Two-Wave Panel Analysis: Comparing Statistical Methods for Studying the Effects of Transitions
Johnson, David, Journal of Marriage and Family
Study of the effect of transitions on individual and family outcomes is central to understanding families over the life course. There is little consensus, however, on the appropriate statistical methods needed to study transitions in panel data. This article compares lagged dependent variable (LDV) and change score (CS) methods far analyzing the effect of events in two-wave panel data. The methods are described, and their performances are compared both with a simulation and a substantive example using the National Survey of Families and Households two-wave panel. The results suggest that CS methods have advantages over LDV techniques in estimating the effect of events on outcomes in two-wave panel data.
Key Words: longitudinal studies, panel studies, regression.
Research on the consequences of family transitions and events impacting families has received considerable attention in the literature. As individuals move through the life course, many of the key family variables of influence relate to transitions, such as marriage, divorce, the birth of a child, and the death of a spouse. Other life events can also impact family outcomes, such as job loss, residential change, major illness or disability, and retirement. Given the central role of understanding the consequences of such changes in the development of family theory, it is surprising to find that the methods used to test empirically hypotheses about these consequences often have not been informed by the more recent developments in quantitative methods for the analysis of change. This state of affairs in part reflects conflicting advice found in the methodological literature and the difficulty in selecting a statistical method that best corresponds to the underlying processes being estimated. Choice of method often involves a trade-off between statistical power and potential bias in the estimators. Making the choice that yields the most valid findings is often difficult because of the lack of clear guidelines or empirical comparisons providing estimates of the consequences of these trade-offs.
The purpose of this article is to compare two alternative methods of analyzing the effects of events (transitions) on outcomes using data from a two-wave panel (prospective) study. Focusing on two-wave panels may seem archaic at first, as recent developments in panel analysis, including growth curve models, longitudinal multilevel linear models, mixture models, and random- and fixed-effects pooled time-series models, have focused on the analysis of three or more waves. Although many of these techniques can be applied to two-wave data (e.g., random- and fixed-effects pooled time-series models), several of the more analytically powerful features of these models require three or more waves (e.g., latent growth curve models). The availability of three or more waves provides substantially more information about the form and structure of the change process and allows the researcher to test hypotheses that are untestable with two waves (Allison, 1994; Cole & Maxell, 2003; Johnson, 1995a, 1995b). Although multiple waves have distinct advantages, I focus on two waves for several reasons. An informal review of recently published panel studies in the family area found that a majority relied on two-wave analysis models, in some cases even when multiple waves were available. This, in part, reflected the availability of large representative two-wave data sets, such as the National Survey of Families and Households (NSFH), but use of two-wave methods was also a function of the research questions examined, particularly those assessing the effects of events or transitions on outcomes. For example, Williams and Umberson (2004) chose to analyze a threewave data set with two-wave models (Wave 1 vs. Wave 2, and Wave 2 vs. Wave 3).
Two-wave data are most commonly used by family researchers to control for the prior levels of the dependent variables in the study. For example, in studying the effect of the transition to divorce on perceived health, the researcher is interested in holding constant prior levels of perceived health, as it is possible that the health levels of those divorcing were initially lower than the levels of those who remained married. …