Poster #122 - Modeling Development When the Construct Itself Changes Over Time: A Longitudinal Measurement Invariance Methodological Study

• In Event: 2-163 - Poster Session 08
  In Poster Session: PS 08 Section - Methods, History, Theory

Integrative Statement

When attempting to study change over time, often it is not developmentally appropriate to measure a skill or behavior the same way across childhood. For example, when measuring reading achievement in a 4-year-old, the developmentally appropriate assessment consists of identifying letters, whereas for a 15-year-old, a reading comprehension assessment is appropriate. Studying children’s growth trajectories via longitudinal data provides important insights in developmental research. However, statistical models designed for studying change over time within individuals assume that the phenomenon of interest can always be measured the same way. Meeting this measurement equivalence assumption is known as achieving longitudinal measurement invariance.
As illustrated by the reading development example, when different measurements must be used across time to adjust for developmental appropriateness, meeting this assumption can become infeasible. Longitudinal data on children’s academic, behavioral, and socioemotional development can easily violate longitudinal measurement invariance when the meaning of the construct changes as children develop. This is of concern since violation of longitudinal measurement invariance is a common reality which can distort research results, the consequences of which may include inaccurate inferences. This is especially of concern since social consequences can ensue when inaccurate results are used to inform parenting and schooling interventions or educational policy and funding decisions. For example, using longitudinal models to examine which environmental factors might explain who will struggle to read initially and who will grow slower as readers can help target research-based prevention and intervention efforts. The logic of this argument, however, rests on the assumption that the research models used are adequate and yield research conclusions that are well-founded.
Although the methodological literature makes clear that violation of longitudinal measurement invariance is problematic, few methodological studies have considered real world conditions mimicking data on child development. Consequently, few guidelines exist which account for nuances specific to the study of child development. The latent growth curve model is a longitudinal model commonly used when studying how children develop and what predicts their growth and development. The main goal of our study is to systematically examine the robustness of the latent growth curve model to violations of longitudinal measurement invariance. We conduct a simulation study designed to examine the degree of distortion of research results under varying violations of longitudinal measurement invariance. By varying the level, location, and amount of invariance, we examine under what conditions violation of longitudinal measurement invariance risks erroneous substantive conclusions, using datasets generated to mimic data on developmental processes across childhood. We explore the consequences of violating this assumption to provide insight regarding under what conditions researchers can trust the results of their analysis and under what conditions to exercise caution when drawing research conclusions regarding child development. We hope our research will provide insights for developmental researchers by helping to establish guidelines for how to address developmental research questions more precisely when using these models.

Authors

• Rachel A. Gross, Indiana University - Bloomington
  Presenting Author

• Dubravka Svetina, Indiana University - Bloomington
  Non-Presenting Author