This was a microeconometrics course covering topics in panel data econometrics and treatment effect estimation. The first three days will focus on modern panel data analysis, beginning with the linear model in situations where we have a large cross section and relatively few time periods. In addition to reviewing the standard estimation procedures we will discuss correlated random effects approaches and how they can be applied to obtain simple specification tests. We also cover models with random slopes, instrumental variables estimation, and dynamic models. An introduction to estimating and interpreting nonlinear models with unobserved heterogeneity – including dynamic models – rounds out the panel data part of the course.

The counterfactual approach to treatment effect estimation is the topic of the last two days. Regression adjustment, propensity score weighting, and matching methods are all covered under the assumption of unconfounded treatment assignment. Difference-in-differences estimation is covered in the counterfactual setting. When the unconfoundedness assumption fails we need instrumental variables. We will study both IV and control function approaches to correcting for selection with both discrete and continuous “treatments.”

Text/Lecture Notes

J.M. Wooldridge, Econometric Analysis of Cross Section and Panel Data, second edition. MIT Press, 2010.

Imbens, G.W. and J.M. Wooldridge (2009), “New Developments in Econometrics.” http://www.cemmap.ac.uk/resources/resources25.php

Day 1: We will begin with the basic linear model with additive heterogeneity and study pooled OLS, random effects, fixed effects, first differencing, and correlated random effects estimation approaches. We will discuss how to test the key assumptions for the RE and FE estimators, including strict exogeneity conditional on the latent effect. We will begin studying models with endogenous explanatory variables that require instrumental variables estimation.

Day 2: On the second day we will consider estimation of dynamic linear models. In addition, we will discuss some recent advances on models with models with random slopes. We will cover material on the consequences of unbalanced panels, sample selection, and attrition, and discuss testing and some remedies.

Day 3: We discuss the nature of nonlinear panel data models with unobserved heterogeneity and ask: What kinds of economic quantities can we estimate? After presenting general settings for models with strictly exogenous regressors we will study binary response, Tobit, and count data examples. We will also study dynamic models with unobserved heterogeneity.

Day 4: The fourth day begins the material on treatment effect estimation. We introduce the counterfactual framework and discuss identification and estimation when treatment assignment is unconfounded. Regression adjustment, propensity score weighting, matching, and combined methods will be discussed. Time permitting we will discuss difference-in-differences estimation in a counterfactual setting.

Day 5: Day five will cover instrumental variables estimation of various treatment effects when selection on observables does not hold. Both traditional control function methods and some extensions will be covered, for linear and nonlinear models. As time permits we will cover treatment effect estimation with general panel data structures.

Lab Work: In the computer lab in the afternoon sessions we will implement most of the methods discussed in lecture, using data sets covering a variety of areas in economics. We will use Stata as the statistical software package.

Reading Assignments

The unlabelled readings are from my text listed above.

1. OLS, Random Effects, Fixed Effects, and First Differencing

Basic Issues: 7.1 7.2, 10.1, 10.2
Estimation by Pooled OLS: 7.8, 10.3
Random Effects Estimation: 10.4
Fixed Effects Estimation: 10.5

2. Violations and Testing of Assumptions

RE versus FE and Correlated Random Effects: 10.7.3
Testing the Strict Exogeneity Assumption: 10.7.2
Imbens and Wooldridge (2009), “Linear Panel Data Models I,” cemmap/UCL Lecture Notes.Serial Correlation: 10.5.4, 10.6.3

G. Chamberlain, “Panel Data,” Handook of Econometrics Volume 2, Chapter 22.

3. Instrumental Variables Estimation

RE and FE Instrumental Variables Estimation: 11.2
FD-IV Estimation: 11.4
Dynamic Models: 11.6
Imbens and Wooldridge (2009), “Linear Panel Data Models I,” cemmap/UCL Lecture Notes.

Arellano, M. and B. Honoré (2001), “Panel Data Models: Some Recent Developments,” in Handbook of Econometrics, Volume 5, ed. J.J. Heckman and E. Leamer. Amsterdam: North Holland, 3229-3296.

