Research Foundations › Module 2: Asking a good question
The FINER test
Lesson 2 of 3 · 2 min
Once you have a question, test it before you spend months on it. A popular checklist comes from medical research, from the textbook "Designing Clinical Research" by Hulley and colleagues. It is called FINER.
| Letter | Means | Ask yourself |
|---|---|---|
| F | Feasible | Do I have the data, time, computing power and skills? Can it be done in my time frame? |
| I | Interesting | Does it interest me, and would it interest others working on this topic? |
| N | Novel | Does it add something: a new result, or a check of an old one? |
| E | Ethical | Can it be done without harming anyone or misusing anyone's data? |
| R | Relevant | Would the answer matter to science, to practice, or to people? |
- F
- Means: Feasible
- Ask yourself: Do I have the data, time, computing power and skills? Can it be done in my time frame?
- I
- Means: Interesting
- Ask yourself: Does it interest me, and would it interest others working on this topic?
- N
- Means: Novel
- Ask yourself: Does it add something: a new result, or a check of an old one?
- E
- Means: Ethical
- Ask yourself: Can it be done without harming anyone or misusing anyone's data?
- R
- Means: Relevant
- Ask yourself: Would the answer matter to science, to practice, or to people?
Feasible is where most student projects fail
Ambition is good, but a project that cannot be finished teaches less than a small one that can. Check:
- Data. Does the dataset exist, and are you allowed to use it? Read its licence.
- Compute. Can you train the models you need on what you have? Free notebook GPUs have strict time limits.
- Time. List the steps and guess how long each will take. Then double it.
- Skills. What do you need to learn first? Is there someone who can help?
Novel does not mean "never done"
Repeating a published study on new data, or with a fair comparison the original missed, is a real contribution. What is not novel is redoing a known exercise and presenting it as new.
RememberIf your question fails one letter of FINER, change the question, not the facts.
Example"Can I build a better model than GPT-4?" fails F for almost any student. "Does a small, fine-tuned model match a large general model on classifying Hindi-English mixed tweets, at a fraction of the cost?" passes all five.