We address this issue by examining the impact of unobserved
attributes on the probability that a firm is in a treatment group,
IT outsourcing in our context (Rosenbaum 1999). To this
end, we construct the odds ratio
(Γ) which measures the
extent to which firms with the same observed attributes differ
in their sourcing mode. If two firms have the same attributes
and
Γ equals 1.5, they differ in their propensity to outsource
by a factor of 1.5, or 50 percent. If there are no unobserved
variables that influence a firm’s sourcing mode, the ratio
equals 1. In practice, however, we do not know the true value
of
Γ. Therefore, we conduct a sensitivity analysis by
changing the values of
Γ and examine how our inference
about the outsourcing effect is altered. Following Rosenbaum
(1999), we conduct Wilcoxon sign–rank tests and our test
statistics suggest that our estimates become sensitive due to
unobserved variables when
Γ = 2 (i.e., firms with the same
observed attributes differ in the propensity to outsource by
100 percent due to unobserved attributes). Given that we
capture a large portion of variations in the sourcing decision
with several key variables, we believe that this is a large
difference. In sum, we can conclude that our results are not
driven by unobserved variables.