TY - CHAP
T1 - Multilevel Mediation Analysis with Categorical Outcomes
T2 - Conducting Causal Inference by the MSEM Framework
AU - Yeung, Jerf W. K.
AU - Chen, CHui-Feng
AU - Low, Andrew You Tsang
AU - Xu, Chi
AU - Xu, Leilei
PY - 2022/6/1
Y1 - 2022/6/1
N2 - Mediation analysis has been a commonly-used statistical procedure to investigate the process or mechanism that links the relationship between an independent variable and a dependent variable through the transmission of at least one intervening variable, usually called a mediator, in a causal order. Nevertheless, in many academic fields, such as social sciences, education, public health, epidemiology, and economics, data are clustered at several levels, which violates the assumption of independent observations in conventional single-level mediation analysis, leading to conflated estimates. Apart from this, a more complicated concern arises when the outcome is categorical, which poses scaling inequivalence and then generates incompatible results. Furthermore, overwhelming existing research presented mediated effects based on cross-sectional data, which deviates the causal chain of events assumed by mediation analysis. Recently, multilevel structural equation modeling (MSEM) provides an efficient and versatile statistical framework to tackle the above-mentioned technical concerns in hierarchical data with categorical outcomes. This chapter uses data from the Children of Immigrants Longitudinal Study (CILS) to demonstrate how to perform MSEM to investigate if academic aspiration of immigrant youths in late adolescence mediates both the effects of family socioeconomic status and school location at individual and school level in early adolescence on their later successful college graduation in young adulthood. Discussions focus on the applications and flexibilities of using MSEM framework in mediation analysis with clustered data.
AB - Mediation analysis has been a commonly-used statistical procedure to investigate the process or mechanism that links the relationship between an independent variable and a dependent variable through the transmission of at least one intervening variable, usually called a mediator, in a causal order. Nevertheless, in many academic fields, such as social sciences, education, public health, epidemiology, and economics, data are clustered at several levels, which violates the assumption of independent observations in conventional single-level mediation analysis, leading to conflated estimates. Apart from this, a more complicated concern arises when the outcome is categorical, which poses scaling inequivalence and then generates incompatible results. Furthermore, overwhelming existing research presented mediated effects based on cross-sectional data, which deviates the causal chain of events assumed by mediation analysis. Recently, multilevel structural equation modeling (MSEM) provides an efficient and versatile statistical framework to tackle the above-mentioned technical concerns in hierarchical data with categorical outcomes. This chapter uses data from the Children of Immigrants Longitudinal Study (CILS) to demonstrate how to perform MSEM to investigate if academic aspiration of immigrant youths in late adolescence mediates both the effects of family socioeconomic status and school location at individual and school level in early adolescence on their later successful college graduation in young adulthood. Discussions focus on the applications and flexibilities of using MSEM framework in mediation analysis with clustered data.
M3 - RGC 12 - Chapter in an edited book (Author)
SN - 978-1-68507-892-8
VL - 31
T3 - Advances in Mathematics Research
BT - Advances in Mathematics Research
A2 - Baswell, Albert R.
PB - Nova Science Publishers
ER -