Data and code download link (Just 91KB): https://drive.google.com/drive/folde...2M?usp=sharing

In this data example I came up with, I get following results.
Command I run Result
1 reg positive & significant
2 ivreg positive & significant
3 probit coefficient positive & significant
4 probit marginal effect positive & significant
5 ivprobit coefficient positive & significant
6 ivprobit marginal effect negative&significant

This is strange in so many levels. #5 and #6 having different sign is strange. #2 and #6 having different sign is strange.

And if this is possible, in what situations would it occur?