social interaction: finding instruments for social interactions

William A. Brock and Steven N. Durlauf. 2001:166-167. Handbook of Econometrics. vol 5.
Interactions-based models.
At the same time, we would argue that the issue of omitted variables is far from insuperable. Both the social psychology and sociology literatures have focused a great deal of attention as to which types of individual and group control variables are most appropriate for inclusion in individual level regressions through the determination of which variables seem to be proximate versus ultimate causes of individual behavior; indeed it is this distiction which is the basis of path analysis; see Sampson and Laub (1995) for what we consider a persuasive example of such a study. In general, we find it likely that these literatures will be able to identify examples of individual variables whose group average analogs are not proximate causes of behavior, and hence are available as instruments. While these literatures are often not driven by formal statistical modelling and further subjected to Sims-Freedman type critiques (Freedman 1991) when formal techniques are employed, this hardly means that these literatures are incapable of providing useful insights [they are hardly devoid of insights]. In this respect, we find arguments to the effect that because an empirical relationship has been established without justification for auxilary assumptions such as linearity, exogeneity of certain variables, etc., one can ignore it, to be far overstated. In our view, empirical work establishes greater or lesser degrees of plausibility for different claims about the world and therefore the value of any study should not be reduced to a dichotomy between full acceptance or total rejection of its conclusions. Hence the determination of the plausibility of any exclusion restriction is a matter of degree and dependent on its specific context, including the extent

Publish Post

to which it has been studied.