Panel data analysis in comparative politics: Linking method to theory

Abstract

Abstract. Re-analyzing a study of Garrett and Mitchell ('Globalization, government spending and taxation in the OECD', European Journal of Political Research 39(2) (2001): 145-177), this article addresses four potential sources of problems in panel data analyses with a lagged dependent variable and period and unit dummies (the de facto Beck-Katz standard). These are: absorption of cross-sectional variance by unit dummies, absorption of time-series variance by the lagged dependent variable and period dummies, mis-specification of the lag structure, and neglect of parameter slope heterogeneity. Based on this discussion, we suggest substantial changes of the estimation approach and the estimated model. Employing our preferred methodological stance, we demonstrate that Garrett and Mitchell's findings are not robust. Instead, we show that partisan politics and socioeconomic factors such as aging and unemployment as expected by theorists have a strong impact on the timeseries and cross-sectional variance in government spending.

Partisan politics: The empirical evidence from OECD panel studies

“…Experts usually discuss econometric issues such as serial correlation or the heteroskedasticity of the error terms. To what extent one may well arrive at wrong conclusions when econometric models are not well specified has been discussed, for example, by Kittel and Winner (2005) and Plümper et al (2005). Many of the OECD panel data studies include fixed country effects to deal with unobserved heterogeneity and assume that all other parameters are identical across countries.…”

Exploiting the Cracks: Wedge Issues in Multiparty Competition

“…17 A common alternative for dealing with serial correlation is the inclusion of a lagged dependent variable on the right-hand side of the equation (Beck and Katz 1995). Yet, more recent work recommends the Prais-Winsten solution, which we used to address the panel specific AR(1) error structure, as a lagged dependent variable absorbs a large part of the trend in the dependent variable and likely biases the estimates (Achen 2000;Greene 1990;Plümper, Troeger, and Manow 2005). Finally, by means of tests we ensured that the dependent and independent variables were stationary (Asteriou and Hall 2007).…”

Spinning the Globe? U.S. Public Diplomacy and Foreign Public Opinion

“…We also considered adding a lagged dependent variable. Although a lagged dependent variable often serves as a powerful control, some argue that its inclusion is often an atheoretical after-thought (Plümper, Troeger, and Manow 2005). Adding it in panel data also introduces methodological complications.…”