How to Choose a Research Question for a High-School Project

Choosing a research question is often harder than choosing a subject. You might enjoy environmental science or computer science but still think, “What could I actually study?” Try to name something specific you'd like to understand, then consider whether you have the time, skills, and information to investigate it. For a project using data or computation, you'll need to decide what to measure, what to compare, and where to find the data.

You don't need to settle every detail yet. Read a little and look at the available data before committing to an idea. We'll use a bike-sharing example to work through those early decisions.

Choose something specific you want to understand

Start with something you've noticed or wondered about. If you're interested in transportation, you might wonder why bike-share rentals vary so much from day to day. Why are there more rentals on some days? Does the weather help explain the difference? You could investigate whether weather information improves estimates of how many bikes people rent.

Read a few relevant papers before committing to the question. Look for what researchers measured, what they found, and what they couldn't explain. You may discover that your question has already been studied or needs to be more specific. The George Mason University Writing Center recommends this kind of preliminary reading when developing a focused question. Keep notes on what you'd need to learn or check before attempting the project.

Check whether the data can answer your question

Find a possible dataset and read its documentation. Check what each row represents, which measurements are included, and whether you're allowed to use it. Then open a few rows yourself. You might find a measurement you don't understand or discover that information you need is missing. It's better to find that out before planning the experiment.

For the bike-sharing example, the UCI Bike Sharing dataset includes rental counts from Capital Bikeshare in 2011–2012, along with weather and seasonal information. You could use it to ask: Does adding weather information improve estimates of daily bike rentals compared with using calendar information alone?

This would be a practice investigation; you'd need to review earlier studies before claiming an original finding. Be careful about what the data can tell you, too. Records from 2011–2012 describe that period, and recorded weather tells you what happened on a given day. If you wanted to forecast rentals in advance, you'd need to use weather information that was available when the forecast would have been made.

Plan a comparison you can explain

For this example, you could compare two versions of the same model: one using calendar information, such as the month and day of the week, and another that also uses weather information. You'd then compare each model's estimates with the actual rental counts. One simple measure is the average size of the errors. If an estimate is 20 rentals too high or 20 too low, both count as an error of 20.

Set aside some data for the final evaluation before you develop the models. Use the remaining data to build and adjust them. If you've already used the final evaluation data to make those adjustments, the results can look better than they really are. The scikit-learn documentation explains this problem, known as data leakage, in more detail.

Decide how you'll evaluate the comparison before looking at the final results. The weather information might help, make little difference, or make the estimates worse. You won't know until you test it. Whatever the result, check whether the method and data support your explanation.

Keep the first project manageable

For a first attempt, one dataset and one comparison are enough. Estimate how much time you'll need to understand the data, learn the method, and run the experiment. If too much is unfamiliar, simplify the question or try a smaller practice exercise first. Leave time to check your work and explain its limitations; running the experiment won't be the last thing you need to do.

Research for college applications

If college applications are part of your motivation, choose a subject you'd still want to understand even if the result is modest. MIT's admissions guidance encourages students to choose activities for their interest and challenge. When describing your project later, be specific about the decisions you made, the work you completed, and what you learned.

Write a short plan before you begin

Once you've read a little and inspected the data, answer these five questions in a few sentences:

1. What do I want to find out, and why does it interest me?
2. What data will I use?
3. What will I compare, and how will I measure the difference?
4. What do I need to learn before I can do this?
5. What will this project leave unanswered?

If an answer is unclear, spend more time on that part before committing to the project. Keep your first plan, even as you revise it. Later, you'll be able to see how your thinking changed and explain why you chose a different question or method.

Question23's free Topic Exploration can help you consider possible research questions and what each would involve. It provides AI guidance, with an initial focus on computational and machine-learning research. You make the research decisions and write your own work. Try free Topic Exploration.