Write your research question
An environmental issue, a dataset you have opened, and a rough sense of the system. If the data is not open in front of you, this sentence cannot be written yet, and step 1 explains why that is not a formality.
- One sentence, as a question or a comparison statement
- The indicator named, with its units
- The years and the places it covers
- No strategy, stakeholders or policy in it
Fifty words, and every one of them has to be checkable against a file.
Everything downstream inherits whatever you fix here, so write five versions and throw four away. The indicator and its units were somebody else's decisions: naming them is the difference between a question a reader can verify and one they take on trust.
Something almost nobody knows
The official teacher support material says so directly and gives its own worked example as a comparison statement. What it must do is make clear what the quantities are and how they relate, so that it guides an appropriate method. A full stop is allowed.
How has mean annual E. coli changed at Geneva's river monitoring stations between 1995 and 2024, and does it differ between urban and agricultural catchments?
A comparison of mean annual E. coli, in colony-forming units per millilitre, at Geneva's river monitoring stations between 1995 to 2000 and 2019 to 2024.
On this route I lean hard towards the statement. "How has X changed" quietly promises that you can say why, and a downloaded series cannot: you controlled nothing. A comparison statement claims exactly what you did, which is compare two sets of numbers you selected.
Everything below is how we suggest you actually do it.
Three ways a question fails before you collect anything
All three are free to fix while the sentence is still a sentence, and expensive once you have collected data against it. Read your question against each one.
Shade is cooler than sun. Water runs downhill. If everybody already knows the answer, there is nothing to find out and your background has nothing to establish.
Anything with an "and how does…" in the middle. One half will go unanswered, and an unanswered half is what the focus strand penalises.
An ambitious question with no method capable of answering it scores below a modest one done properly. Simpler and well executed beats complex and unfinished.
Two quick tests. Say your question to somebody outside the class: if they can tell you the answer, you have not got one yet. And can you see, today, what the last paragraph of your report says? If not, the question is too big.
Four parts, all present
Four things have to be in the sentence: what varies, what you measure, where, and across which years.
Name the indicator, not the concept. "Soil compaction" is a concept, and you never decided how to measure it: you chose a column somebody else had already defined, and the sentence should say which one. "Deforestation" becomes "annual net forest loss in hectares"; "water quality" becomes "mean annual E. coli in colony-forming units per millilitre".
Then the fourth part: the boundaries. Which years, and which places. They are not decoration, they are the inclusion rule you will have to justify in step 4, appearing early because it is a promise about what your answer covers.
One independent variable, and one dependent variable: several of either reads as an investigation that never decided what it was asking.
The thing you compare or track. Two periods, five catchments, thirty years.
The actual column, as its publisher names it, and what it is measured in.
The places the data covers, which may be narrower than the issue.
The years you will use, chosen for a reason you can state.
Keep the strategy out of it
Some advice recommends a two-part research question ending with something about stakeholder tensions. Do not follow it.
Your question is a promise about what your data will answer. A column of bacterial counts cannot answer a question about who paid for the sewers, so a two-part question guarantees that half of it goes unanswered.
Criterion B (Strategy) asks only that your issue has a clearly stated and credible connection to your question. The connection lives in the text, not in the sentence.
How has E. coli in Geneva's rivers changed since 1995, and how have tensions between the canton, the communes and French authorities shaped investment in the sewer network?
A comparison of mean annual E. coli, in colony-forming units per millilitre, at Geneva's river monitoring stations between 1995 to 2000 and 2019 to 2024.
Three drafts, and what each fixed
Not a tidy illustration. This is the real sequence the Geneva example went through, including one wrong turn that was only caught by opening the file.
No indicator, no units, no period, and a scale with no strategy attached to it. Every word of it is a category rather than a thing.
Operationalised, located, dated, and dead on arrival. The dataset has exactly one lake bathing station and it reads zero in 24 of its 28 years. There was nothing to analyse, and no amount of rewriting the sentence would have produced any. Only opening the file revealed it.
The indicator is named with its units. The comparison is stated rather than a causal claim. The two periods are the inclusion rule, declared in the question. And the sites moved from the lake to the rivers feeding it, which is where the variation was all along.
The lesson is the middle card. A question can be perfectly well formed, perfectly specific and perfectly impossible, and the only way to find out is to open the data before you commit to the sentence.
A hypothesis, and the temptation that comes with one
Optional and not assessed. If it clarifies the relationship you expect, include it; if it is decoration, leave it out.
What this route carries with it is a temptation. Your results already exist. You can plot the thing in ninety seconds, see which way it goes, and then write a hypothesis that predicts it. Nobody can tell from the finished report, and it is not a hypothesis: a prediction made after the fact predicts nothing and is worth no marks, because the thinking it was supposed to evidence never happened.
So if you want one, write it before you plot anything, and keep it even when the data disagrees. A refuted hypothesis is a perfectly good result and an interesting one to write about. A hypothesis that was always going to be right is a sentence you wasted.
This is the small version of the integrity question that runs through the whole route. Writing a hypothesis after peeking is harmless enough. Choosing which years or which places to keep after peeking is not, and step 4 deals with it properly.
Both have the same fix and it costs nothing: write the thing down first, in a file, with the date on it.
It can argue with a question you have already written: is the indicator named, are the units there, could the file you have actually answer it?
It cannot write it for you, and cannot tell you what your dataset contains. It will describe columns your file does not have, confidently. The sentence has to be checked against the file, not against a description of the file.
Ready for step 3?
Secondary data checklist0 of 18Read your sentence against every one of these, then read it against your open file. Fifty words can afford that many checks.
Your question is deliberately silent about people. Step 3 is where they arrive, and it is the same page on both routes: somebody has a plan for the place your data describes, and not everybody agrees with it.
The Geneva rivers investigation used on the secondary-data route of this guide is the author’s own analysis of a published cantonal dataset. Every figure quoted comes from the Canton of Geneva’s open data, downloaded on 10 September 2026; the duplication, the missing year and the varying campaign counts described in the guide are as found on that date and may since have been corrected. The selection rules, the analysis and the conclusions are the author’s and not the canton’s.