The paper examines the determinants of scientific productivity (number of articles and journals' impact factor) for a panel of about 3600 French and Italian academic physicists active in 2004-05. Endogeneity problems concerning promotion and productivity are addressed by specifying a generalized Tobit model, in which a selection probit equation accounts for the individual scientist's probability of promotion to her present rank, and a productivity regression estimates the effects of age, gender, cohort of entry, and collaboration characteristics, conditional on the scientist's rank. We find that the size and international nature of collaborative projects and co-authors' past productivity have very significant impacts on current productivity, while age and gender, and past productivity are also influential determinants of both productivity and probability of promotion. Furthermore we show that the stop-and-go policies of recruitment and promotion, typical of the Italian and French centralized academic systems of governance, can leave significant long-lasting cohort effects on research productivity.