P-hacking can be done by analyzing the data in different ways until a significant effect is found, and then failing to report all of the analyses done. There is no universal way to analyze data (especially what collected data to include); there are choices that are made and this leads to a “garden of forking paths” where if you look around the garden enough noise alone will give you ‘signicance’. Preregistration makes this impossible by declaring your analysis before collecting data.
