Analysing PIAAC Data with Structural Equation Modelling in Mplus

Abstract

Structural equation modelling (SEM) has become one of the most prominent approaches to testing substantive theories about the relations among observed and/or unobserved variables. Applying this multivariate procedure, researchers are faced with several methodological decisions, including the treatment of indicator variables (e.g. categorical vs. continuous treatment), the handling of missing data, and the selection of an appropriate level of analysis. The PIAAC data pose additional issues, such as the clustering of individual-level data, the large number of participating countries, the representation of performance scores by a set of plausible values, and the differences in the selection probabilities. Therefore, a flexible software package is required to handle them. This chapter introduces readers to analysing PIAAC data with SEM in the software Mplus by (a) presenting the key concepts behind SEM, (b) discussing the complexities of the PIAAC data and their possible handling, (c) illustrating the specification and evaluation of measurement and structural models, and (d) pointing to current developments in the areas of measurement invariance testing and multilevel SEM. Sample input and output files are provided. Structural equation modelling (SEM) represents a broad range of multivariate approaches that allow researchers to test hypotheses related to the means, variances, and covariances of manifest and latent variables (Kaplan 2009). It includes approaches such as path analysis, confirmatory factor analysis, and structural models that are based on researchers' hypotheses and theories about the relations among variables. In his seminal book, Kline (2016) emphasised that SEM requires

The German Multi-Dimensional Perceived Autonomy Support Scale for Physical Education: Adaption and Validation in a Sample of Lower Track Secondary School Students

“…This finding was unusual because CFI and RMSEA normally are equally sensitive to different factor loadings across groups [57]. Above all, with regard to the influence of the type of factor models or the types of invariance models being compared, the cut-offs for maximum change in model fit in invariance testing should not be considered as strict rules [77]. Therefore, multigroup analyses rather speak for measurement invariance across the age groups included in this study.…”

School leadership and achievement gaps based on socioeconomic status: a search for socially just instructional leadership

“…So we conducted a three-level multilevel analysis to examine how student and school-related variables predict student achievement in reading, math and science, controlling for variation due to the countries in question. More precisely, we followed the suggestions for handling large-scale assessment data with a large number of countries made by Muth en and Asparouhov (2018) and Scherer (2020), and thus treated countries as random with schools nested within countries and students nested in schools. However, several issues need to be considered when analyzing complex surveys (Lorah, 2018).…”

“…We handled PVs by writing out ten data sets – one for each PV – and combined these data sets in Mplus using the IMPUTATION function. Thus, analyses are conducted for each set of PVs and combined automatically afterward in Mplus (Scherer, 2020). Scholars also recommend the use of sample weights due to the unequal probability of selection (Lorah, 2018), but it is largely unclear which weights should be used and how to achieve robust results (Laukaityte and Wiberg, 2018).…”

Empathic concern and personal distress, two facets of empathy, have opposing relations to prosocial behavior: A 63-country study

“…This study adopted the Multilevel Confirmatory Factor Analysis approach for the measurement invariance examination, which is recommended when the number of groups is large (Kim et al, 2017;Jak, 2018;Scherer, 2020). Especially when the number of groups is larger than 50 (we have 63 countries in this study), the Multilevel Confirmatory…”