nominated but the duration of the service is practically unknown, and her/his function is scarcely
available. Thus we take the board composition as if it was present in 2001-2003 as well. For each
member with available data, we calculate her/his age at the time of the nomination within the board,
as well as the age in 2001, 2002 and 2003. For each firm we have calculated an average age of the
board. We excluded from the dataset those firms whose board name appeared to be another
company, not a physical person. We also excluded those firms whose board members’ appointment
appears to have occurred before their birth dates.
Firms in the Capitalia-Unicredit dataset (with information relative to the 2001-2003 period) also
present in 2007 to match AIDA information are 3562 (that is 85.3% of the sample). Firm-individual
observations are 21081. We first test for potential sample selection of these firms, in terms of age,
size, location and sector of production (younger, bigger or particular sectors could have a higher
survival rate, higher productivity or innovation capacity). The discussion of the potential selection
bias is placed in the Appendix section of this paper. The data then need a cleaning procedure
because of inconsistencies between birth dates and appointment dates of the individual board
members, implausible firm age, non-individual board members, missing values. Only 7977 (about
40% of total observations) distributed in 1042 firms contain sensible information on birth and
service dates and other variables, which finally becomes our longitudinal or “quasi”-panel dataset
with firms as units and board members as the longitudinal dimension, in the years 2001-2003.
15