776 Computational Considerations in Microeconometrics
15.4.2 Semiparametric estimation 793
15.4.2.1 Example: efficient estimation with heteroskedasticity of
unknown form 794
15.4.2.2 Example: partially linear model 796
15.4.2.3 Example: binary choice model 797
15.4.2.4 Further considerations 800
15.5 Modeling heterogeneity 801
15.5.1 Example: quantile regression 802
15.5.2 Example: finite mixture model 805
15.5.2.1 EM algorithm for model estimation 806
15.5.2.2 FM of count models 808
15.6 Simulation-based maximum likelihood 809
15.6.1 Model specification 810
15.6.2 Estimation algorithm 811
15.6.3 Example: MSL estimation 812
15.7 Concluding remarks 813
15.1 Introduction
Since the 1980s there has been a huge growth in the availability of both census
and survey data, in part due to the expansion of electronic recording and col-
lection of data. As data, computational power, and modeling opportunities have
grown, so have the variety and complexity of the modeling objectives of empir-
ical researchers. These developments have created modeling opportunities and
challenges that were largely absent when only aggregated market-level data were
available.
Explosive growth in the volume and types of data has also given rise to numerous
methodological issues. Certain features of micro-data are also ultimately respon-
sible for added computational complexity. Sample survey data, the raw material
of microeconometrics, are often subject to problems of sample selection, mea-
surement errors, incomplete and/or missing data, all of which generate additional
uncertainty about the econometric specifications used by empirical researchers and
impede the empirical generalizations from sample to population. One response to
these issues is that empirical researchers often explore several modeling strategies,
leading to additional computational complexity.
Model specification, estimation, testing, and then revision are essential com-
ponents of a modeling cycle. Whether econometric models are intended to be
exploratory and descriptive, or whether they aim to quantify structural relation-
ships, computation is central to every step in the modeling cycle. Important
features of modern applied econometrics include the following:
- Disaggregation. Microeconometrics is about regression-based modeling of eco-
nomic relationships using data at the levels of individuals, households, and
firms. The low level of aggregation in the data has immediate implications for
the functional forms used to model the relationships of interest. Disaggregation