Partner Selection into Policy Relevant Field Experiments

This study investigates the issue of self-selection of stakeholders into participation and collaboration in policy-relevant experiments. We document and test the implications of self-selection in the context of randomised policy experiment we conducted in primary schools in the UK. The main questions we ask are (1) is there evidence of selection on key observable characteristics likely to matter for the outcome of interest and (2) does selection matter for the estimates of treatment eff ects. The experimental work consists in testing the e ffects of an intervention aimed at encouraging children to make more healthy choices at lunch. We recruited schools through local authorities and randomised schools across two incentive treatments and a control group. We document the selection taking place both at the level of local authorities and at the school level. Overall we nd mild evidence of selection on key observables such as obesity levels and socio-economic characteristics. We find evidence of selection along indicators of involvement in healthy lifestyle programmes at the school level, but the magnitude is small. Moreover, We do not find signifi cant di erences in the treatment e ffects of the experiment between variables which, albeit to a mild degree, are correlated with selection into the experiment. To our knowledge, this is the rst study providing direct evidence on the magnitude of self-selection in fi eld experiments.

Sharp Bounds on Heterogeneous Individual Treatment Responses

This paper discusses how to identify individual-specific causal effects of an ordered discrete endogenous variable. The counterfactual heterogeneous causal information is recovered by identifying the partial differences of a structural relation. The proposed refutable nonparametric local restrictions exploit the fact that the pattern of endogeneity may vary across the level of the unobserved variable. The restrictions adopted in this paper impose a sense of order to an unordered binary endogeneous variable. This allows for a uni.ed structural approach to studying various treatment effects when self-selection on unobservables is present. The usefulness of the identi.cation results is illustrated using the data on the Vietnam-era veterans. The empirical findings reveal that when other observable characteristics are identical, military service had positive impacts for individuals with low (unobservable) earnings potential, while it had negative impacts for those with high earnings potential. This heterogeneity would not be detected by average effects which would underestimate the actual effects because different signs would be cancelled out. This partial identification result can be used to test homogeneity in response. When homogeneity is rejected, many parameters based on averages may deliver misleading information.

Baysian Model Averaging, Learning and Model Selection

Agents have two forecasting models, one consistent with the unique rational expectations equilibrium, another that assumes a time-varying parameter structure. When agents use Bayesian updating to choose between models in a self-referential system, we find that learning dynamics lead to selection of one of the two models. However, there are parameter regions for which the non-rational forecasting model is selected in the long-run. A key structural parameter governing outcomes measures the degree of expectations feedback in Muth's model of price determination.