Panel surveys: Adding the fourth dimension

Abstract

Practitioners and academics are often faced with the problem that some time series of interest are not available at high frequencies. When related series exist, the Chow and Lin (1971) methodology can be used to disaggregate a low frequency series into its high frequency counterpart using the available related series. The aim of this paper is to increase accuracy of the Chow and Lin (1971) methodology by exploiting information from the cross-sectional dimension. Contribution We suggest jointly estimating multiple Chow and Lin (1971) equations, one for each cross-sectional unit, restricting the coefficients to be the same across units in order to interpolate unit-specific data. In contrast to the estimation of single equations, unobservable time-varying characteristics (e.g. structural breaks) that are common across units can be taken into account. The proposed approach is straightforward to implement and can readily be applied to various settings. Results We provide empirical evidence that the panel-based approach can improve accuracy compared to single equation models, in particular when the time dimension of available data is short. Furthermore, the results suggest that controlling for unobservable time-varying characteristics can improve accuracy of the resulting interpolated series. Nichttechnische Zusammenfassung Fragestellung Forscher werden häufig mit dem Problem konfrontiert, dass betrachtete Zeitreihen nicht in hoher Frequenz verfügbar sind. Liegen jedoch verwandte, hochfrequente Zeitreihen vor, kann eine niederfrequente Zeitreihe nach Chow und Lin (1971) disaggregiert werden. Ziel des vorliegenden Papiers ist es, die Genauigkeit des Chow und Lin (1971)-Ansatzes durch Einbeziehung von Querschnittsinformationen zu verbessern. Beitrag Mehrere Chow und Lin (1971)-Gleichungen werden (bei Annahme identischer Regressi-onskoeffizienten für die einzelnen Querschnittseinheiten) simultan geschätzt. Auf Basis dieser Schätzungen können einheitenspezifische Zeitreihen interpoliert werden. Im Ge-gensatz zu einer Schätzung von Einzelgleichungen können unbeobachtbare, zeitvariable Charakteristika (z.B. Strukturbrüche), die allen Querschnittseinheiten gemeinsam sind, berkücksichtigt werden. Der vorgeschlagene Ansatz ist einfach zu implementieren und kann leicht an verschiedene Anwendungen angepasst werden. Ergebnisse Die empirische Evidenz zeigt, dass der Panel-basierte Ansatz inbesondere dann zu ge-naueren interpolierten Zeitreihen führen kann, wenn die verfügbaren Daten eine gerin-ge Zeitdimension aufweisen. Darüber hinaus deuten die Ergebnisse darauf hin, dass die Berücksichtigung unbeobachtbarer, zeitvariabler Charakteristika die Genauigkeit der in-terpolierten Reihen verbessern kann. Abstract Single equation models are well established among academics and practitioners to perform temporal disaggregation of low frequency time series using available related series. In this paper, we propose an extension that exploits information from the cross-sectional dimension. More specifically, we suggest jointly estimating mult...

Are we keeping the people who used to stay? Changes in correlates of panel survey attrition over time

“…A repeated panel survey (Kalton and Citro, 1993), in which wave one and wave two of a panel survey are fielded at different points in time, is needed for this study. The American National Election Study (ANES) fits this criterion.…”

Culture, participative decision making and job satisfaction

“…Thus, responses may be conditioned by the previous experiences of the participant in the study. Due to the nature of humans as research subjects, the very act of observation will transform the behaviour of those being observed (Kalton & Citro, 2000). Panel conditioning was identified as a potential limitation in the study by Morse and Reimer (1956), which took place over one and one-half years.…”

The Long-Term Effect of Material Incentives on Participation in Online Panels

“…For example, unless random sampling from a defined population is done (instead of asking for volunteers), there is no scientific basis for generalizing results from online panels to any larger population. Furthermore, panel conditioning (i.e., changes in respondent answers to later surveys) might occur through panelists' repeated survey participation (see Kalton and Citro 1993;Dennis 2001). Moreover, panels that offer lucrative rewards for study participation may face the challenge of identifying respondents who lie about themselves to fit into the target sample.…”