Microdata
Selecting a Model

The criteria used to select models estimated using microeconomic data are likely to differ from those applied in the specification of macroeconomic models. In Discussion Paper No. 57, Costas Meghir and Associate Programme Director Richard Blundell discuss the relative importance which should be given in selecting microeconomic models to the flexibility of the model, to its consistency with economic theory, and to its random stochastic properties.

Microeconomic modelling is attractive because of the important relationship between the economic theory, its stochastic specification of the model and the appropriate estimation procedure. But this close relationship also makes the properties of the model estimates quite sensitive to stochastic specification. This is particularly so in microeconometric models where individuals are assumed to choose between different observed 'regimes' of behaviour, such as participation or non- participation in the labour force. In such situations the pattern in which individuals are distributed across 'regimes' enters the analysis and the method of estimation. It therefore becomes important to verify that the assumed stochastic specification is appropriate, and here the authors draw on the new and expanding literature in this area.

It is also necessary to assess the flexibility of the estimated model and its consistency with economic theory. The condition of flexibility is of particular importance for microdata sets, where there is a wide variety of behaviour across individuals with quite different economic and demographic characteristics which must be accommodated within the model. Blundell and Meghir note that a model's consistency with economic theory cannot be assessed unless its stochastic specification has been chosen appropriately in the first instance. Although it is convenient for many purposes to have a model that is globally consistent with economic theory, the authors argue that this should not be bought at the cost of imposing overly strong restrictions on the model, which may prevent the model from properly reflecting the behaviour observed in the data.

Blundell and Meghir illustrate these factors by examining the specification of a labour supply model for married women based on data from the UK Family Expenditure Survey.


Selection Criteria for a Microeconometric Model of Labour Supply
Richard Blundell and Costas Meghir

Discussion Paper No. 57, March 1985 (ATE)