View Item 
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        JavaScript is disabled for your browser. Some features of this site may not work without it.

        Browse

        All of UU Student Theses RepositoryBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

        Automated model specification search in CFA using meta-heuristics

        Thumbnail
        View/Open
        Master_Thesis_ADS_OscarKromhof.pdf (2.437Mb)
        Publication date
        2023
        Author
        Kromhof, Oscar
        Metadata
        Show full item record
        Summary
        [""Confirmatory Factor Analysis is an essential tool in psychometrics to indirectly measure abstract psychological constructs. It is therefore important to have a well fitting CFA model on empirical data-sets. It turns out that in practice many theoretical models don’t fit the empirical data well which could potentially be resolved by model re-specification. Research on the topic has led to the use of meta-heuristics for this task. Numerous meta-heuristics have been proposed for model specification. From previous research the most promising meta-heuristics seem to be the Particle Swarm Optimisation and the Genetic Algorithm. The final goal is to be able to automate the process of model specification. In this thesis we investigate which set of parameters in the objective functions (hyperparameters) for each of the named metaheuristics yield the best result for model specification according to certain fit-measures (CFI, RMSEA, SRMR) that are specific to Structural Equation Models. From the analysis of the selected data-sets from the OSF-website we conclude that having the modelcomplexity penalty term, λ, in the objective-function in the range of [0, 0.2] will lead to descent results. For the slope-parameter we find that values beyond a threshold −b < −100 give the better values for the fit-measures. The threshold values/centring values c1, c2, c3 seem to have a less significant impact on the resulting fit-measures. None of these conclusions should be interpreted as hard cut-off values but rather suggestions for avoiding bad performance of the algorithms in the context of CFA. We also gathered evidence that the Genetic Algorithm is consistently performing better than the Particle Swarm Optimisation on the selected data-sets while taking up less computing time, suggesting that the Genetic Algorithm might be preferred over the PSO for model specification in Structural Equation Modelling""]
        URI
        https://studenttheses.uu.nl/handle/20.500.12932/44957
        Collections
        • Theses
        Utrecht university logo