Dissertations / Theses on the topic 'Latent class method'

INTRODUCTION The dietary pattern approach is useful to study the effect of the overall diet on health outcomes, through considering the network of complex interactions between foods or nutrients. The main methods traditionally used to identify dietary patterns are principal components analysis, factor analysis, principal components factor analysis and cluster analysis. Latent class analysis (LCA) is a latent variable approach, that has some advantages in comparison to the previous methods. Unlike principal component, factor and principal component factor analysis, it can be used to classify individuals into mutually exclusive groups conceived as dietary patterns and differently from cluster analysis, which has the same aim of grouping subjects, it permits quantification of the uncertainty of class membership, and assessment of goodness of fit. Moreover, it allows for adjustment for covariates directly in the pattern identification. OBJECTIVES As latent class analysis has rarely been applied in dietary pattern studies, the aim of this research is to apply the recent developments of the techniques to this area of research. We targeted to address the issue of dietary pattern identification in the case-control setting using latent class analysis and latent class trees. We provided estimation of pattern sizes and their characterization, taking into account correlations between dietary variables (local dependencies), and covariate adjustment. We also evaluated the robustness of the identified dietary patterns to total non-alcoholic energy intake adjustment, for different types of correction. Finally, we illustrated the method’s properties in the assessment of the relation between the identified dietary patterns and selected health outcomes, given the all the above. DIETARY PATTERNS AND THE RISK OF ORAL AND PHARYNGEAL CANCER We analyzed data from a multicentric case-control study on oral and pharyngeal cancer (OPC) carried out between 1992 and 2009, including 946 cases and 2492 hospital controls. Information on diet was collected through a food frequency questionnaire (FFQ). Using LCA, we identified 4 dietary patterns, conceived as mutually exclusive groups of people who shared a common dietary behaviour within groups. The first pattern, labelled ‘Prudent pattern’, showed higher probability of consuming more leafy and fruiting vegetables, citrus fruit and all other kinds of fruits, tea while showing lower probability of consuming red meat. The second pattern, that we named ‘Western pattern’, reported higher consumption of red meat and lower consumption of fruits, cruciferous and fruiting vegetables. We termed the third pattern ‘Lower consumers-combination pattern’ as people in it were less likely to eat fruits, leafy and fruiting vegetables, pulses, potatoes, fish, white and red meat, bread and tea/decaffeinated coffee. The last pattern had higher probability to eating fruiting, leafy and other vegetables, white and red meat and bread, while showed a lower probability to consume coffee, tea, processed meat, cheese, fish, sugary drinks and desserts. We called this last pattern ‘Higher consumers-combination pattern’. Dietary patterns were adjusted for total non-alcoholic energy intake and correlation between certain foods item (sugar-coffee, soups-pulses) was allowed during classes identification. Compared to the Prudent pattern, the Western and the Lower consumers-combination ones were positively related to the risk of OPC (OR=2.56, 95% CI: 1.90 – 3.45 and OR=2.23, 95% CI: 1.64 – 3.02). Higher consumers-combination pattern didn’t differ significantly from the Prudent pattern (OR=1.28, 95% CI: 0.92 – 1.77). ENERGY INTAKE ADJUSTMENT IN DIETARY PATTERN RESEARCH USING LATENT CLASS ANALYSIS Using data from the multicentric case-control study on OPC (Italy, 1992-2009), we identified and compared dietary patterns adjusting or not for total non-alcoholic energy intake in the classes identification phase of the analysis. Three possible ways to correct for total energy intake in class identification were presented, corresponding to different hypothesis on the effect of this variable. In general unadjusted and adjusted solutions were comparable. The main difference was related to the patterns that showed highest/lowest non-alcoholic energy intake, that resulted in a variation of number of classes (4/5/7 patterns for the different adjusted solutions and 5 patterns for the unadjusted one). Then, to determine the effect of adjustment in predicting an health outcome, we compared the effect of unadjusted dietary patterns, unadjusted dietary patterns with non-alcoholic energy intake variable also included in the model as a confounder, and adjusted dietary patterns on the risk of OPC . Differences in the estimations for the distinct solutions were found when Odds Ratios (ORs) were not corrected for known/potential risk factors. In general, adjustments for non-alcoholic energy intake results in a mitigation of the effects, thus remaining in the same order. When adjusting for known/potential risk factors, estimations of ORs and related confidence intervals (CIs) remained consistent in all the models we fitted. In the end, specific suggestions on how to perform energy correction in dietary patterns research using LCA were delivered, basing on the results of the current analysis. DIETARY PATTERNS INSPECTION THROUGH LATENT CLASS TREE We analyzed data from two Italian case–control studies, the first included 946 cases with OPC and 2492 hospital controls, and the second included 304 cases with squamous cell carcinoma of the esophagus (ESCC) and 743 hospital controls. In our application of latent class analysis on the combined dataset of the two case-control studies (Italy, 1992-2009), we found the best fit for a solution that was difficult to interpret and included minor differences between clusters. To address these issues, the Latent Class Tree method was proposed. Three fit statistics (AIC, AIC3, BIC) were used for their different level of penalty that resulted in different lengths of the tree and consequently, different granularity in the analysis. For the first split we allowed for a 4-class solution which identified a pattern characterized by high intake of leafy and fruiting vegetable and fruits (‘Prudent pattern’), a pattern with a high intake of red meat and low intake of certain fruits and vegetables (‘Western pattern’) and two patterns which showed a combination-type of diet. The first ‘combination’ pattern showed a low intake of the majority of foods (‘Lower consumers-combination pattern’), and the other one high intake of various foods (‘Higher consumers-combination pattern’). Compared to the Prudent pattern, the Western one was positively related to OPC (OR=1.91, 95% CI: 1.41-2.58) and to ESCC (OR=3.22, 95% CI: 1.78 – 5.82). The Lower consumers-combination pattern was positively associated to OPC (OR=2.14, 95% CI: 1.58-2.91) and to ESCC (OR=2.85, 95% CI: 1.47-5.55). No significant association was found between the Higher consumers-combination pattern and OPC (1.04, 95% CI: 0.74-1.46) and ESCC (OR=0.89, 95% CI: 0.39-1.99). In the ‘Prudent pattern’ branch of the tree, at the third level, we found two classes that differed in the risk of both cancer types. These two classes differed mainly for the intake of citrus fruit, showing respectively, OR=1.85, 95% CI:1.07-3.19 for OPC and OR=5.37, 95% CI: 1.48-19.44 for ESCC for the class that reported low intake of citrus fruit with respect to the class which exhibit a high intake of citrus fruit. No other significant differences were found between the other pairs of classes at any other level of the tree. CONCLUSION We presented latent class methods as powerful tools to determine dietary patterns conceived as mutually exclusive homogeneous groups of subjects which shared common dietary habits. These methods exhibit some advantages, with respect to classical approaches, that can address important issues in dietary pattern research. For example, it is possible to obtain estimation for pattern prevalence in the population, and to perform energy intake adjustment in the pattern identification phase of the analysis. Moreover, class formation inspection, comparison between different solutions and the analysis of subgroups that may be relevant for the research at hand are features offered by the newly developed latent class tree approach.

