“Whatever is worth doing at all is worth doing well.”
- Philip Stanhope, 4th Earl of Chesterfield

“Anything not worth doing is worth not doing well.”
- Robert Fulghum

Your time and energy are limited, use them wisely.

Gerd Altmann/Pixabay

• Why do I want to do this project?
• How will a certain project fit my overall (career/life) plans?
• Remember that one Yes means many No’s.

ijmaki/Pixabay

• Define tangible goals/outcomes/deliverables for your project.
  • Not that good: “My goal is to write a paper.”
  • Better: “My goal is to have a paper submitted within 12 months to a 1st tier journal.”

• Use e.g. the SMART approach (specific, measurable, achievable, relevant, time-based) or something similar.

Gerd Altmann/Pixabay

• Overconfidence with regard to time-line and impact are common (and a bit of it might be good).
• Estimate time/effort requirements based on prior performance.
• Get input from others.
• Find someone who can play devil’s advocate.

Gerd Altmann/Pixabay

• Explore rapidly what has already been done.
• Talk to others, read the literature to determine if it’s a good project.
• Do some rapid exploring/prototyping to see if things might work out the way you envision it.
• Assess feasibility given your time/resources.

The Goal is to come to a quick go/no-go decision.

Gerd Altmann/Pixabay

“Whatever is worth doing at all is worth doing well.”

• Once you start the project for real, be meticulous.
• Carefully think through everything.
• Make sure you understand every step/result, every table/figure.
• You should still have a time-line and at some point decide you are done.

Gerd Altmann/Pixabay

This fast/slow idea is similar to the explore/exploit approach.

• Is the project on track?
• Does the project still fit with my overall goals?
• Do I need to make adjustments?

Gerd Altmann/Pixabay

If you find yourself off course, don’t keep going and hoping that magic will happen and you’ll get a good final product after all.

Blog post discussing high failure rate of data science projects.

• Unless you produce tangible/measurable results, it doesn’t count.
• Some possible tangible results:
  • Published paper or final report.
  • Presentation to stakeholders.
  • Documented improvement in some process.
  • Blog posts on ‘what we learned’.
  • …

Even if your project doesn’t finish as planned (e.g. canceled, not further needed, clear it won’t work, …), do try to produce some tangible products that show what you did and learned.

• For academic work, think in terms of papers.
• A project needs to lead to at least one good paper.
• A project can consist of multiple papers.
• A paper should be substantial but focused.

OpenClipart-Vectors/Pixabay

Other outcomes exist (e.g. a grant proposal, a presentation, etc.), but the main academic product is the peer-reviewed paper.

• Option 1: Start with a question/idea, find suitable data.
• Option 2: Start with data, find a suitable question.
• Option 3: Iterate between question/idea & data.

Not all projects require data.

• You have some cool data that you are sure can lead to at least one paper.
• Start with some exploratory analysis.
• Based on your explorations come up with some interesting question(s).
• Check the literature and colleagues to make sure your question has not been answered yet and is interesting.

Do this fairly quickly.

• Collect it yourself.
• Get it from a collaborator.
• Get it from some (online) database/resource.

Pexels/Pixabay

Combining data sources often makes for very interesting projects.

• You have an idea (question or hypothesis) that you are sure will lead to a paper.
• Check the literature and colleagues to make sure your idea has not been covered and is interesting.
• Look around to see if you can get the necessary data.

Do this fairly quickly.

• Recent reviews of the topic by experts.
• Discussion with colleagues.
• Your previous work.
• Something just came to you in the shower.

Carlos Alvarenga/Pixabay

Keep an idea notebook/log, revisit occasionally.

• You might have an idea, then you look for data. You find some potentially suitable data, but realize that it does not allow you to address your original question. However you can address a slightly modified question.
• You might have data, and formulate a question that you could ask. Doing literature search, you notice that question has been answered before. But you realize you could ask a related but different question with your data that has not been answered.

Do this fairly quickly.

• To find questions (& data) that are both “big enough” (to be interesting) and “small enough” (to be doable).
• To get suitable data in a reasonable amount of time.
• To understand what makes a good question.
• Your project/paper needs to pass the So What? test.

• Once you decided you have an interesting and doable project, do it 😃.
• Visualize and draft the end product (i.e. paper) early.
• Keep good notes, document everything well.
• Be as reproducible as possible.
• Make sure you don’t cut corners.
• Set yourself a deadline (you won’t hit it, but have one anyway).
• Evaluate and reflect regularly, make changes as needed.

• Projects that produce useful outcomes no matter what happens are best.
  • If you have a hypothesis and the results are only interesting if it turns out to be right, it’s not that great.
  • If you can ask a question that will be of interest no matter the outcome, it’s better.

The Art of Data Science - a pay what you want/can ebook. Focuses on data science, but the first several chapters also apply more generally to projects.