Yes, although there are several factors that can influence an outcome, if the researcher carries out an extensive search for the review of existing studies and if they opt for the most appropriate theory for their study, then they are likely to end up opting for the most relevant variables for their regression analysis as well. Furthermore, if the sampling and data collection processes are carried out scientifically, the results are also likely to be accurate.

No. They generally cannot deal with endogeneity

Interesting last comment. If you follow all the required steps for regression modelling it is a very robust and reliable approach. It is only a statistical method though, and is only ever as good as the investigation to which is applied. I was surprised by the previous answer, so left field and so packed full of assumptions. Perhaps a misleading comment in this context. A statistical method should never be judged except in the context of the situation in which it is used. Regression modelling (read this as meaning all types of multivariate statistics using similar linear relationship assumptions) has been used with great benefits for nearly 100 years in the social sciences. I would call it a robust approach whether all the variance is captured or not. But like most tools it is only as good as the scientist using it and their knowledge of how to optimise it for the intended purpose. Endogeneity was rarely ever mentioned in my experience (The previous author may care to explain what they think this means, and provide some mathematical evidence for their claim). Typically in linear regression modelling, whatever the hypothesised source of variance (including endogeneity) might be, one can always create a dummy variable to represent it, and specify some parameters for it, and include it in the model. Clever statisticians do that all the time. More reading on the history and development of linear statistical models might help as well.