Research under this theme examines the nature of the restrictions that endow an econometric model with the power to identify particular structural features. For any particular feature there is special interest in the content of sets of restrictions which just identify the feature and in determining sets of non-falsifiable identifying restrictions. This interest leads to the study of non-parametric identification, that is the study of the identifying power of models which do incorporate parametric restrictions. There is research under three main themes.

(i) Continuous variation. Models which permit non-additivity of latent variates are of interest because they permit stochastic across-individual variation in individuals responses to policy interventions. Results on identification in such models is reported in cemmap working papers: CWP 01/01, 08/01 and 05/02, the last of these published in revised form in Econometrica in September 2003. These papers propose models embodying weak local restrictions which identify the sensitivity of smooth structural functions to ceteris paribus continuous variation in their arguments.

(ii) Discrete variation. Discrete variation in covariates limits the identifying power of a model and discrete variation in outcomes gives rise to particular problems. These issues are studied in cemmap working papers CWP 17/02, 06/03, 19/03 and 11/04.

(iii) Excess heterogeneity. The models considered in (i) and (ii) permit no more unobservable variables than outcomes. This restriction is relaxed in cemmap working paper CWP 20/02. Further work on this topic is being undertaken jointly with Whitney Newey (cemmap Fellow and MIT) and Frank Vella (EUI).