This project develops new statistical tools for economists and other social scientists interested in studying individual behavior and outcomes over time. These tools will be used to answer empirical questions related to the joint retirement decisions of husbands and wives and to questions in the area of health economics.
The research is centered on three general methodological themes. The tools that are developed within each of these will be applicable in a number of areas in economics and the social sciences.
The first theme involves situations in which a researcher is interested in the duration of some event. The tools developed in this part of the project will be useful in many different areas of economics: in labor economics, to study the duration of unemployment; in public economics, to study participation in welfare programs; in marketing, to study the long-run effects of advertising; and in health economics, to study the evolution in health status. The innovation relative to existing methods, is that this project will explicitly focus on situations in which two individuals choose their durations jointly. For example, a husband and a wife may jointly decide on their retirement times taking their own as well as their spouses preferences into account. This joint decision will lead to an interaction between the two durations that is fundamentally different from the interaction that one would see if the durations were chosen by, say, two competing firms. Both types of interactions imply that the impact of a policy that affects one person?s behavior, will be enlarged or diminished by the fact that the person?s behavior will in turn influence others. It is therefore important to have tools for measuring the importance of these interactions.
The second general scenario addressed in this research involves situations in which two outcomes interact with each other over time. For example, it is well-established that health and socioeconomic status are related, both contemporaneously and over time. In interpreting such correlations, however, it is important to determine whether the correlation exists because a change in one of the two causes the other to change in the future (and, if so, which causes which), or because they are both determined by the same underlying driving forces. The project will continue the development of methods that can be used to answer such questions.
The third theme is smaller and more focused. It addresses statistical problems in situations in which a researcher is interested in durations, but only has data on the durations if they fall in some interval. It turns out that this is relevant for studying mortality (and how it related to various economic factors) because certain data sets only contain the dates of death if it occurs over the period in which that data was collected. However, the methodological contributions made here will have applications that are much more general.
Broader impacts: The most direct broader impact comes from the fact that software for the new statistical tools will be made available online, and from the training of the students involved in the project. The project will also improve the way economists and other social scientists think about their empirical findings.
This project has made contribution to the econometric analysis of nonlinear panel data model and of duration models. It has produced both methodological and empirical findings in both areas. The empirical finding are closely connected with the methodological breakthroughs. It is therefore most natural to integrate them in the discussion below. I have continued to develop new techniques for dealing with panel data models. These are statistical models that are appropriate for analyzing data sets in which a number of individuals (or firms or countries, etc.) are observed in more than one time period. The most important specific contribution here is to a set of statistical models called regression model with double censoring and truncation in a fixed-effects setting. This is a (surprisingly cumbersome) extension of some of my old work on panel data censored and truncated regression models. The techniques that have been developed here have potential use in many areas of applied economics. I have complemented my theoretical research on panel data with empirical research in macroeconomics. From the point of view of my own research agenda, this serves as an illustration of the econometric methods that I develop, but it is also interesting in its own right. Indeed, it was a collaboration with two research at the Swiss National Bank, Daniel Kaufmann and Sarah Lein, whose research interest are entirely in the area of macroeconomics. In this work, we modified existing econometric methods so that they can be used to cast new light on price setting behavior. Our contribution relative to the existing literature is to propose an estimator that relaxes distributional assumptions on the unobserved heterogeneity. We the use the estimator to examine the prevalence of positive price changes in a low-inflation environment. The main contribution to the econometrics analysis of duration data has been concerned with the formulation of 'simultaneous equations duration models' in which the timing of two events can interact with each other in an economically sensible way. This is joint work with Aureo de Paula, and it has applications in many areas of economics. To illustrate the contribution, we apply our simultaneous equations duration model to the retirement decisions of husbands and wives. Using data from the United States, we find some evidence that couples coordinate their retirement decisions in a way that is consistent with our econometric framework. Specifically, we find and that there is some preference for simultaneous retirement. The project period 10/01/2013 to 09/30/2014 has primarily been devoted to the construction of new econometric methods that can been applied to estimate panel data models in which one wants to control for both time-varying and time-invariant explanatory variables. A classic solution to this already existed for linear models. My research (with Michaela Kesina) has been concerned with extending this to more general nonlinear models. This has resulted in a new set of statistical methods. We illustrate the usefulness of the research by applying the new methods to topics as disperse as the annual number of doctors’ visits to the level of cross-country trade-flow. In this project period, I have also extended my research on 'simultaneous equations duration models' by applying our methodology to data sets covering various European countries and by comparing the results from these to the results from the Unites States.
