Comparison of cobweb and rational expectations models of PhD student enrollments.

***pStandard errors robust to arbitrary heteroskedasticity are presented in parentheses below each coefficient estimate.All results presented are from instrumental variables estimation, with outcome variable the log of the ratio of first-time, full-time enrollment in U.S. biomedical sciences graduate programs each year to bachelor's degrees in biological sciences and chemistry-related fields from U.S. institutions in the previous academic year. Models (1), (2), and (4) are dynamic first-order autoregressive (AR(1)) models, whereas models (3) and (5) are instead first-differenced to remove autocorrelation. Models (1), (2), and (3) correspond to cobweb expectations, with all explanatory variables measured at time of students' entry into graduate programs. Models (4) and (5) assess rational or forward-looking expectations, with the relative wage and employment variables measured six years after entry into the graduate program. We instrument for wages and employment with third and fourth lags of each of our two measures of pharmaceutical industry R&D expenditures, divided by the U.S. Gross Domestic Product.