner suspects that younger customers spend more online. Curiosity is the spark — but statistics needs more than a spark. It needs a question that can actually be tested with data.
A hypothesis is that testable version. It is not a question — it is a statement that names the variables involved and predicts how they are related.
- Vague idea: “Does exercise affect mood?”
- Testable hypothesis: “People who exercise for 30 minutes or more per day will report higher wellbeing scores than people who exercise for less than 30 minutes per day.”
The second version tells you exactly what to measure (exercise time and wellbeing score), how to split your groups, and what a supporting result would look like. Everything that follows in an investigation — what data to collect, how to sample it, and how to judge the result — depends on getting this first step right.
A hypothesis must name the variables and state a predicted direction or difference. The exam board does not require a formal null hypothesis — a clearly worded statement is enough.
Why Only Data Can Test a Hypothesis
It is tempting to think a convincing argument proves a hypothesis right. It does not. You can reason your way to a conclusion that sounds perfectly sensible and still be wrong once real data is collected. A hypothesis can only be tested through the appropriate collection and analysis of data — opinion and intuition can suggest a hypothesis, but they cannot confirm one.
This is exactly why the design stage of an investigation matters so much. A biased sample, a badly worded question, or a measurement error at the collection stage means any conclusion drawn later — however confidently stated — is built on shaky ground.
The Constraints on Real Investigations
No investigation happens in ideal conditions. You need to be able to identify and discuss five practical limits: time constraints, cost constraints, ethical issues, confidentiality, and convenience.
Time: a two-week survey gives a different picture than a year-long one. Investigating something seasonal — hot drink sales, hay fever cases, holiday bookings — with only one month of data risks missing the pattern entirely.
Cost: a large random sample is statistically better, but interviewing hundreds of people in person costs far more than an online form sent to the same number. Budgets often force a smaller sample or a cheaper (and sometimes less reliable) method.
Ethical issues: an investigation must not embarrass, distress, or harm participants. Topics like income, mental health, or family circumstances need particular care, informed consent, and a judgement about whether the value of the findings outweighs any risk.
Confidentiality: people who fear their answers could be traced back to them are more likely to refuse to take part, or to answer dishonestly. Guaranteeing anonymity usually raises both the response rate and the honesty of responses.
Convenience: the theoretically ideal population is not always reachable. Surveying a genuinely random sample of every teenager in the country is logistically enormous, so real researchers work with who they can access.
“Is there a link between screen time and sleep?” is a reasonable research question — but it is not a hypothesis, because it does not predict anything.
A hypothesis makes a specific, directional claim: “Students who use screens for more than three hours each evening will sleep fewer hours than students who use screens for less than one hour.” Now the variables, groups, and expected direction are all clear.
If an exam question asks you to write a hypothesis and you write a question mark at the end, you have not answered it. Check your answer is a statement, not a question.
Planning for Problems Before They Happen
Careful investigators do not just hope nothing goes wrong — they plan for the problems that commonly occur, including difficulty identifying the population, non-response, and unexpected outcomes in the data.
Identifying the population: some populations are obvious — all Year 10 students at a school. Others are fuzzy: what exactly counts as a “frequent gamer”? Once a week? Every day? Defining the population precisely before collecting anything prevents arguments later about who should have been included.
Non-response: if a large share of your sample ignores the questionnaire, your results may no longer represent the population, because non-responders often differ systematically from responders — people with strong views tend to reply more than people with mild ones. Shorter questionnaires, reminders, and anonymity all reduce this risk.
Unexpected outcomes: real data throws up surprises. An unusually large value might be a genuine finding, a recording error, or a sign the original hypothesis needs revising. Building in a review stage — rather than assuming the plan will run perfectly — catches these problems early.
Piloting a questionnaire, guaranteeing anonymity, defining the population precisely, and planning a reminder for non-responders are all proactive strategies that stop small problems becoming reasons to distrust the whole investigation.
A student wants to find out whether a long commute makes students more likely to arrive late.
Hypothesis: “Students with a commute of more than 30 minutes will have a higher rate of late arrivals than students with a commute of under 30 minutes.”
Constraints identified:
- Time — data can only be collected for one half-term, so winter weather delays may not be captured.
- Ethical — lateness records are sensitive; parental consent is needed for under-16s.
- Non-response — some students may not return the travel-time form, so a reminder is planned in advance.