OLS and TSLS

[Zainab Ali]

I have been working on OLS and TSLS. I created an index which will be my dependent variable. Earlier I created an index using pca and then I came to know that all my variable of index are binary so I shifted to mca method (indicator). Now the issue is, I got my TSLS results as per my requirement (significant and positive) but my OLS results are negatively significant. I also tested for multicollinearity and there isn't any. I added controls and did robustness as well. What can be the issue or are the results accurate?

[Andreas Backhaus]

There are various assumptions required for either OLS or TSLS to be "accurate"; these assumptions differ by method. As OLS and TSLS (which I assume you're using as part of an instrumental variable strategy) are estimating different effects, different signs of the coefficients are not necessarily surprising or indicative of anything. Maybe checking out the Angrist/Pischke books and lecture slides would be helpful.