As I do not have any econometric background I find it difficult to understand which one to use when outcomes differ a lot. My dependent (female labor-force participation status), independent (has more than three children) and instrumental (combination of gender of first two children) variables are all binary. OLS, logit, and probit models all have shown similar results (as mentioned in "Mostly harmless econometric" when it comes to average marginal effect there is no difference between these models). But when I run ivreg (2SLS), ivprobit, and biprobit results are totally different. Previous literature just simply use OLS and 2SLS, but I can not find convincing solid reason behind it. (I checked "Mostly harmless econometrics" by Angrist and "Econometric analysis of cross section and panel data" 2nd edition by Wooldridge where explanations were quite vague) What are the trade-off between these models, and how to interpret difference in estimations and how to choose right one?