Cross-sectional age variance extraction: What's change got to do with it?

Abstract

In cross-sectional age variance extraction (CAVE), age, the indicator of a hypothesized developmental mechanism, and a developmental outcome are specified as independent, mediator, and target variables, respectively, to test hypotheses about behavioral development. We show that: (a) longitudinal change in a mediator variable accounting for substantial cross-sectional age-related variance in the target variable need not correlate with the target variable's longitudinal change; and, conversely, (b) longitudinal change in a mediator not sharing cross-sectional age-related variance with the target variable may nevertheless correlate highly with that variable's longitudinal change. We discourage use of CAVE for testing multivariate hypotheses about behavioral development.Keywords: cross-sectional data, longitudinal data, variance partitioning, correlated change, longitudinal mediator A central goal of developmental psychology is to identify the number and nature of causes that drive age-related changes in

Behavior and Neural Correlates of Empathy in Adolescents

“…Furthermore, it would be interesting to investigate the predictive role of age in combination with specific, context-dependent and individual empathic abilities for prosocial behavior. In order to detect age-related changes, future longitudinal studies should examine whether these relations are based on trait versus state characteristics and their stability patterns across adolescence [45]. …”

Life contexts make a difference: Emotional stability in younger and older adults.

“…A second particular strength of this study is its focus on matching rather than regression analysis. Admittedly, results of the matching procedure as applied in this study are no closer to causality and longitudinal change than the results generated with regression analysis in cross-sectional data sets (for criticisms of the use of cross-sectional data for generating insights on change and mediation, see Lindenberger et al, 2011;Maxwell & Cole, 2007). Nevertheless, matching procedures are advantageous in comparison with multiple regression analysis as detailed in the Introduction and Method sections (numerous contextual variables were adjusted for simultaneously and before testing the effect of age on emotional stability; no linearity assumptions on the relationship between covariates and the outcome variable were needed; it was not necessary to extrapolate, which would rely on the assumption that the same relationships exist between variables across subgroups and for those expressions of variables that are outside a range common across subgroups-an assumption that is often criticized as unrealistic).…”

“…Although matching as carried out in this study cannot overcome the limits of cross-sectional comparisons for establishing causal relationships between study variables (Lindenberger, von Oertzen, Ghisletta, & Hertzog, 2011;Maxwell & Cole, 2007), this analytical procedure has three important advantages over multiple regression: (1) Numerous variables characterizing daily life can be adjusted for simultaneously, which corresponds to the notion that affective experiences in different age groups occur in life contexts that differ in many regards; (2) age-related differences between matched and nonmatched samples can be compared without necessarily meeting the assumptions formulated in multiple regression analysis, in particular, on linearity among all variables (i.e., adjustment is separated from this analytical step); and (3) the matching procedure deleted the observations that would require substantial extrapolation (Cook et al, 2009;Ho et al, 2007). In multiple regression, the adjustment for covariates can lead to estimates of group differences that imply values on the covariates for which no observed cases actually exist in one or more of the groups to be compared.…”

Getting at Developmental Processes Through Experiments

“…In the current context, I would like to mention only the problem of inferring causality and of identifying psychological processes resulting in an association between different variables, including age-related differences in such associations. Mediation analyses are generally not well suited to test psychological processes (see Fiedler, Meiser, & Schott, 2011) and extracting age variance in crosssectional designs in an attempt to test developmental processes is even less appropriate (Hofer & Sliwinski, 2001;Lindenberger, Oertzen, Ghisletta, & Hertzog, 2011). Given Experiments targeting developmental processes-5 these difficulties in investigating causal mechanisms driving age-related changes and provide insights into developmental processes, how can the field move beyond the description of age-related differences or age-differential covariations of different variables?…”