Measuring Change with Longitudinal Data: Challenges and Strategies for Researchers

BACKGROUND AND PURPOSE.  Examining change among research participants is an endeavor sought by many researchers.  Thus, doing it well is an important undertaking for researchers. Ultimately, in the best possible scenarios, the more accurately change is measured and understood, the greater the chance such research will have a positive impact for the population studied; understanding the predictors and correlates of change are a valued and necessary component.  While there is no single methodology identified as superior for the analysis of longitudinal data, an investigator may select from  a set of methods based on the objectives of the study, the nature of the data, and the underlying assumptions of the analysis) (Arellano, 2003). These methods include:  1) raw change; 2) analysis of covariance/residualized change; 3) repeated measures; 4) multilevel modeling; and, 5) linear growth modeling.  The characteristics of each method are discussed, with a conclusion that each method can be appropriate under certain circumstances.  The purpose of this workshop is to apply what is known (from published research and the authors’ experiences) about measuring change when using longitudinal data.

SESSION FORMAT & CONTENT.  The educational methods will include lecture, illustration, demonstration, discussion, and, exercises.  This workshop examines five methods for measuring change involving longitudinal data.  The authors used data from the National Survey of Child and Adolescent Well-being (NSCAW) to examine influential factors in depression change among youth in the CWS. The authors demonstrate how the choice of research questions guides the selection of an analytic strategy, and how the specific analysis implemented directly impacts the results and their interpretation. Methodological requirements such as satisfying reliability assumptions and handling the correlation between the change score and its initial component measure are discussed, as well as how these assumptions, when left unaddressed and/or violated, can lead to statistical errors and incorrect conclusions about the data (Bergh & Fairbank, 2002; Cronbach & Furby, 1970; Linn & Slinde, 1977; Lord, 1963).  The authors also demonstrate strengths and weaknesses of each statistical approach and discuss when it is and is not appropriate to use each one when analyzing longitudinal data.  All participants will receive a resource and reference list and a copy of syntax files for workshop illustrated methods.

WORKSHOP OBJECTIVES.  As a result of this session, participants will achieve the following objectives: (1) understand different methods for examining change when using longitudinal data; (2) understand the strengths and weaknesses of each method; (3) understand how to satisfy reliability assumptions and how to manage correlated data; and, (4) recognize the impact of ignoring or violating statistical assumptions on outcomes.

References

Arellano, M. (2003). Panel data econometrics. New York: Oxford University Press.

Bergh, D.D. & Fairbank, J.F. (2002).  Measuring and testing change in strategic management research. Strategic Management Journal, 23(4), 359-366.

Cronbach, L.J. & Furby, L. (1970). How should we measure “change” – Or should we?  Psychological Bulletin, 74, 68-80.

Linn, R.L. & Slinde, J.A. (1977).  The determination of the significant of change between pre and posttesting periods. Review of Education Research, 47, 121-150.