I come up with three reasons:
1) It’s easy. So much of science is built on trust; generally, nobody comes into your lab and checks your notebook, equipment, computer code, or raw data. This is true of PIs as well – they trust that their grad students and postdocs are not faking their data.
2) There are (some short-term) incentives to cheat in science. In today’s hypercompetitive scientific community, there can be great pressure to cheat when you think your future in science is threatened. However, I think the long term incentives don’t favor cheating. Most serious cheaters seem to be caught quickly, the risks are huge, and the benefits of cheating scientists are more ephemeral than the benefits of many other types of fraud – scientists aren’t stashing laundered money away in offshore bank accounts.
3) When the data doesn’t go your way, it can be hard to accept that your idea is wrong. So much of science, especially experimental science, is a matter of judgment – what anomalous data is significant, and what data is simply a screw-up. Scientific publications by necessity are a selection of the work done by the authors, not a report of everything they tried. There are moments in every scientists career when some idea you knew just had to be true turns out to be wrong. Some cheaters are scientists who can’t deal with being wrong.