An IA is marked on the quality of its design and analysis, not on whether the result came out. The marks are lost before you collect anything: a question too broad to answer, no control for the obvious confounder, a sample that cannot support the claim, and an evaluation written as apology rather than analysis.
Students treat the IA as a writing task with an experiment attached. It is the reverse: most of the marks are decided at the design stage, weeks before you write anything.
Where the marks actually go
Across subjects the criteria differ in wording and agree in substance. You are assessed on:
- whether the question is focused enough to answer
- whether the method can actually answer it
- whether the analysis is appropriate to the data
- whether the evaluation identifies what genuinely limits the conclusion
Note what is not on that list: whether the result came out. The same principle governs academic publishing — the joint COPE, DOAJ, OASPA and WAME principles are built around process integrity rather than whether a finding is convenient. A well-designed study producing a null result, honestly interpreted, scores better than a confounded study that happened to confirm the hypothesis. Students do not believe this and it is true.
The four decisions that decide your grade
1. The question, named in variables
If your question does not contain the specific things being varied and measured, the design marks have nothing to attach to.
| Weak | Why it fails | Stronger |
|---|---|---|
| "How does temperature affect enzymes?" | No enzyme, no measure, no range | "How does temperature between 20 and 60°C affect the initial rate of catalase activity in potato tissue?" |
| "Does social media affect wellbeing?" | Neither variable is measurable as stated | "Does self-reported daily screen time predict self-reported sleep duration in the 2023 Monitoring the Future sample?" |
| "Is the river polluted?" | Not a comparison | "Does nitrate concentration increase downstream of the sewage outflow relative to upstream?" |
The test is the same one that applies to any research question: can you say in one sentence what evidence would change your answer? Our method for narrowing between candidate questions scores two options in an hour.
2. The confounder you will be asked about
Before you collect anything, name what else could produce your expected result.
Measuring river nitrate downstream of an outflow? Downstream is also further from the source, lower, slower and warmer. Comparing two schools' exam results? They also differ in intake. Testing a plant growth variable on a windowsill? Light intensity varies with position.
A design that controls for the obvious confounder scores. A design that ignores it loses marks twice — once in design, again in evaluation when you fail to mention it.
3. The sample, honestly sized
The commonest hard limit on an IA's conclusion.
Three repeats is not a sample. Twenty classmates is not a population. Students know this and proceed anyway because the alternative is more work, and then write a conclusion the data cannot support.
Two legitimate ways out:
Collect more. More repeats, more sites, more respondents. Dull and effective.
Use secondary data, where your subject permits it. Economics, geography, psychology and environmental systems IAs can often be built on public datasets with samples in the thousands. Our list of free datasets covers twenty-two sources. Check your subject's requirements first, because not every course allows it.
If you genuinely cannot increase the sample, the third option is to quantify the uncertainty and say what it means — which is worth more marks than pretending the problem is not there.
4. The evaluation, written as analysis not apology
This is where most marks are lost, and the fix is a change of register.
Most students write: "I could have been more accurate with the measuring cylinder. If I did this again I would repeat it more times."
What scores: "The downstream sites were also 4°C warmer, and nitrate solubility rises with temperature, so part of the increase I attributed to the outflow may be temperature-driven. Measuring at matched temperatures, or sampling both sites at dawn, would separate the two."
The difference is that the second one says which weakness affects which part of the conclusion and how. That is analysis. The first is contrition, and contrition is not on the mark scheme.
By subject group
The design move differs, and generic IA advice ignores this.
Group 3 — Individuals and societies. Secondary data is usually your friend. The strongest economics and geography IAs take a public dataset and ask something local of it. Beware questions about attitudes that would need a survey you cannot run properly.
Group 4 — Sciences. The constraint is equipment, and it is worth knowing early that most universities bar under-eighteens from BSL-2, radioactive and animal work, so anything requiring an outside lab is likely closed. Design within what your school has, or go computational — research without a lab covers the fields where that works.
Group 5 — Mathematics. The risk is restating textbook material. The exploration needs something of your own: a dataset you analysed, a model you built, a result you tested rather than quoted.
Psychology. Ethics is a design constraint, not a formality. Yale asks applicants to state their specific contributions where work was collaborative, and the same precision about your own role applies to how you describe a study involving other people. Anything involving deception, distress or vulnerable participants will not be approved, and designing around that before you commit saves a rewrite.
The sequence that works
- Pick the dataset or method you can actually access. Not what would be ideal. What you can start on Monday.
- Write the question with variables named. Twenty minutes.
- Name the confounder and design the control. This is the hour that decides your grade.
- Run a pilot. Three measurements, or one afternoon of data. Pilots reveal the problem you did not anticipate, and finding it now costs nothing.
- Collect properly. More than you think you need.
- Analyse before you write. Know what you found before you start describing it.
- Write the evaluation as analysis.
Step four is the one students skip and the one that most reliably pays for itself.
What this has to do with everything after
The IA is the first time most students design research rather than follow a recipe, and the skill transfers directly.
The same four decisions govern an Extended Essay, an EPQ, an AP Research paper and any journal submission. The scale changes; the design discipline does not.
If you later want to take an IA further, our guide to publishing an Internal Assessment is honest about the obstacle — which is almost always sample size, decided at the design stage covered here.
Where we fit
The Oxford Centre for Advanced Research pairs students with PhD candidates and early-career researchers at Oxford, Cambridge, other leading UK universities and Ivy League institutions. The Oxford Scholar Programme is £2,000 — about $2,600 — for ten contact hours over ten to fourteen weeks.
You have a subject teacher for the IA, and for this task they are the right person: they know the criteria, they have moderated dozens, and they are free. Use them at the design stage rather than bringing a finished draft, because by then the marks are already allocated.
The hour that matters
Before you collect anything, write down what else could produce the result you expect.
If you cannot name one, you do not yet understand your own design — and that hour, spent now, is worth more than any amount of careful writing later.
Frequently asked questions
What makes a good IA research question?
It names the independent and dependent variables, specifies the population or material, and can be answered with the data you can actually collect in the time available. If your question does not contain the specific things being varied and measured, you have a topic rather than a question and the design marks have nowhere to attach.
Does my IA have to produce a positive result?
No. You are marked on design, analysis and evaluation, not on whether the hypothesis held. A well-designed study producing a null result and honestly interpreted scores better than a confounded study that happened to produce the expected answer.
How big does my sample need to be?
Big enough that the difference you are reporting is unlikely to be noise, which usually means more repeats than students plan for. If you cannot increase the sample, say so explicitly in the evaluation and quantify the uncertainty rather than hoping nobody notices.
Can I use data somebody else collected?
In several subjects yes, and it is often the stronger choice. Economics, geography, psychology and environmental systems all support secondary-data IAs using public datasets, which gives you a sample size no classroom study can match. Check your subject's requirements first.
What do students lose the most marks on?
The evaluation. Most students write it as apology — listing what went wrong and saying they would be more careful. The marks are for identifying which weaknesses actually affect the conclusion and how, which is analysis rather than contrition.
How is the IA different from the Extended Essay?
The IA is smaller, sits inside a course you are already taking, and is marked against that subject's own criteria by your teacher before moderation. The Extended Essay is longer, freestanding and externally assessed. The design discipline is the same; the scale is not.