Methodology and Statistics
Faculty of Social and Behavioural Sciences
Academic webpage Jeroen Mulder
Concerning Causes: Evaluation of Methods to Study Causes and Their Effects in Developmental Processes
On the website of the Consortium Individual Development (CID) it is stated that: “CID examines how the environment (…) and child characteristics (…) affect the development of social competence (SC) and behavioral control (BC), skills that are essential for functioning in society and for reducing risk of behavioral and emotional problems.”.
Trying to understand how diverse factors result in individual differences in developmental outcomes, implies a strong interest in the causal mechanisms that drive the developmental processes. However, from a methodological point of view, studying causality in the context of developmental processes is a challenging task. While experiments based on random assignment form a powerful tool to study causal mechanisms, most (human) developmental processes cannot readily be studied using this methodology for practical or ethical reasons. In contrast, using correlational research—such as cross-sectional and longitudinal panel studies—to study causal mechanisms, is hampered by the threat of omitted variables. Researchers are keenly aware of this problem, and therefore tend to avoid the use of strong causal language in the context of non-experimental research (Hernán, 2018); however, this does not make them less interested in it.
The aim of the current PhD project is to: a) investigate the extent to which CID research is based on a causal interest versus an interest in mere description and/or prediction; b) investigate how particular experimental designs that include mediation can be used to study causal relationships, but also what threats may exist in this context; c) investigate how diverse longitudinal models may align with research questions about the causal connection between developmental processes; and d) investigate how instrumental variables may be put to use in CID to allow for causal inference.
This project is divided into 4 parts. Subproject 1 will consist of a literature review that is concerned with the current aims and practices within social science research. A preliminary review of a random sample of 100 CID publications (published before 2019) reveals that many researchers are interested in causal inference. 18 studies employed an experimental design by including some sort of manipulation and 15 of these studies drew causal conclusions. In contrast, 49 studies used correlational data. Of these studies, however, 23 expressed an interest in causal inference in the introduction, and 18 studies recommended future research into causal effects.
Subproject 2 will focus on the use of experimental manipulations in which the interest is in determining to what extent the effect of the treatment is mediated through another variable. Due to the possibility of unmeasured confounding, a causal interpretation of the relationship between mediator and outcome is problematic. Therefore decomposing the total effect into a direct and indirect effect is problematic as well (VanderWeele, 2015; Cox, Kisbu-Sakarya, Miočević, & MacKinnon, 2013). We will propose a related design that may be able to account for this, and we will perform a simulation study to investigate how well this would work in practice.
Subproject 3 will focus on the relation between two developmental processes and how to investigate possible causal connections between them in longitudinal research. We will view this challenge through the lens of directed acyclical graphs (DAGs) and consider whether including time as a covariate can be considered from a mediation perspective.
Subproject 4 will focus on the potential of using instrumental variables (IV) in the context of CID. Within modern econometrics, IV’s are most importantly used as a solution to the omitted variable bias, and are therefore promising as a solution to some of CID’s methodological issues.
prof. dr. Ellen Hamaker
Consortium Individual Development (CID)
1 May 2018 – 31 April 2023