본문

서브메뉴

Bayesian Identification, Estimation, and Evaluation of Growth Mixture Models
Bayesian Identification, Estimation, and Evaluation of Growth Mixture Models
Bayesian Identification, Estimation, and Evaluation of Growth Mixture Models

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260202103503
ISBN  
9798288862229
DDC  
151
저자명  
Xiao, Xingyao Doria.
서명/저자  
Bayesian Identification, Estimation, and Evaluation of Growth Mixture Models
발행사항  
[Sl] : University of California, Berkeley, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
138 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-01, Section: A.
주기사항  
Advisor: Rabe-Hesketh, Sophia.
학위논문주기  
Thesis (Ph.D.)--University of California, Berkeley, 2025.
초록/해제  
요약Growth Mixture Models (GMMs) are widely used in education and psychology to model individual differences in developmental trajectories through discrete latent classes and continuous random effects. Despite their capability to represent complex longitudinal structures, effectively identifying, estimating, and evaluating GMMs poses significant methodological challenges. This dissertation addresses these challenges through theoretical analysis, empirical demonstrations, and simulation studies.The first chapter investigates problematic behaviors in Markov chain Monte Carlo (MCMC) estimation caused by degenerate nonidentifiability in finite mixture models. It identifies underlying causes, introduces diagnostics to detect such issues, and demonstrates through simulations that using more informative priors significantly reduces these problems. An application of GMMs to data from the National Longitudinal Survey of Youth (NLSY), analyzing reading skill development from ages 6 to 14, illustrates these points. Additionally, this chapter proposes methods to characterize within-class variability, reviews the literature on likelihood and Bayesian identification, introduces a practical definition of Bayesian identification using marginal likelihoods (integrated over the latent variables), and provides a concise overview of Hamiltonian Monte Carlo (HMC) in Stan.The second chapter reviews information criteria commonly used for model selection, including frequentist approaches, such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), and Bayesian approaches, such as the Deviance Information Criterion (DIC), Watanabe-Akaike Information Criterion (WAIC), and Leave-One-Out Cross-Validation (LOO-CV). For the DIC, the penalty term is often found to be negative, especially in finite mixture models. A new version of the DIC is proposed that uses a variance-based penalty term (as previously suggested) and also does not rely on the plug-in deviance. For each information criterion, the target quantity is defined to clarify the motivation for the criterion. For hierarchical or latent variable models, information criteria can be defined based on the marginal likelihood, integrated over the latent variables, or the conditional likelihood, conditioning on the latent variables. The chapter underscores the theoretical advantages of using the marginal likelihood. The chapter ends with a brief review of information criteria for GMMs.The third chapter presents a comprehensive simulation study assessing the performance of frequentist and Bayesian estimation and of model selection based on a range of information criteria. The data-generating model is a two-class GMM, and the model selection problem is to identify the correct number of latent classes, choosing among 1-class, 2-class, 3-class, and 4-class models. Simulation conditions include equal versus unequal class probabilities, two sample sizes, smaller or larger class separation, and smaller or larger residual variances. Both frequentist and Bayesian estimation performed poorly in some conditions with three or more classes, the former converging to local maxima and the latter suffering from degenerate nonidentifiability. Interestingly, the Bayesian issue does not pose problems for information criteria that do not rely on a plug-in deviance, such as our proposed variance-based formulation of the DIC and Congdon's Bayesian criteria (CAIC and CBIC). This observation leads to another possible diagnostic for degenerate nonidentifiability, namely a negative DIC penalty term. In contrast, convergence to suboptimal local maxima is problematic for all frequentist information criteria and needs to be addressed. Unfortunately, increasing the number of initializations in flexmix does not always solve the problem. , whereas traditional frequentist estimators frequently converge to local maxima under weak identification.Collectively, these chapters offer theoretical foundations, practical diagnostics, and empirical guidance to advance rigorous and reproducible GMM research.
일반주제명  
Quantitative psychology
일반주제명  
Statistics
일반주제명  
Applied mathematics
일반주제명  
Information science
키워드  
Bayesian identification
키워드  
Growth Mixture Models
키워드  
Model evaluation
키워드  
Markov chain Monte Carlo estimation
키워드  
Hamiltonian Monte Carlo
기타저자  
University of California, Berkeley Education
기본자료저록  
Dissertations Abstracts International. 87-01A.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017357378
■00520260202103503
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798288862229
■035    ▼a(MiAaPQ)AAI32001869
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a151
■1001  ▼aXiao,  Xingyao  Doria.
■24510▼aBayesian  Identification,  Estimation,  and  Evaluation  of  Growth  Mixture  Models
■260    ▼a[Sl]▼bUniversity  of  California,  Berkeley▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a138  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-01,  Section:  A.
■500    ▼aAdvisor:  Rabe-Hesketh,  Sophia.
