Advances in research methodology and analytic strategies
Knowledge about social development expanded at an unprecedented rate during modern and near‐contemporary eras and this accomplishment largely was attributable to changes in scientific practice. Two progressive research innovations were particularly influential: (a) the recruitment and utilization of larger and more diverse samples, and (2) the implementation of enhanced longitudinal research designs and data analytic strategies.
Samples and sampling
The pursuit of larger samples was motivated by many factors including the challenge of improving sample representativeness and generalizability and the need to obtain larger ns for complex, multivariate analyses. Evidence of this trend was apparent in the progression of studies that were published on social development from the 1970s to the present.
Gains in sample sizes were accompanied by a movement toward sample diversification. Whereas Caucasian samples predominated in published studies during the early decades of this era, this convention eventually was supplanted by ongoing efforts to increase sample representativeness, validate findings in understudied populations, and ensure that underrepresented groups or strata (e.g., minorities, girls and women, low‐income families) were included in scientific research.
Research methods, designs, and analyses
As implied by the term “development,” the principal aim of empirical research on social development has been to account for change (e.g., describe, predict; ultimately “explain” when, how, or why growth occurs) in child characteristics that are conceived as “social.” Near the inception of this era, investigators commonly made inferences about developmental changes from data obtained with cross‐sectional designs (comparing children of different ages concurrently). Longitudinal studies were less prevalent, relatively narrow in scope (e.g., limited to a few focal variables), and often implemented with follow‐back or follow‐up designs. Even rarer were prospective or follow‐through longitudinal studies in which targeted constructs were progressively tracked (i.e., repeatedly measured) across many months or years.
Advances in developmental theory and multivariate statistics expedited a transition toward longitudinal research and increased the utilization of prospective and cross‐sequential longitudinal designs (Baltes, 1968, 1987). The emergence of developmental process models encouraged researchers to devise and implement multiwave, multivariate longitudinal studies in which assessment intervals were sequenced to capture the intervening processes (e.g., variable associations, interactions, transactions) that were hypothesized to drive developmental change.
In particular, the creation of multivariate statistical tools such as structural equations models (SEM; Jöreskog & Sörbom, 1979) furthered this transition by providing researchers with a means for evaluating complex networks of relations among multiple constructs across multiple occasions (i.e., both within and across time). Moreover, the algorithms contained in these tools (i.e., measurement models) allowed for the estimation of unobserved latent variables and, accordingly, provided investigators strategies for addressing critical measurement objectives such as improving construct specification and validity and evaluating and reducing measurement error (MacCallum & Austin, 2000).
One consequence of these developments was that researchers incorporated larger numbers of variables into their longitudinal investigations. Another was that they began to propose and test more complex patterns of construct relations (e.g., stability, invariance, independence/collinearity of constructs; direct and indirect effects; mediated, moderated, and cross‐lagged associations) that corresponded to (and provided a test of) hypotheses about developmental processes, mechanisms, pathways of influence, and so on.
Full‐panel prospective longitudinal designs increasingly were utilized because the data generated were well‐suited for evaluating hypotheses about emergent, shifting, and continuing pathways of influence among constructs (e.g., stabilities, alternative directions of effect, bidirectional effects, mediated effects, transactions) as specified within developmental process models. Investigators who utilized these designs assessed all focal constructs plus relevant covariates (e.g., control variables) at all times of measurement (i.e., repeatedly across assessment waves) and often relied on statistical tools such as cross‐lag panel models (CLPM; Kenny & Harackiewicz, 1979) to analyze their data. Among the advantages of CLPM, relative to its multivariate forerunners (e.g., multiple regression, path analysis) was the ability to construct and utilize latent variables, estimate multiple pathways over time including mediated pathways, and simultaneously estimate alternative directions of effects.
Other, more recent analytic innovations improved researchers’ ability to describe, distinguish, and quantify intra‐individual (within‐person) growth patterns (i.e., trajectories). Scientists used tools such as latent growth modeling (LGM; Bollen & Curran, 2006; Meredith, & Tisak, 1990) to map developmental trajectories and index specific trajectory parameters (e.g., initial status, shape or contour, and direction of change over time) for many dimensions of social development. Common applications included using LGM to: (a) contrast the growth trajectories of different classes of children (e.g., individuals manifesting different temperamental or behavioral characteristics); (b) assess whether growth on one variable was related to growth on another variable (trajectory covariance); (c) determine whether particular background variables (e.g., organismic or contextual factors) predicted specific aspects of children’s trajectories (e.g., rate of growth); and (d) test hypotheses about developmental continuities and discontinuities.
Among the latest advances made in the analysis of longitudinal panel data stem from tools such as Latent Curve Model with Structured Residuals (LCM‐SR, Curran et al., 2014), Autoregressive Latent Trajectory with Structured Residuals (ALT‐SR, Berry & Willoughby, 2017), and Random Intercept Cross‐Lag Panel Model (RI‐CLPM; Hamaker et al., 2015). Developmental scientists began to utilize these tools because they overcame a basic limitation inherent within CLPM: the aggregation of between‐person and within‐person covariance (Berry & Willoughby, 2017; Curran et al., 2014).
CLPM proliferated because it offered researchers an analytic framework for testing hypotheses about alternative pathways of influence (e.g., by simultaneously estimating cross‐lagged associations among constructs). Eventually, however, questions arose about the reliability of CLPM findings. Principally, criticism stemmed from the fact that, due to the aggregation of between‐person and within‐person covariance, significant cross‐lagged paths could emerge absent of any systemic within‐person increases or decreases in the predicted variable. This being the case, it was argued that CLPM could overestimate (or underestimate) cross‐lagged paths and thereby cause investigators to misinterpret the nature and strength of predictive links.
Strategies such as ALT‐SR overcame this limitation by utilizing repeated measurements to apportion the variance of each variable to stable between‐person differences (i.e., a latent intercept), between‐person differences in linear change (i.e., latent slope), and within‐person deviations from the estimated linear trajectory (i.e., structured residuals). Covariances between latent intercepts reflect stable between‐person associations, potentially due to unmeasured covariates or patterns established earlier in development. Cross‐lagged associations between structured residuals indicate that deviation from one’s developmental trajectory on one variable (e.g., an increase over and above what would be expected) is predictive of deviation from one’s developmental