These conclusions about the interpretation of the coefficients on the observed
variables and the area dummy variables are reinforced when account is taken of
omitted variable bias. This can arise in the usual way if omitted variables are
correlated with included variables. It can also arise from endogeneity. We can test
and allow for endogenous included variables but if there are unobserved endogenous
variables they will in general be functions of the exogenous variables, both
observed and unobserved and will impart bias to the coefficients on the observed
variables.