FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2800458972> ?p ?o ?g. }

• W2800458972 abstract "Author(s): Furr, Daniel Coulter | Advisor(s): Rabe-Hesketh, Sophia | Abstract: The chapters of this dissertation are intended to be three independent, publishable papers, but they nevertheless share the theme of predictive inferences for explanatory item models. Chapter 1 describes the differences between the Bayesian and frequentist statistical frameworks in the context of explanatory item response models. The particular model of focus, the doubly explanatory model, is a model for dichotomous item responses that includes covariates for person ability and covariates for item difficulty. It includes many Rasch-family models as special cases. Differences in how the model is understood and specified within the two frameworks are discussed. The various predictive inferences available from the model are defined for the two frameworks. Chapter 2 is situated in the frequentist framework and focuses on approaches for explaining or predicting the difficulties of items. Within the frequentist framework, the linear logistic test model (LLTM) is likely to be used for this purpose, which in essence regresses item difficulty on covariates for characteristics of the items. However, this regression does not include an error term, and so the model is in general misspecified. Meanwhile, adding an error term to the LLTM makes maximum likelihood estimation infeasible. To address this problem, a two-stage modeling strategy (LLTM-E2S) is proposed: in the first stage Rasch model maximum likelihood estimates for item difficulties and standard errors are obtained, and in the second stage a random effects meta-analysis regression of the Rasch difficulties on covariates is performed that incorporates the uncertainty in the item difficulty estimates. In addition, holdout validation, cross-validation, and Akaike information criteria (AIC) are discussed as means of comparing models that have different sets of item predictors. I argue that AIC used with the LLTM estimates the expected deviance of the fitted model when applied to new observations from the same sample of items and persons, which is unsuitable for assessing the ability of the model to predict item difficulties. On the other hand, AIC applied to the LLTM-E2S provides the expected deviance related to new observations arising from new items, which is what is needed. A simulation study compares parameter recovery and model comparison results for the two modeling strategies.Chapter 3 takes a Bayesian outlook and focuses on models that explain or predict person abilities. I argue that the usual application of Bayesian forms of information criteria to these models yields the wrong inference. Specifically, when using likelihoods that are conditional on person ability, information criteria estimate the expected fit of the model to new data arising from the same persons. What are needed are likelihoods that are marginal over the distribution for ability, which may be used with information criteria to estimate the expected fit to new data from a new sample of persons. The widely applicable information criterion (WAIC), Pareto-smoothed importance sampling approximation to leave-one-out cross-validation, and deviance information criterion (DIC) are discussed in the context of these conditional and marginal likelihoods. An adaptive quadrature scheme for use within Markov chain Monte Carlo estimation is proposed to obtain the marginal likelihoods. Also, the moving block bootstrap is investigated as a means to estimate the Monte Carlo error for Bayesian information criteria estimates. A simulation study using a linear random intercept model is conducted to assess the accuracy of the adaptive quadrature scheme and the bootstrap estimates of Monte Carlo error. These methods are then applied to an real item response dataset, demonstrating the practical difference between conditional and marginal forms of information criteria." @default.
• W2800458972 created "2018-05-17" @default.
• W2800458972 creator A5055907112 @default.
• W2800458972 date "2017-01-01" @default.
• W2800458972 modified "2023-09-26" @default.
• W2800458972 title "Bayesian and frequentist cross-validation methods for explanatory item response models" @default.
• W2800458972 cites W1480376833 @default.
• W2800458972 cites W1513861746 @default.
• W2800458972 cites W1598813349 @default.
• W2800458972 cites W1648888400 @default.
• W2800458972 cites W1954062058 @default.
• W2800458972 cites W1967495697 @default.
• W2800458972 cites W1967931779 @default.
• W2800458972 cites W1968371014 @default.
• W2800458972 cites W1972625678 @default.
• W2800458972 cites W1985054565 @default.
• W2800458972 cites W1988301765 @default.
• W2800458972 cites W2010796907 @default.
• W2800458972 cites W2015558203 @default.
• W2800458972 cites W2017541934 @default.
• W2800458972 cites W2017779924 @default.
• W2800458972 cites W2017966270 @default.
