- Ensure you have a good mandate and a supportive institution to take on the question.
- Use the fewest assumptions possible to design your experiment. Assume that half or more of the existing literature (and tools) may not be reproducible or may have shortcomings.
- Design experiments that give the maximum informational value. Testing whether just one treatment cures a disease yields only one bit of information: yes or no. If you can test which of five treatments has efficacy, that is a much more informative experiment.
- Use orthogonal approaches to ask the same question. This avoids the artifacts and biases innate to each approach and allows for orthogonal validation.
Well-trained scientists and their passion are important rate-limiting assets of our society. Given how many important scientific questions there are to ask, reducing failure and ensuring each of you succeeds is important even beyond your personal career.
Share your own advice below!