Browse ORBi by ORBi project

- Background
- Content
- Benefits and challenges
- Legal aspects
- Functions and services
- Team
- Help and tutorials

Recent advances and improvements in computerized adaptive testing with the R package catR Magis, David ; Conference (2014, February 14) The purpose of this talk is to present recent advances and developments of the R package catR. This package allows for random generation of response patterns under computerized adaptive testing (CAT) and ... [more ▼] The purpose of this talk is to present recent advances and developments of the R package catR. This package allows for random generation of response patterns under computerized adaptive testing (CAT) and holds various options to select the first items, estimate ability, select the next item, stop the test and return final results. Two main improvements were realized. First, several rules for next item selection were added, among others, Kullback- Leibler, progressive and proportional methods. Second, catR was limited to dichotomous IRT models so far. The most recent update allows now for several polytomous IRT models, such as partial and generalized partial credit model, graded and modified graded response models, rating scale model, and nominal response model. All improvements will be shortly presented, both from a theoretical point of view and in terms of practical implementation in catR. If possible, several illustrative examples will be displayed in a short live demonstration of catR. [less ▲] Detailed reference viewed: 86 (4 ULg)The effect of melody and technique on the singing voice accuracy of trained singers Larrouy, Pauline ; Magis, David ; Morsomme, Dominique in Logopedics, Phoniatrics, Vocology (2014), 39 A previous study highlighted the effect of vocal technique on the singing voice accuracy of trained singers (1). The intervals' precision between the notes of the tune was altered when the singers used ... [more ▼] A previous study highlighted the effect of vocal technique on the singing voice accuracy of trained singers (1). The intervals' precision between the notes of the tune was altered when the singers used Western operatic singing technique. In order to better understand these results, we have recorded two different melodies sung with two different vocal techniques. A large panel of trained singers (N = 50) participated in the study. The analytical method described in the reference paper (1) has been applied. The results confirm the effect of vocal technique on the vocal accuracy of trained singers. In addition, these results provide an answer about the melodic effect and guide future work on the perception process of operatic voices. [less ▲] Detailed reference viewed: 75 (27 ULg)Effects of melody and technique on acoustical and musical features of Western operatic singing voices Larrouy, Pauline ; Magis, David ; Morsomme, Dominique in Journal of Voice (2014), 28 Objective: The operatic singing technique is frequently employed in classical music. Several acoustical parameters of this specific technique have been studied but how these parameters combine remains ... [more ▼] Objective: The operatic singing technique is frequently employed in classical music. Several acoustical parameters of this specific technique have been studied but how these parameters combine remains unclear. This study aims to further characterize the Western operatic singing technique by observing the effects of melody and technique on acoustical and musical parameters of the singing voice. Methods: Fifty professional singers performed two contrasting melodies (popular song and romantic melody) with two vocal techniques (with and without operatic singing technique). The common quality parameters (energy distribution, vibrato rate and extent), perturbation parameters (standard deviation of the fundamental frequency, signal-to-noise ratio, jitter and shimmer) and musical features (fundamental frequency of the starting note, average tempo, and sound pressure level) of the 200 sung performances were analyzed. Results: The results regarding the effect of melody and technique on the acoustical and musical parameters show that the choice of melody had a limited impact on the parameters observed, whereas a particular vocal profile appeared depending on the vocal technique employed. Conclusions: This study confirms that vocal technique affects most of the parameters examined. In addition, the observation of quality, perturbation and musical parameters contributes to a better understanding of the Western operatic singing technique. [less ▲] Detailed reference viewed: 189 (12 ULg)Analyse du fonctionnement différentiel des items des versions papier et informatisée d’un test de classement en anglais, langue seconde, en présence de patrons de réponses inappropriés ; Magis, David ; in Blais, Jean-Guy; Gilles, Jean-Luc; Tristan-Lopez (Eds.) Bienvenue au 21e siècle : Evaluation des apprentissages et technologies de l’information et de la communication (2014) Detailed reference viewed: 25 (2 ULg)On the asymptotic standard error of a class of robust estimators of ability in dichotomous item response models Magis, David in British Journal of Mathematical & Statistical Psychology (2014), 67 In item response theory, the classical estimators of ability are highly sensitive to response disturbances and can return strongly biased estimates of the true underlying ability level. Robust methods ... [more ▼] In item response theory, the classical estimators of ability are highly sensitive to response disturbances and can return strongly biased estimates of the true underlying ability level. Robust methods were introduced to lessen the impact of such aberrant responses onto the estimation process. The computation of asymptotic (i.e., large sample) standard errors (ASE) for these robust estimators, however, has not been fully considered yet. This paper focuses on a broad class of robust ability estimators, defined by an appropriate selection of the weight function and the residual measure, for which the ASE is derived from the theory of estimating equations. The maximum likelihood (ML) and the robust estimators, together with their estimated ASE, are then compared through a simulation study. It is concluded that both the estimators and their ASE perform similarly in absence of response disturbances, while the robust estimator and its estimated ASE are