This paper examines the robustness of determinants of IIT to changes i
n the set of included variables and to measurement errors. On applying
extreme bounds analysis (EBA) on Swedish IIT, it is found that most v
ariables are sensitive to changes in the selected control variables. W
hen examining the effects of errors of measurement and when only robus
t variables are included in the model, the estimated coefficients are
bounded although they are unbounded when more variables are included.
It is also found that the proxy-variables for product differentiation
need to be of much higher quality than actually seems to be the case,
if we are to ensure that the signs of the proxy-variables correspond t
o the signs of the 'true' variables.