Comparison of multivariate approaches in meta-analytic factor analysis of the ucla loneliness scale

In the classical meta-analysis approach, studies focusing on a single effect size are in the majority. However, in recent years researchers have stated that focusing on a single effect size limits the capacity to contribute to the development of conceptual theories. In recent years, it has been tried to develop model-based meta-analysis techniques that try to simultaneously analyze the complex set of relations and deal with latent variables and measurement errors, in order to be more suitable for studies in social sciences. In this context, it has been observed that interest in multivariate effect sizes and complex models in which meta-analytical techniques and structural equation modeling (SEM) are synthesized have increased. This technique, also called model-based meta-analysis, in which SEM studies are synthesized with meta-analysis, has been also expressed as meta-analytic structural equation modeling (MASEM). Although meta-analytic CFA or MASEM is a new research area, its use is increasing day by day in the international literature. Especially in the modeling of psychological structures, the use of techniques under SEM (correlation analysis, regression analysis, ANOVA, MANOVA, factor analysis) made it more important to investigate methods related to model-based meta-analysis. In this study, it was aimed to conduct a meta-analysis of the primary studies in which the correlation matrices of the factorial structure of the UCLA loneliness scale, which has been one of the most frequently used scales to measure the loneliness structure in the field of psychology and social sciences. The most suitable model for the factorial structure of the UCLA loneliness scale was compared according to three different models and estimation methods. The study group of the research consisted of primary studies (articles, theses and papers) that meet the criteria determined by the researcher. At the end of the identification, screening, eligibility and inclusion processes, 52 studies were included in the meta-analysis. Two points required for MASEM were checked prior to data analysis: correlation matrices and sample sizes of primary studies. For the first stage of the analysis, correlation matrices were synthesized under the random effects model according to the GLS and TSSEM estimation methods. Model parameters and model fit indices were obtained and compared according to two different methods for the three factorial structures proposed in the second stage of MASEM. The findings of the study showed that the two-factor and second-level three-factor models were the models that best fit the synthesized correlation matrix for UCLA loneliness scale. In addition, when the performance between MASEM methods was compared, it was seen that the TSSEM method performed better than the GLS method. In this study, the effect of moderator variables that could explain the heterogeneity of correlation matrices was also investigated. When the results obtained for the moderator variables of sample size and sample age is summarized, it was seen that the two-factor and the second-level three-factor model fit better than the one-factor model, in parallel with the analyzes conducted with all studies. However, considering the model fit indexes and error indexes obtained from the second-order three-factor model, it can be said that the second-order three-factor model is the model that best fits the data. In addition, it can be said that the second-order three-factor structure of the UCLA loneliness scale did not change in the subgroups formed according to the moderator variables of sample size and sample age. As a result of the meta-analytic CFA analysis and based on findings in the literatüre, it was seen that the most suitable model for the factor structure of the UCLA loneliness scale is the three-factor model in which the second level global loneliness factor is. For practitioners who plan to use the UCLA loneliness scale in their studies, it is recommended to use the scale's three-factor structure.