• W2800458972 cites W2023517070 @default.
• W2800458972 cites W2024159897 @default.
• W2800458972 cites W2030553221 @default.
• W2800458972 cites W2039332020 @default.
• W2800458972 cites W2043096067 @default.
• W2800458972 cites W2043241258 @default.
• W2800458972 cites W2043466393 @default.
• W2800458972 cites W2046376969 @default.
• W2800458972 cites W2047694274 @default.
• W2800458972 cites W2052342715 @default.
• W2800458972 cites W2053752283 @default.
• W2800458972 cites W2057765075 @default.
• W2800458972 cites W2061925911 @default.
• W2800458972 cites W206627562 @default.
• W2800458972 cites W2077514460 @default.
• W2800458972 cites W2078351332 @default.
• W2800458972 cites W2086206379 @default.
• W2800458972 cites W2089603425 @default.
• W2800458972 cites W2089965822 @default.
• W2800458972 cites W2101895213 @default.
• W2800458972 cites W2109844396 @default.
• W2800458972 cites W2112995480 @default.
• W2800458972 cites W2113338391 @default.
• W2800458972 cites W2122977572 @default.
• W2800458972 cites W2123222757 @default.
• W2800458972 cites W2127374659 @default.
• W2800458972 cites W2127682847 @default.
• W2800458972 cites W2129092711 @default.
• W2800458972 cites W2130902307 @default.
• W2800458972 cites W2137387582 @default.
• W2800458972 cites W2142635246 @default.
• W2800458972 cites W2148534890 @default.
• W2800458972 cites W2153882688 @default.
• W2800458972 cites W2168175751 @default.
• W2800458972 cites W2168934042 @default.
• W2800458972 cites W2245463310 @default.
• W2800458972 cites W2800420890 @default.
• W2800458972 cites W281417399 @default.
• W2800458972 cites W55848811 @default.
• W2800458972 cites W958651444 @default.
• W2800458972 hasPublicationYear "2017" @default.
• W2800458972 type Work @default.
• W2800458972 sameAs 2800458972 @default.
• W2800458972 citedByCount "0" @default.
• W2800458972 crossrefType "journal-article" @default.
• W2800458972 hasAuthorship W2800458972A5055907112 @default.
• W2800458972 hasConcept C101266164 @default.
• W2800458972 hasConcept C105795698 @default.
• W2800458972 hasConcept C107673813 @default.
• W2800458972 hasConcept C119043178 @default.
• W2800458972 hasConcept C121332964 @default.
• W2800458972 hasConcept C149782125 @default.
• W2800458972 hasConcept C151730666 @default.
• W2800458972 hasConcept C151956035 @default.
• W2800458972 hasConcept C160234255 @default.
• W2800458972 hasConcept C162376815 @default.
• W2800458972 hasConcept C171606756 @default.
• W2800458972 hasConcept C19875794 @default.
• W2800458972 hasConcept C2779343474 @default.
• W2800458972 hasConcept C33923547 @default.
• W2800458972 hasConcept C41008148 @default.
• W2800458972 hasConcept C61797465 @default.
• W2800458972 hasConcept C62520636 @default.
• W2800458972 hasConcept C86803240 @default.
• W2800458972 hasConceptScore W2800458972C101266164 @default.
• W2800458972 hasConceptScore W2800458972C105795698 @default.
• W2800458972 hasConceptScore W2800458972C107673813 @default.
• W2800458972 hasConceptScore W2800458972C119043178 @default.
• W2800458972 hasConceptScore W2800458972C121332964 @default.
• W2800458972 hasConceptScore W2800458972C149782125 @default.
• W2800458972 hasConceptScore W2800458972C151730666 @default.
• W2800458972 hasConceptScore W2800458972C151956035 @default.
• W2800458972 hasConceptScore W2800458972C160234255 @default.
• W2800458972 hasConceptScore W2800458972C162376815 @default.
• W2800458972 hasConceptScore W2800458972C171606756 @default.
• W2800458972 hasConceptScore W2800458972C19875794 @default.
• W2800458972 hasConceptScore W2800458972C2779343474 @default.
• W2800458972 hasConceptScore W2800458972C33923547 @default.

• next