| We present the analysis, evaluation and modification of current econometrics practice in estimation of Capital-Skill Complementarities (CSC) and its impact on derived technology parameter estimations. Capital Skill Complementarities is a key concept in modern labor economics and macroeconomics that is useful in explaining various economic dynamics observed over time. However, there are several issues in the econometrics practice that need attention. Specifically, we address three aspects of the econometric techniques used in CSC study. First, actual skilled labor is not observed therefore should be modeled as a latent variable; secondly, pooling many countries of very diverse backgrounds in estimation shadows the heterogeneity that should be corrected; thirdly, production factors are highly correlated, therefore directly using them in the econometric model will subject them to multicollinearity issues. In this study, we use Measurement Error Model to model the latent actual skilled labor. The unobserved skilled labor is modeled as a Berkson Model on a logit scale, which turns out to be a Random Coefficient Model. To address the heterogeneity among countries used in estimation, a conditional linear mixed model using a SVD-based conditioning matrix is studied. This technique is a generalization of current conditioning technique. At last, we explore the consequence of multicollinearity on the estimates of derived technology parameters and discuss several ways of dealing with multicollinearity in the literature. The study shows that for OECD countries, a threshold between Complete Primary and Some Secondary is appropriate, while for non-OECD countries, a more country specific threshold is desired. This result is consistent with and provides indirect support for the findings from current literature. Based on clustering the result of the relative positioning of each country and the underlying heterogeneous technology, we found significant explanation power in CSC and skilled labor wage premium. At the last part, we showed that handling multicollinearity appropriately is necessary to obtain sensible estimates of derived technology parameters. |
