Abstract
This paper proposes a new methodology for the identification of the effect of time-invariant variables in a linear panel data framework with fixed effects (FE). We show that identification of time-invariant variables can be achieved under the assumptions that (i) the correlation between the individual fixed effect and the regressors is constant over time, and (ii) the time-invariant variables are correlated with the within variability of the time-variant variables (e.g., their growth rate). Monte Carlo experiments show that the proposed approach allows to estimate the coefficients of time-invariant variables, with marginal decrease in the precision of the estimates of the coefficients of the time-variant variables as compared to FE. The proposed methodology is applied to the identification of the role of « Science » at the firm level. We gauge whether and how much firm-level publications’ stock (proxy of « Science ») might affect labour productivity. The results show that inputs from Science can be beneficial to those firms with higher level of R&D investments already in place, further supporting the dual role (direct and indirect) of R&D.