This paper presents the fully Bayesian analysis of the latent class model using a new approach towards MCMC estimation in the context of mixture models. The approach starts with estimating unidentified models for various numbers of classes. Exact Bayes' factors are computed by the bridge sampling estimator to compare different models and select the number of classes. Estimation of the unidentified model is carried out using the random permutation sampler. From the unidentified model estimates for model parameters that are not class specific are derived. Then, the exploration of the MCMC output from the unconstrained model yields suitable identifiability constraints. Finally, the constrained version of the permutation sampler is used to estimate group specific parameters. Conjoint data from the Austrian mineral water market serve to illustrate the method. (author's abstract)
Series: Report Series SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

In this paper we head for a fully Bayesian analysis of the latent class model with a priori unknown number of classes. Estimation is carried out by means of Markov Chain Monte Carlo (MCMC) methods. We deal explicitely with the consequences the unidentifiability of this type of model has on MCMC estimation. Joint Bayesian estimation of all latent variables, model parameters, and parameters determining the probability law of the latent process is carried out by a new MCMC method called permutation sampling. In a first run we use the random permutation sampler to sample from the unconstrained posterior. We will demonstrate that a lot of important information, such as e.g. estimates of the subject-specific regression coefficients, is available from such an unidentified model. The MCMC output of the random permutation sampler is explored in order to find suitable identifiability constraints. In a second run we use the permutation sampler to sample from the constrained posterior by imposing identifiablity constraints. The unknown number of classes is determined by formal Bayesian model comparison through exact model likelihoods. We apply a new method of computing model likelihoods for latent class models which is based on the method of bridge sampling. The approach is applied to simulated data and to data from a metric conjoint analysis in the Austrian mineral water market. (author's abstract)
Series: Forschungsberichte / Institut für Statistik

