control function

The ordinary least squares (OLS) estimators will be biased If there is a relationship between observable input levels and unobservable productivity shocks in the production function. Along instrumental variables (IV) and generalized moments method (GMM) approaches, the control function approach is the most widely used in production function estimation to solve this problem. In this article, the production function was estimated by the conventional panel data method and the control function approach and compared with using Iran’ provinces manufacturing industries data of in the period of 2011-2018. The findings showed that the OLS estimator in the panel data methods overestimates the free variables coefficients (labor force), which this is consistent with the argument of Levinson and Petrin (LP) (2003). In the control function approach methods, Wooldridge (WR) method estimators are more efficient than LP method. Among the optimization algorithms, the Nelder and Mead (NM) algorithm results are more efficient than others. By assuming that the capital stock is endogenous, the results indicated that the ACF correction method (by controlling the labor force functional dependence) is more efficient than LP method. Finally, estimators’ efficiency was increased by using dynamic panel instruments in the MR approach. The Article findings point out that the labor force functional dependence and the endogeneity of the capital stock can be important issues in the production function estimation that should not be ignored. Thus, it seems that GMM nonlinear models in general, and control function models, namely OP, LP with ACF and MR correction in particular, should be used more in the production function estimation for controlling unobservable productivity shocks.