This summer, 58 students in Research 101 handed in a full research proposal: a question, background, method, ethics, timeline, significance and limitations. Grades and written feedback went back this weekend, and each piece of feedback is a paragraph of specific comments, not a rubric score.
Reading them side by side, the gap between the strongest and the rest was almost never the topic, and it was rarely talent. Some of the most interesting ideas in the whole cohort sat inside proposals that were not finished yet. What separated the top of the pile was a short list of habits, and every one of them can be learned in an afternoon.
1. Ask a question you could actually answer
The most common problem was a topic dressed up as a question. How do cancer cells affect the body? is a textbook chapter. The answer is already known, and no single project could produce it. How can BMI be improved? is a redesign job, not something you can test.
The fix is to narrow until the question names what you would measure. One cancer, one organ, one mechanism. Or: does waist-to-height ratio predict recovery after hip surgery better than BMI does, in a public dataset you can name? That is a question with a population, a variable, and an answer you could turn out to be wrong about.
- Is it one sentence, ending in a question mark?
- Does it name who or what you are studying, and what you would measure?
- Could the answer honestly come back no?
- Could you get to an answer in months, not years?
2. Find the gap before you write the method
The best proposal we received found something very specific: a published study had collected both of the variables it cared about and never tested one against the other. That is a gap. More research is needed is not.
Gaps live at the end of papers. Authors list what they could not do in the limitations and future-work paragraphs, which makes those two sections the most useful thing to read when you are hunting for a question. Start with Google Scholar or PubMed, read five abstracts in your area, then read the last page of the two that interest you most.
A gap also answers the question every reader asks: what is new here? One well-built machine learning proposal lost marks for no error at all. It used a dataset that has been analysed hundreds of times, and never said what one more analysis would add. If your data is famous, your contribution has to be named.
3. Make the method answer the question you asked
In week four of Research 101 we describe a proposal as seven sections that have to agree with each other. The two most likely to disagree are the question and the method.
It happens quietly. A proposal asks about brain activity and measures a maths test score. Another asks about diabetes risk in adults and measures one blood-sugar spike in mice. Both methods measure something real. Neither measures the thing in the question.
The test is simple. Read your question, then read your method, and ask what number you will be holding at the end. If that number would not answer the question, one of the two has to change, and it is usually the question.
4. Specify the method until a stranger could run it
I will do a survey is not a method. Who takes it, how many of them, what are the questions, and what analysis turns the answers into a result? The methods we scored highest read like instructions. They fixed the order of conditions so it could not bias the result, named the most likely confound and said how they would rule it out, and had the actual consent forms and questions ready to use.
The same goes for every measure. One student wrote, almost in passing, that there is no good way to measure effort. That sentence was the whole project. Decide what effort means, hours logged or a checklist of study habits, and the question becomes answerable.
5. A literature review is a real method
Most high school students cannot run a lab experiment, and they do not need to. A careful review of published studies is legitimate research, and some of the best proposals this summer were reviews.
What made them good was that they were systematic. They named the databases they would search and the rules for which studies were in or out. They built a table recording the same things from every paper: who was studied, what was measured, what was found. And the best of them said how they would weigh different kinds of evidence against each other, like a controlled animal experiment against a correlational study in people.
6. Do not claim more than your evidence can hold
Ambition is good. Overclaiming is not. A computer model's score is not a measured result, and a correlation is not a cause. A reader trusts a proposal that says exactly what it can show and exactly what it cannot.
That makes the limitations section a strength, not a confession. The strongest ones were concrete: not the sample is small, but what a sample of that size can and cannot tell you. And if one number carries your whole argument, test what happens to the result when that number moves.
7. Finish every section, and make each one do its own job
The easiest marks lost were the ones for sections that were not really there. Headings followed by a dash. Template instructions still sitting in the text. An ethics section describing participants the project would never have. A significance section that was really a second list of limitations.
- Significance: who would act on your result, and what would they do differently?
- Timeline: phases with weeks attached, and a flag on the one that will run long. Data collection almost always does.
- Ethics: what actually happened or will happen, in your own words. Not what a template says should.
- Background: cite real studies by name. Two sentences and no sources is a placeholder.
If yours came back lower than you hoped
Read the feedback as a to-do list, not a verdict. Almost every comment we wrote names a fix you could make this week. The hard part of research is noticing a question worth asking, and plenty of you did that. Everything on this list is craft, and craft is the part that gets better with practice.
Then keep going. A finished proposal is the first page of a project, and the project is what you bring to The Symposium this winter.
Research 101 is free and online. Its next cohort starts November 9, and week four is where these seven habits become your own proposal, with written feedback from us at the end.