{"id":9325,"date":"2026-04-07T13:56:52","date_gmt":"2026-04-07T13:56:52","guid":{"rendered":"https:\/\/www.proefschriftmaken.nl\/portfolio\/mattis-van-den-bergh\/"},"modified":"2026-04-23T08:15:15","modified_gmt":"2026-04-23T08:15:15","slug":"mattis-van-den-bergh","status":"publish","type":"us_portfolio","link":"https:\/\/www.proefschriftmaken.nl\/en\/portfolio\/mattis-van-den-bergh\/","title":{"rendered":"Mattis Van Den Bergh"},"content":{"rendered":"","protected":false},"excerpt":{"rendered":"","protected":false},"author":8,"featured_media":13359,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"footnotes":""},"us_portfolio_category":[45],"class_list":["post-9325","us_portfolio","type-us_portfolio","status-publish","has-post-thumbnail","hentry","us_portfolio_category-new-template"],"acf":{"naam_van_het_proefschift":"Latent Class Trees","samenvatting":"Er is geen Nederlandse samenvatting beschikbaar. De Engelse samenvatting vind je hier<\/a>.","summary":"People differ, but some people are more alike to each other than to others. Within the social sciences, answer patterns based on variables of interest are used to cluster similar people. For example, based on variables on healthy and risky behavior (e.g., \u2018How often do you sport?\u2019 or \u2018How much do you smoke?\u2019) one could identify clusters of people showing qualitative different risk behavior. A popular method for this is Latent Class (LC) analysis which identifies unobserved homogenous subgroups or classes within a data set. A very important and sometimes difficult issue of LC analysis is the decision on how many classes should be used. This is usually decided by comparing models with a different number of classes on some measure that indicates how well the model describes\/fits to the observed data. The number of classes is increased until the fit measure does not improve. Once the number of classes has been decided, the classes are interpreted based on the class specific probabilities or means.\n\nThis is a theoretically sound procedure, but in applied settings it is often problematic. For instance, when a LC analysis is applied to a large data set (with many respondents and\/or many variables) there is frequently a large number of classes identified. Such a large number of classes is often more specific than intended and interpretation can become very hard. Moreover, the number of classes is only based on fit, as it is very hard to substantively compare models with different numbers of classes. Finally, the fact that different fit measures can indicate a different optimal number of classes does not help. In this thesis LC models with different numbers of classes that are substantively related have been developed. These models are based on the so-called Latent Class Tree (LCT) procedure.\n\nLCT modeling starts as a standard LC analysis by determining whether two classes fit better than one class. If the two class solution is preferred, the two classes are separated in a new data set for each of the classes and every respondent is proportionally assigned to each class. Subsequently a 1-class and a 2-class model are estimated for each of these classes. This procedure is repeatedly applied until only 1-class models are preferred. This results in a hierarchical tree structure of latent classes. The main advantage of this approach is that classes will be substantially related. This allows the use of both statistical and substantive reasoning to decide on a number of classes. These binary LCTs are described in Chapter 2 and illustrated with an empirical data set on social capital.\n\nThe first split of a LCT is the most important split, as the rest of tree depends on this first result. Moreover, in some situations binary splits are too much a simplification and it is important to investigate this. Therefore in Chapter 3 a measure of the relative improvement in fit has been proposed to investigate models with different numbers of classes. Standard fit measures would result in standard LC analysis, but with the measure of relative improvement in fit it can be assessed whether it is preferable to increase the number of classes at the first split of the LCT. For subsequent splits there is more substantive information on the classes available to guide decisions on split sizes, while the relative fit improvement can also be used. The measure of relative improvement in fit is described in Chapter 3 and illustrated with several LCTs on social capital and (post-)materialism.\n\nLC analysis is applied to cross-sectional data (assessed on one moment in time). For longitudinal data latent class growth curves are used. This identifies similar patterns over time, for example respondents with different mood patterns during the day. A large number of classes is quite common with longitudinal data and the tree approach can be very suitable for assessing classes with different patterns over time. Therefore in Chapter 4 the LCT approach has been expanded to also construct so-called Latent Class Growth Trees (LCGT). The LCGTs illustrated with empirical examples on mood regulation during the day and the probability of drugs use given a respondents age.\n\nAssessing the latent classes is usually the first step of a study. Subsequently researchers often want to relate the classes to some external variables. For example, do some classes differ in the amount of men and women or can we predict class membership based on age. A procedure has been developed to compare the distributions of external variables among classes and to predict the class memberships based on external variables. This is applied to every split of a tree and therefore gives a clear overview on how the external variable is related to every class of the tree. This procedure is illustrated for both LCTs and LCGTs in Chapter 5.\n\nThe last chapter discusses possibilities for future research, such as resampling methods or specific fit measures. In principle the traditional LC analysis will be a first starting point for applied researchers, but with this work (and also the development of software to build LCTs) there is now the possibility in difficult situations to use the LCT method to have substantive information to decide on the optimal number of classes.","auteur":"Mattis Van Den Bergh","auteur_slug":"mattis-van-den-bergh","publicatiedatum":"5 januari 2018","taal":"EN","url_flipbook":"https:\/\/ebook.proefschriftmaken.nl\/ebook\/mattisvandenbergh?iframe=true","url_download_pdf":"","url_epub":"","ordernummer":"FTP-202604071353","isbn":"978-94-6295-790-9","doi_nummer":"","naam_universiteit":"Tilburg University","afbeeldingen":13359,"naam_student:":"","binnenwerk":"","universiteit":"Tilburg University","cover":"","afwerking":"","cover_afwerking":"","design":""},"_links":{"self":[{"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/us_portfolio\/9325","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/us_portfolio"}],"about":[{"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/types\/us_portfolio"}],"author":[{"embeddable":true,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/comments?post=9325"}],"version-history":[{"count":1,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/us_portfolio\/9325\/revisions"}],"predecessor-version":[{"id":9326,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/us_portfolio\/9325\/revisions\/9326"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/media\/13359"}],"wp:attachment":[{"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/media?parent=9325"}],"wp:term":[{"taxonomy":"us_portfolio_category","embeddable":true,"href":"https:\/\/www.proefschriftmaken.nl\/en\/wp-json\/wp\/v2\/us_portfolio_category?post=9325"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}