In this paper we head for a fully Bayesian analysis of the latent class model with a priori unknown number of classes. Estimation is carried out by means of Markov Chain Monte Carlo (MCMC) methods. We deal explicitely with the consequences the unidentifiability of this type of model has on MCMC estimation. Joint Bayesian estimation of all latent variables, model parameters, and parameters determining the probability law of the latent process is carried out by a new MCMC method called permutation sampling. In a first run we use the random permutation sampler to sample from the unconstrained posterior. We will demonstrate that a lot of important information, such as e.g. estimates of the subject-specific regression coefficients, is available from such an unidentified model. The MCMC output of the random permutation sampler is explored in order to find suitable identifiability constraints. In a second run we use the permutation sampler to sample from the constrained posterior by imposing identifiablity constraints. The unknown number of classes is determined by formal Bayesian model comparison through exact model likelihoods. We apply a new method of computing model likelihoods for latent class models which is based on the method of bridge sampling. The approach is applied to simulated data and to data from a metric conjoint analysis in the Austrian mineral water market. (author's abstract)
Series: Working Papers SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

This thesis examines the interactions between road users and the factors that contribute to the occurrence of traffic accidents, and discusses the implications of these interactions with regards to driver behaviour and accident prevention measures. Traffic accident data is collected on a macroscopic level by local police authorities throughout the UK. This data provides a description of accident related factors on a macroscopic level which does not allow for a complete understanding of the interaction between the various road users or the influence of errors made by active road users. Traffic accident data collected on a microscopic level analysis of real world accident data, explaining why and how an accident occurred, can further contribute to a data driven approach to provide safety measures. This data allows for a better understanding of the interaction of factors for all road users within an accident that is not possible with other data collection methods. In the first part of the thesis, a literature review presents relevant research in traffic accident analysis and accident causation research, afterwards three accident causation models used to understand behaviour and factors leading to traffic accidents are introduced. A comparison study of these accident causation coding models that classify road user error was carried out to determine a model that would be best suited to code the accident data according to the thesis aims. Latent class cluster analyses were made of two separate datasets, the UK On the Spot (OTS) in-depth accident investigation study and the STATS19 national accident database. A comparison between microscopic (in-depth) accident data and macroscopic (national) accident data was carried out. This analysis allowed for the interactions between all relevant factors for the road users involved in the accident to be grouped into specific accident segmentations based on the cluster analysis results. First, all of the cases that were collected by the OTS team between the years 2000 to 2003 were analysed. Results suggested that for single vehicle accidents males and females typically made failures related to detection and execution issues, whereas male road users made diagnosis failures with speed as a particularly important factor. In terms of the multiple vehicle accidents the interactions between the first two road users and the subsequent accident sequence were demonstrated. A cluster analysis of all two vehicle accidents in Great Britain in the year 2005 and recorded within the STATS19 accident database was carried out as a comparison to the multiple vehicle accident OTS data. This analysis demonstrated the necessity of in-depth accident causation data in interpreting accident scenarios, as the resulting accident clusters did not provide significant differences between the groups to usefully segment the crash population. Relevant human factors were not coded for these cases and the level of detail in the accident cases did not allow for a discussion of countermeasure implications. An analysis of 428 Powered Two Wheeler accidents that were collected by the OTS team between the years 2000 to 2010 was carried out. Results identified 7 specific scenarios, the main types of which identified two particular looked but did not see accidents and two types of single vehicle PTW accidents. In cases where the PTW lost control, diagnosis failures were more common, for road users other than the PTW rider, detection issues were of particular relevance. In these cases the interaction between all relevant road users was interpreted in relation to one another. The subsequent study analysed 248 Pedestrian accidents that were collected by the OTS team between the years 2000 to 2010. Results identified scenarios related to pedestrians as being in a hurry and making detection errors, impairment due to alcohol, and young children playing in the roadside. For accidents that were initiated by the other road user s behaviour pedestrians were either struck after an accident had already occurred or due to the manoeuvre that a road user was making, older pedestrians were over-represented in this accident type. This thesis concludes by discussing how (1) microscopic in-depth accident data is needed to understand accident mechanisms, (2) a data mining approach using latent class clustering can benefit the understanding of failure mechanisms, (3) accident causation analysis is necessary to understand the types of failures that road users make and (4) accident scenario development helps quantify accidents and allows for accident countermeasure implication discussion. The original contribution to knowledge is the demonstration that when relevant data is available there is a possibility to understand the interactions that are occurring between road users before the crash, that is not possible otherwise. This contribution has been demonstrated by highlighting how latent class cluster analysis combined with accident causation data allows for relevant interactions between road users to be observed. Finally implications for this work and future considerations are outlined.