■5021  ▼aThesis  (Ph.D.)--University  of  California,  Berkeley,  2025.
■520    ▼aGrowth  Mixture  Models  (GMMs)  are  widely  used  in  education  and  psychology  to  model  individual  differences  in  developmental  trajectories  through  discrete  latent  classes  and  continuous  random  effects.  Despite  their  capability  to  represent  complex  longitudinal  structures,  effectively  identifying,  estimating,  and  evaluating  GMMs  poses  significant  methodological  challenges.  This  dissertation  addresses  these  challenges  through  theoretical  analysis,  empirical  demonstrations,  and  simulation  studies.The  first  chapter  investigates  problematic  behaviors  in  Markov  chain  Monte  Carlo  (MCMC)  estimation  caused  by  degenerate  nonidentifiability  in  finite  mixture  models.  It  identifies  underlying  causes,  introduces  diagnostics  to  detect  such  issues,  and  demonstrates  through  simulations  that  using  more  informative  priors  significantly  reduces  these  problems.  An  application  of  GMMs  to  data  from  the  National  Longitudinal  Survey  of  Youth  (NLSY),  analyzing  reading  skill  development  from  ages  6  to  14,  illustrates  these  points.  Additionally,  this  chapter  proposes  methods  to  characterize  within-class  variability,  reviews  the  literature  on  likelihood  and  Bayesian  identification,  introduces  a  practical  definition  of  Bayesian  identification  using  marginal  likelihoods  (integrated  over  the  latent  variables),  and  provides  a  concise  overview  of  Hamiltonian  Monte  Carlo  (HMC)  in  Stan.The  second  chapter  reviews  information  criteria  commonly  used  for  model  selection,  including  frequentist  approaches,  such  as  Akaike  Information  Criterion  (AIC)  and  Bayesian  Information  Criterion  (BIC),  and  Bayesian  approaches,  such  as  the  Deviance  Information  Criterion  (DIC),  Watanabe-Akaike  Information  Criterion  (WAIC),  and  Leave-One-Out  Cross-Validation  (LOO-CV).  For  the  DIC,  the  penalty  term  is  often  found  to  be  negative,  especially  in  finite  mixture  models.  A  new  version  of  the  DIC  is  proposed  that  uses  a  variance-based  penalty  term  (as  previously  suggested)  and  also  does  not  rely  on  the  plug-in  deviance.  For  each  information  criterion,  the  target  quantity  is  defined  to  clarify  the  motivation  for  the  criterion.  For  hierarchical  or  latent  variable  models,  information  criteria  can  be  defined  based  on  the  marginal  likelihood,  integrated  over  the  latent  variables,  or  the  conditional  likelihood,  conditioning  on  the  latent  variables.  The  chapter  underscores  the  theoretical  advantages  of  using  the  marginal  likelihood.  The  chapter  ends  with  a  brief  review  of  information  criteria  for  GMMs.The  third  chapter  presents  a  comprehensive  simulation  study  assessing  the  performance  of  frequentist  and  Bayesian  estimation  and  of  model  selection  based  on  a  range  of  information  criteria.  The  data-generating  model  is  a  two-class  GMM,  and  the  model  selection  problem  is  to  identify  the  correct  number  of  latent  classes,  choosing  among  1-class,  2-class,  3-class,  and  4-class  models.  Simulation  conditions  include  equal  versus  unequal  class  probabilities,  two  sample  sizes,  smaller  or  larger  class  separation,  and  smaller  or  larger  residual  variances.  Both  frequentist  and  Bayesian  estimation  performed  poorly  in  some  conditions  with  three  or  more  classes,  the  former  converging  to  local  maxima  and  the  latter  suffering  from  degenerate  nonidentifiability.  Interestingly,  the  Bayesian  issue  does  not  pose  problems  for  information  criteria  that  do  not  rely  on  a  plug-in  deviance,  such  as  our  proposed  variance-based  formulation  of  the  DIC  and  Congdon's  Bayesian  criteria  (CAIC  and  CBIC).  This  observation  leads  to  another  possible  diagnostic  for  degenerate  nonidentifiability,  namely  a  negative  DIC  penalty  term.  In  contrast,  convergence  to  suboptimal  local  maxima  is  problematic  for  all  frequentist  information  criteria  and  needs  to  be  addressed.  Unfortunately,  increasing  the  number  of  initializations  in  flexmix  does  not  always  solve  the  problem.  ,  whereas  traditional  frequentist  estimators  frequently  converge  to  local  maxima  under  weak  identification.Collectively,  these  chapters  offer  theoretical  foundations,  practical  diagnostics,  and  empirical  guidance  to  advance  rigorous  and  reproducible  GMM  research.
■590    ▼aSchool  code:  0028.
■650  4▼aQuantitative  psychology
■650  4▼aStatistics
■650  4▼aApplied  mathematics
■650  4▼aInformation  science
■653    ▼aBayesian  identification
■653    ▼aGrowth  Mixture  Models
■653    ▼aModel  evaluation
■653    ▼aMarkov  chain  Monte  Carlo  estimation
■653    ▼aHamiltonian  Monte  Carlo
■690    ▼a0632
■690    ▼a0723
■690    ▼a0364
■690    ▼a0463
■71020▼aUniversity  of  California,  Berkeley▼bEducation.
■7730  ▼tDissertations  Abstracts  International▼g87-01A.
■790    ▼a0028
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357378▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    Подробнее информация.

    • Бронирование
    • не существует
    • моя папка
    • Первый запрос зрения
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    материал
    Reg No. Количество платежных Местоположение статус Ленд информации
    TF17375 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Бронирование доступны в заимствований книги. Чтобы сделать предварительный заказ, пожалуйста, нажмите кнопку бронирование

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.