As an example, imagine you're running the faculty of computer science and decide which books to use for the next semester.
- You pick book A for all students -> no IRB review needed
- You pick book B for all students -> no IRB needed
- You pick book A for half the classes and book B for the other half (due to professor preference, for example) -> no IRB needed
- You decide to run an experiment where you want to see whether book A works better than book B -> IRB needed
- You retroactively decide to analyze whether book A was better than book B -> simplified version of IRB is needed for retrospective data analysis
What is so special about something being an experiment?