Of Fixed-Effects and Time-Invariant Variables

Autor: Nathaniel Beck
Rok vydání: 2011
Předmět:
Zdroj: Political Analysis. 19:119-122
ISSN: 1476-4989
1047-1987
DOI: 10.1093/pan/mpr010
Popis: What follows is a longish controversy (two critiques, a reply and two rejoinders) over the quality of the estimates and associated SEs provided by Plumper and Troeger's (2007) "fixed-effect vector decompo sition" (FEVD) procedure; Plumper and Troeger (PT) will refer to that article and not any persons. My role is to lay out some issues that separate the authors rather than to adjudicate between them. As with many controversies, a bit of heat is generated along with some light. Readers care a bit less than the authors about what was said when, but they do care a lot about what appropriate method to use when a panel data model has both unit-specific intercepts' and variables that are invariant over a unit. Thus, I also take it upon myself to discuss some things that I gleaned from this controversy; this discussion has a bit less heat than what follows, but of course readers should judge the evidence for themselves. Why should anyone care about this controversy? Panel data always suffer from some potential omitted variable bias, bias due to a lack of measured data on important attributes of the individual or country that do not change over time. Thus, in labor economics, we usually cannot measure ability, but we know it is an important determinant of wages and it is likely to be correlated with other important variables (education), thus leading to omitted variable bias. Similar problems occur in comparative political economy, where institutional features of a country may be important but difficult to measure. The standard solution is to adjoint unit-specific intercepts to the specification. At the same time, there are variables we care about that do not vary over time (time-invariant variables [TIV's]). In labor economics, these would be individual attributes such as gender, race, and education; in comparative political economy, these are often institutional variables related to the electoral system or issues of central bank independence.2 Scholars well beyond the authors in this controversy care a lot about the issues raised. The first econo metric solution to the problem, instrumental variables, was proposed three decades ago by Hausman and Taylor (1981). HT, as I shall denote it, has almost 1500 Google Scholar citations; PT has about 250 such citations. Many scholars using panel data not only care about the effects of TIV's but also must rely on unit-specific intercepts to deal with unobserved heterogeneity. Users of FEVD want to know whether it performs appropriately (both in terms of coefficient estimates and SEs) when applied to data that appear to fit its underlying assumptions. They also want to know if there is something new and different about the method. Both these issues are discussed analytically by Greene. As we shall see, the short answers are that FEVD performs well when its assumptions are met and that there is little new in this method. Researchers also want to know whether the assumptions behind FEVD are plausible, and how FEVD compares to other methods that make different assumptions. This is the basis of the very lucid Breusch, Ward, Nguyen, and Kompas contribution, and to my mind, this is the most
Databáze: OpenAIRE
Externí odkaz: