Methodological and sampling considerations for creating global norms

• In Event: Creating Global Developmental Norms for Young Children: Opportunities and Challenges

Abstract

Building on the work of Paper 1, this paper outlines the methods that will allow us to explore whether, as McCoy (2021) suggests, that there is “universality with specificity” in child development and if global norms are feasible. This work poses some methodological and practical challenges under three main areas:

1) How can we ensure that conditional independence assumption is met?
2) What is the best sampling design to construct a restricted sample of children without known environmental and health constraints on their development?
3) How can we adequately model the distribution of conditional-on-age scores? Are two moments sufficient?

Global norms are only viable if restriction by the criteria discussed in Paper 1 create a conditionally independent sample. If children develop similarly across contexts conditional on a set of criteria that includes only those without significant environmental and health constraints, then norm reference comparisons can be made without collecting a representative, global sample. Testing this assumption is thus a prerequisite for further global norming work. We discuss the planned psychometric exploration of the test as a whole and our specific plans to investigate differences in developmental trajectories by country.

The sampling question largely rests on examining the tradeoffs between a cross-sectional design and a longitudinal design. A cross-sectional design has the advantage of, by design, ensuring independence of observations at the child-level. By contrast, a longitudinal design induces dependencies that reduce the precision of estimated means and variances. However, longitudinal designs also present favorable practical opportunities by reducing the number of required children and also allow for the exploration of variations in individual trajectories. We review these tradeoffs and discuss the results of simulations that led us to pursue a longitudinal design.

This paper concludes by briefly discussing the methods we plan to employ to estimate the age-specific centiles to create Development-for-Age Z-score (DAZ) scores. Other work has focused on the first two moments of the age-conditional distribution, but focusing on the first two moments may be inadequate. We overview the GAMLSS framework (Stasinopoulos & Rigby, 2007) we plan to apply to fit a univariate distribution to the conditional-on-age distributions of GSED D-scores.

After discussing the normative constraints for creating a sample of children without significant health and environmental constraints to their development in Paper 1, this paper summarizes the methodological processes and challenges inherent in global norming work. We discuss the important work of assessing the viability of the conditional independence assumption, balancing practical and statistical considerations in sampling, and how modeling distributions beyond the mean and variance may be necessary to create valid DAZ scores.

Group Authors

GSED Team

Authors

• Jonathan Seiden, Harvard University
  Presenting Author

• Gareth McCray

• Stef van Buuren, 1) TNO, Leiden 2) Utrecht University

• Gillian Lancaster, Keele University

• Marcus Waldman, Univeristy of Nebraska Medical Center