Suchergebnisse

My project assessed different poverty indicators and their correlates in Laos. I was interested in poverty-related variables on the household level (such as household assets and health) and village level (such as infrastructure and population). Therefore, I chose a multilevel model that estimates connections on both these levels simultaneously. Multilevel modeling was necessary also because the sampling was clustered and hence, the observations (households) were not independent of each other. As I searched for quantitative methods that could flexibly integrate multilevel modeling with ordinal or non-normally distributed data, I encountered the idea of latent variable modeling, often referred to as structural equation modeling. In practice, structural equation modeling can be complex but also rewarding because it provides a vast range of options for model building and comprehensive tools for evaluating models as a whole. Structural equation modeling usually features latent variables or factors, meaning that observed variables are not assumed to be complete, but they contain measurement error. In this research methods case, I briefly describe the main issues I encountered while conducting a multilevel structural equation modeling analysis for multidimensional poverty and its correlates. I also provide some useful resources for additional information on these methods.

Multilevel structural equation modeling (multilevel SEM) has become an established method to analyze multilevel multivariate data. The first useful estimation method was the pseudobalanced method. This method is approximate because it assumes that all groups have the same size, and ignores unbalance when it exists. In addition, full information maximum likelihood (ML) estimation is now available, which is often combined with robust chi‐squares and standard errors to accommodate unmodeled heterogeneity (MLR). In addition, diagonally weighted least squares (DWLS) methods have become available as estimation methods. This article compares the pseudobalanced estimation method, ML(R), and two DWLS methods by simulating a multilevel factor model with unbalanced data. The simulations included different sample sizes at the individual and group levels and different intraclass correlation (ICC). The within‐group part of the model posed no problems. In the between part of the model, the different ICC sizes had no effect. There is a clear interaction effect between number of groups and estimation method. ML reaches unbiasedness fastest, then the two DWLS methods, then MLR, and then the pseudobalanced method (which needs more than 200 groups). We conclude that both ML(R) and DWLS are genuine improvements on the pseudobalanced approximation. With small sample sizes, the robust methods are not recommended.

Testing for invariance of measurements across groups (such as countries o r time points) is
essential before meaningful comparisons may be conducted. However, when tested, invariance is often absent. As a result, comparisons across groups are potentially problematic and may be
biased. In the current study, we propose utilizing a multilevel structural equation modeling (SEM)
approach to provide a framework to explain item bias. We show how variation in a contextual
variable may explain non invariance. For the illustration of the method, we use data from the
second round of the European Social Survey (ESS). (author's abstract)

The current study explores the reasons for noninvariance of the measurements of gender role attitudes across countries. While previous studies have shown that noninvariance is a problem for comparative research and pointed out methods to alleviate the risks of drawing invalid conclusions, none has so far tried to explain why measurements of gender role attitudes are nonequivalent. Therefore, we use multilevel structural equation modeling to exploring measurement invariance and explain its absence. We use data assessing peoples' views on the specialization of roles by gender and the consequences of female employment on family's well-being from the International Social Survey Programme. We can replicate the findings from prior research indicating that scalar measurement invariance across countries is absent. Furthermore, we use two country-level variables to explain the noninvariance of particular items. The cultural value embeddedness explains noninvariance to a considerable degree while the Gender Inequality Index from the United Nations Development Programme does not. Therefore, we conclude that issues of comparability of gender role attitudes are related mainly to cultural rather than structural differences between countries.