Direct marketing problems have been intensively reviewed in the marketing literature recently, such as purchase frequency and time, sales profit, and brand choices. However, modeling the customer response, which is an important issue in direct marketing research, remains a significant challenge. This thesis is an empirical study of predicting customer response to direct marketing and applies a Bayesian approach, including the Bayesian Binary Regression (BBR) and the Hierarchical Bayes (HB). Other classical methods, such as Logistic Regression and Latent Class Analysis (LCA), have been conducted for the purpose of comparison. The results of comparing the performance of all these techniques suggest that the Bayesian methods are more appropriate in predicting direct marketing customer responses. Specifically, when customers are analyzed as a whole group, the Bayesian Binary Regression (BBR) has greater predictive accuracy than Logistic Regression. When we consider customer heterogeneity, the Hierarchical Bayes (HB) models, which use demographic and geographic variables for clustering, do not match the performance of Latent Class Analysis (LCA). Further analyses indicate that when latent variables are used for clustering, the Hierarchical Bayes (HB) approach has the highest predictive accuracy.

Thesis (M.S.)--Worcester Polytechnic Institute.
Keywords: latent class model; Poisson regression model; Metropolis-Hastings sampler; order restriction; disease mapping. Includes bibliographical references.

L'accident vasculaire cérébral et la sclérose en plaques figurent parmi les maladies neurologiques les plus destructrices du système nerveux central. L'accident vasculaire cérébral est la deuxième cause de décès et la principale cause de handicap chez l'adulte dans le monde alors que la sclérose en plaques est la maladie neurologique non traumatique la plus fréquente chez l'adulte jeune. L'imagerie par résonance magnétique est un outil important pour distinguer le tissu cérébral sain du tissu pathologique à des fins de diagnostic, de suivi de la maladie, et de prise de décision pour un traitement personnalisé des patients atteints d'accident vasculaire cérébral ou de sclérose en plaques. La prédiction de l'évolution individuelle de la maladie chez les patients atteints d'accident vasculaire cérébral ou de sclérose en plaques constitue un défi pour les cliniciens avant de donner un traitement individuel approprié. Cette prédiction est possible avec des approches statistiques appropriées basées sur des informations cliniques et d'imagerie. Toutefois, l'étiologie, la physiopathologie, les symptômes et l'évolution dans l'accident vasculaire cérébral et la sclérose en plaques sont très différents. Par conséquent, dans cette thèse, les méthodes statistiques utilisées pour ces deux maladies neurologiques sont différentes. Le premier objectif était l'identification du tissu à risque d'infarctus chez les patients atteints d'accident vasculaire cérébral. Pour cet objectif, les méthodes de classification (dont les méthodes de machine learning) ont été utilisées sur des données d'imagerie mesurées à l'admission pour prédire le risque d'infarctus à un mois. Les performances des méthodes de classification ont été ensuite comparées dans un contexte d'identification de tissu à haut risque d'infarctus à partir de données humaines codées voxel par voxel. Le deuxième objectif était de regrouper les patients atteints de sclérose en plaques avec une méthode non supervisée basée sur des trajectoires individuelles cliniques et d'imagerie tracées sur cinq ans. Les groupes de trajectoires aideraient à identifier les patients menacés d'importantes progressions et donc à leur donner des médicaments plus efficaces. Le troisième et dernier objectif de la thèse était de développer un modèle prédictif pour l'évolution du handicap individuel des patients atteints de sclérose en plaques sur la base de données démographiques, cliniques et d'imagerie obtenues a l'inclusion. L'hétérogénéité des évolutions du handicap chez les patients atteints de sclérose en plaques est un important défi pour les cliniciens qui cherchent à prévoir l'évolution individuelle du handicap. Le modèle mixte linéaire à classes latentes a été utilisé donc pour prendre en compte la variabilité individuelle et la variabilité inobservée entre sous-groupes de sclérose en plaques
Stroke and multiple sclerosis are two of the most destructive neurological diseases of the central nervous system. Stroke is the second most common cause of death and the major cause of disability worldwide whereas multiple sclerosis is the most common non-traumatic disabling neurological disease of adulthood. Magnetic resonance imaging is an important tool to distinguish healthy from pathological brain tissue in diagnosis, monitoring disease evolution, and decision-making in personalized treatment of patients with stroke or multiple sclerosis.Predicting disease evolution in patients with stroke or multiple sclerosis is a challenge for clinicians that are about to decide on an appropriate individual treatment. The etiology, pathophysiology, symptoms, and evolution of stroke and multiple sclerosis are highly different. Therefore, in this thesis, the statistical methods used for the study of the two neurological diseases are different.The first aim was the identification of the tissue at risk of infarction in patients with stroke. For this purpose, the classification methods (including machine learning methods) have been used on voxel-based imaging data. The data measured at hospital admission is performed to predict the infarction risk at one month. Next, the performances of the classification methods in identifying the tissue at a high risk of infarction were compared. The second aim was to cluster patients with multiple sclerosis using an unsupervised method based on individual clinical and imaging trajectories plotted over five 5 years. Clusters of trajectories would help identifying patients who may have an important progression; thus, to treat them with more effective drugs irrespective of the clinical subtypes. The third and final aim of this thesis was to develop a predictive model for individual evolution of patients with multiple sclerosis based on demographic, clinical, and imaging data taken at study onset. The heterogeneity of disease evolution in patients with multiple sclerosis is an important challenge for the clinicians who seek to predict the disease evolution and decide on an appropriate individual treatment. For this purpose, the latent class linear mixed model was used to predict disease evolution considering individual and unobserved subgroup' variability in multiple sclerosis