4. Random Trends and Slopes

Models with Individual-Specific Slopes: 11.7

Wooldridge, J. (2005), “Fixed Effects and Related Estimators for Correlated Random-Coefficient and Treatment Effect Panel Data Models,” Review of Economics and Statistics 87, 385-390.

Mutazashvili, I. and J. Wooldridge (2008), “Fixed Effects Instrumental Variables Estimation in Correlated Random Coefficient Panel Data Models,” Journal of Econometrics 142, 539-552.

5. Sample Selection and Attrition in Linear Panel Data Models

Fixed Effects on Unbalanced Panels: 19.9.1
Testing and Correcting for Sample Selection Bias: 19.9.2
Attrition: 19.3.3

Imbens and Wooldridge (2009), “Missing Data,” cemmap/UCL Lecture Notes.

Wooldridge, J. (1995), “Selection Corrections for Panel Data Models Under Conditional Mean Independence Assumptions,” Journal of Econometrics 68, 115-132.

Semykina, A. and Wooldridge, J. (2010), “Estimating Panel Data Models in the Presence of Endogeneity and Selection,” Journal of Econometrics 157, 375-380.

Wooldridge, J. (2009), “Correlated Random Effects Models with Unbalanced Panels,” manuscript, Michigan State University

6. Overview of Nonlinear Models and Quantities of Interest

General Nonlinear Models with Unobserved Effects: 13.9

Imbens and Wooldridge (2009), “Nonlinear Panel Data Models,” cemmap/UCL Lecture Notes.

7. Examples of Nonlinear Models Response Models for Panel Data

Pooled Probit and Logit: 15.8.1
Unobserved Effects Probit Models: 15.8.2
Unobserved Effects Logit Models: 15.8.3
Dynamic Models: 15.8.4

Wooldridge, J. (2005), “Simple Solutions to the Initial Conditions Problem for Dynamic, Nonlinear Panel Data Models with Unobserved Heterogeneity,” Journal of Applied Econometrics 20, 39-54.

Arulampalam, W. and M. Stewart (2009), “Simplified Implementation of the Heckman Estimator of the Dynamic Probit Model and a Comparison with Alternative Estimators,” Oxford Bulletin of Economics and Statistics 71, 659-681.

Multinomial Responses: 16.2.4
Ordered Responses: 16.3.4

Pooled Tobit: 17.8.1
Unobserved Effects Tobit Models: 17.8.2
Dynamic Unobserved Effects Tobit Models: 17.8.3
Count Data Models, Fixed Effects: 18.7.4
Count Data, Dynamic Models: 18.7.5

Arellano, M. and B. Honoré (2001), “Panel Data Models: Some Recent Developments,” in Handbook of Econometrics, Volume 5.

8. Treatment Effect Estimating Assuming Unconfoundedness

Sections 21.1, 21.2, 21.3, 21.5

Imbens and Wooldridge (2009), Lectures 1 and 2

Imbens, G.W. and J.M. Wooldridge (2009), “Recent Developments in the Econometrics of Program Evaluation,” Journal of Economic Literature, 47(1): 5–86.

Abadie, A. and G.W. Imbens (2011), “Matching on the Estimated Propensity Score,” mimeo, Harvard University Department of Economics.

Imbens, G.W. and K. Kalyanaraman (2011), “Optimal Bandwidth Choice for the Regression Discontinuity Estimator,” mimeo, Harvard University Department of Economics.

9. Estimating Treatment Effects under Self Selection

Section 6.2, Section 15.7.2, Section 16.2.3, Section 17.5.2

Imbens and Wooldridge (2009), Lecture 5

Imbens and Wooldridge (2009), Lecture 14

Blundell, R. and J.L. Powell (2003), “Endogeneity in Nonparametric and Semiparametric Regression Models,” in Advances in Economics and Econonometrics: Theory and Applications, Eighth World Congress, Volume 2, M. Dewatripont, L.P. Hansen and S.J. Turnovsky, eds. Cambridge: Cambridge University Press, 312-357.

Blundell, R. and J.L. Powell (2004), “Endogeneity in Semiparametric Binary Response Models,” Review of Economic Studies 71, 655-679.