Ebook Human Capital In Growth Regressions: How Much Difference Does Data Quality Make? An Update And Further Results
Recent empirical investigations of the contribution of human capital accumulation to economic growth have often produced discouraging results. Educational variables frequently turn out to be insignificant or to have the "wrong" sign in growth regressions, particularly when these are estimated using first-differenced or panel specifications.
The accumulation of such negative results in the recent literature has fueled a growing skepticism about the role of schooling in the growth process, and has even led some researchers (notably Pritchett, 1999) to seriously consider possible reasons why educational investment may fail to contribute to productivity growth.
In this paper we provide evidence that these counterintuitive results on human capital and growth can be attributed to deficiencies in the data, and show that improvements in data quality lead to larger and more precise estimates of schooling coefficients in growth regressions. In the first part of the paper, we review the main schooling data sets that have been used in the empirical growth literature and document a number of suspicious features and inconsistencies that suggest that these data contain substantial measurement error.
Next, we attempt to increase the signal-to-noise ratio in the data by constructing new schooling series for a sample of 21 OECD countries. These series make use of previously unexploited sources, including unpublished data supplied by the OECD and by a number of national statistical agencies, and rely on a heuristic procedure to obtain plausible time profiles for attainment levels by removing sharp breaks in the data that can only be due to changes in classification criteria. Following Krueger and Lindhal (2001), we use estimates of reliability ratios to measure the amount of measurement error in different cross-country data sets and find that, for the case of the OECD sample we work with, our series have the highest information content of all those available, followed closely by the attainment estimates constructed by Cohen and Soto (2001).
In the final part of the paper, we systematically compare the performance of our series and several other attainment data sets in a number of standard growth specifications and find a clear positive correlation between estimated schooling coefficients and reliability ratios. Finally, we use an extension of the classical errors-in-variables model to construct estimates of reliability ratios that allow for correlated measurement errors and to obtain a set of meta-estimates of the coefficient of schooling that correct for attenuation bias. The exercise suggests that the coefficient of years of schooling in an aggregate Cobb-Douglas production function is well above 0.50.
Posted in :
