It seems the startup half of the survey was ran on Angel while the FAANG/MSFT was using a larger or more diverse data set from Blind. Unless I'm missing something this seems like a huge selection bias, since people actively focusing on startups (indicating this subgroup makes them happier) are more likely to be interested/happy with the work they find.

I had the impression that Microsoft treated its employees particularly well. Maybe things have changed since the 1990s.