economy. Our results seem to indicate that both workers’ and managerial experience matter for
productivity growth.
As to managerial age, definite patterns of correlation are present once the whole sample is split into
innovative and non-innovative clusters. Age, in particular - a measure of overall experience - in the
labor market appears to be (positively correlated or) uncorrelated with productivity growth in non-
innovative firms, while it is robustly negatively correlated with productivity growth in the sample of
innovative firms. Results are also strongly statistically significant for our other variable of interest:
the share of temporary workers is in most cases negatively correlated with productivity growth.
This result seems to differ across groups in absolute value, being more important for non innovative
firms.
The cross-sectional statistical analysis of long-differences based on firm-averaged data is not
problem-free. A big issue is potential reverse causation. The statistical relations we intend to
analyze posit that (say) age is the independent variable and productivity the dependent variable. But
cross-section data as such (be they observed at a given point in time or averaged over time) may
only indicate correlation, not causation. Therefore, if the estimated coefficient linking age and
productivity is negative, this may not indicate that the firms where aged managers are employed are
less productive. Rather, the negative correlation may simply signal that older managers tend to stay
longer in less productive and older firms, featuring outdated machines and methods of production,
probably because they managed to put in place successful “relations”, while new, innovative and
high-productivity plants may be more often matched to young and brilliant managers. If this is the
case, we would be wrongly interpreting what causes what, attributing to age a causal influence on
plant productivity, which may go the other way around. This is why we implement our 2SLS
specification. Our expectation is that by choosing predetermined instruments, which also include
age of the firm, we are lessening the simultaneity problems.
Surely, a lot of unobserved heterogeneity in plant productivity is still there in the data even once we
have augmented the list of productivity determinants with dummies and other control variables. Yet
25