Pass fixed regressors, go straight to sampling

Over the past 20 years, the combination of growing computer power, widening availability of large cross-sectional data sets and the lengthening of the time-series dimension to microeconomic panels has encouraged development of increasingly sophisticated microeconometric techniques and has enabled their application.

Jeffrey Wooldridge's excellent textbook well meets the need for an up-to-date text explaining state-of-the-art methods to potential practitioners among graduate students. The book covers familiar and less familiar ground with clarity and with attention to the issues of greatest relevance to practical application.

It assumes acquaintance with basic econometric material - stochastic explanatory variables are assumed throughout (wasting no time on the practical fiction of fixed regressors), and it concentrates on large sample properties of estimators. From the beginning, the book draws attention to the structural model of interest, typically parametrically specified, distinguishing this clearly from other conditional expectations that may be recovered from data by particular techniques. It draws a clear distinction between the population model and the sampling assumptions, recognising issues raised by non-standard sampling schemes.

With these concepts in place, issues of identification can be given due prominence. The treatment of linear models, for example, begins with a discussion of ordinary least squares, focusing on the central concerns evident in empirical practice - omitted variables, simultaneity and measurement error. This is discussed in an accessible way that reflects the reality of empirical practice and that points to the economic content of issues encountered in applications. Instrumental variables estimation is treated next, and it figures subsequently with appropriate prominence for such a commonly applied technique.

The book also discusses pitfalls. It includes a measured examination of recent work on problems raised by weak instruments. Subsequent chapters give special attention to the panel data settings. Topics hitherto largely absent from textbooks include panel data methods for non-linear models and recent work on evaluation methods for heterogenous treatment effects.

This is a book that I expect to find extremely valuable for reference in research contexts, and it should also be a first-rate basis for teaching. Any graduate student with an interest in applied microeconometric work would benefit from access to it.

Ian Preston is reader in economics, University College London.