碩士
國立臺北大學
統計學系
96
Survey is a popular research tool, but often causes missing values for some reasons. When the proportion of the missing value is high, it can seriously affect the conclusion. Imputation is an alternative is to handle missing data. For categorical missing data, both model-based and non- model based imputation methods have been proposed, for example, hot deck imputation and loglinear models. However, there are still some problems for these methods. Latent class model (LCM) is a popularly used method for categorical variable. We extended the research of Vermunt al (2007) to study what are the important factors on accuracy rate of imputation for categorical data. Four imputation methods and 6 other independent variables were examined for their effects on accuracy of imputation. The imputation methods were evaluated in terms of accuracy rates. The result shows the significant factors are conditional probability, latent class proportions, number of manifest variables, imputation method, sample size, missing data mechanism. The accuracy rate of imputation is higher with substantially different conditional probability and latent class proportions, more manifest variables, method2 or method3, larger sample sizes, MCAR, and lower missing rate.

Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2006.
Source: Dissertation Abstracts International, Volume: 67-11, Section: B, page: 6485. Adviser: Jeffrey Douglas. Includes bibliographical references (leaf 58) Available on microfilm from Pro Quest Information and Learning.

碩士
輔仁大學
心理學系
98
Clustering is an important multivariate technique in social science which assigns individuals into group under particular rules. Due to the difference on the use of variable, there are different approaches in clustering methods. The purpose of this study is to apply three clustering methods: Two-Step Cluster (TSC), Latent Class Modeling (LCM), and Self-Organizing Map (SOM) work on the separation of mixture distribution. This study includes two sub-researches, simulation and empirical study. In the first study, the 2000 Monte Carlo simulation data is created by R software, including three variables which are normal distributed. In the second study, survey data regarding occupation interest of 617 undergraduates students were selected from Taiwan Integrated Postsecondary Education Database (TIPED). The occupation interest consists of six dimensions: interests in mathematics, machinery, creation, social, leading, and administration. The TSC was processed by using SPSS 17, the LCM was processed by using LatentGOLD 4.5, and the SOM was processed by Clementine 12. At first, the characteristics of the sample were examined and then followed by the TSC, LCM, and SOM analysis. Finally, the association of true score and discrimination among the outcome cluster of methods were applied. The results of simulation study showed that the same outcome of clustering by TSC and LCM was reported. However, SOM tend to cluster data in more detail. Moreover, the outcomes of all methods have high association and good discrimination. In the other hand, the results of empirical study showed difference in outcomes, but the structure of occupation interest was confirmed. In association test, middle association between TSC outcome and department, high association between LCM outcome and department, and low association between SOM outcome and department. In discrimination test, good discrimination exists in TSC and LCM outcome, but bad discrimination exists in SOM outcome. In the end, the difference in algorithm which might affect the performance of methods and their value in application were discussed. The research shown the importance and significance of clustering methods.