less biased and outperform their ML counterparts in presence of response disturbances with large impact on the item response process. [less ▲] Detailed reference viewed: 26 (4 ULg)The evaluation of vocal pitch accuracy: The case of operatic singing voices Larrouy, Pauline ; Magis, David ; Morsomme, Dominique in Music Perception (2014), 32(1), 1-10 The objective analysis of Western operatic singing voices indicates that professional singers can be particularly “out of tune”. This study aims to better understand the evaluation of operatic voices ... [more ▼] The objective analysis of Western operatic singing voices indicates that professional singers can be particularly “out of tune”. This study aims to better understand the evaluation of operatic voices, which have particularly complex acoustical signals. Twenty-two music experts were asked to evaluate the vocal pitch accuracy of 14 sung performances with a pairwise comparison paradigm, in a test and a retest. In addition to the objective measurement of pitch accuracy (pitch interval deviation), several performance parameters (average tempo, fundamental frequency of the starting note) and quality parameters (energy distribution, vibrato rate and extent) were observed and compared to the judges’ perceptual rating. The results show high intra- and inter-judge reliability when rating the pitch accuracy of operatic singing voices. Surprisingly, all the parameters were significantly related to the ratings and explain 78.8% of the variability of the judges’ rating. The pitch accuracy evaluation of operatic voices is thus not based exclusively on the precision of performed music intervals but on a complex combination of performance and quality parameters. [less ▲] Detailed reference viewed: 49 (15 ULg)deltaPlotR: An R package for differential item functioning analysis with Angoff's Delta Plot Magis, David ; in Journal of Statistical Software (2014), 59(1), 1-19 Angoff's delta plot is a straightforward and not computationally intensive method to identify differential item functioning (DIF) among dichotomously scored items. This approach was recently improved by ... [more ▼] Angoff's delta plot is a straightforward and not computationally intensive method to identify differential item functioning (DIF) among dichotomously scored items. This approach was recently improved by proposing an optimal threshold selection and by considering several item purification processes. Moreover, to support practical DIF analyses with the delta plot and these improvements, the R package deltaPlotR was also developed. The purpose of this paper is twofold: to outline the delta plot by describing the original method and its recent improvements in a user-oriented way, and to illustrate the structure and performances of the deltaPlotR package. A real data set about language skill assessment is being analyzed as an illustrative example. [less ▲] Detailed reference viewed: 47 (1 ULg)Type I error inflation in DIF identification with Mantel-Haenszel: an explanation and a solution Magis, David ; in Educational & Psychological Measurement (2014), 74 It is known that sum score-based methods for the identification of differential item functioning (DIF), such as the Mantel-Haenszel (MH) approach, can be affected by Type I error inflation in the absence ... [more ▼] It is known that sum score-based methods for the identification of differential item functioning (DIF), such as the Mantel-Haenszel (MH) approach, can be affected by Type I error inflation in the absence of any DIF effect. This may happen when the items differ in discrimination and when there is item impact. On the other hand, outlier DIF methods have been developed that are robust against this Type I error inflation, while they are still based on the MH DIF statistic. The present paper gives an explanation for why the common MH method is indeed vulnerable to the inflation effect while the outlier DIF versions are not. In a simulation study we were able to produce the Type I error inflation by inducing item impact and item differences in discrimination. At the same time and in parallel with the Type I error inflation the dispersion of the DIF statistic across items was increased. As expected, the outlier DIF methods did not seem sensitive to impact and differences in item discrimination. [less ▲] Detailed reference viewed: 16 (1 ULg)Asymptotic standard errors of ability estimators in dichotomous item response models Magis, David Report (2014) Detailed reference viewed: 29 (2 ULg)Snijders’ correction of Infit and Outfit indexes with estimated ability level : an analysis with the Rasch model Magis, David ; ; in Journal of Applied Measurement (2014), 15 The Infit mean square W and the Outfit mean square U are commonly used person fit indexes under Rasch measurement. However, they suffer from two major weaknesses. First, their asymptotic distribution is ... [more ▼] The Infit mean square W and the Outfit mean square U are commonly used person fit indexes under Rasch measurement. However, they suffer from two major weaknesses. First, their asymptotic distribution is usually derived by assuming that the true ability levels are known. Second, such distributions are even not clearly stated for indexes U and W. Both issues can seriously affect the selection of an appropriate cut-score for person fit identification. Snijders (2001) proposed a general approach to correct some person fit indexes when specific ability estimators are used. The purpose of this paper is to adapt this approach to U and W indexes. First, a brief sketch of the methodology and its application to U and W is proposed. Then, the corrected indexes are compared to their classical versions through a simulation study. The suggested correction yields controlled Type I errors against both conservatism and inflation, while the power to detect specific misfitting response patterns gets significantly increased. [less ▲] Detailed reference viewed: 43 (1 ULg)Accuracy of asymptotic standard errors of the maximum and weighted likelihood estimators of proficiency levels with short tests Magis, David in Applied Psychological Measurement (2014), 38 The maximum likelihood (ML) and the weighted likelihood (WL) estimators are commonly used to obtain proficiency level estimates with pre-calibrated item parameters. Both estimators have the same ... [more ▼] The maximum likelihood (ML) and the weighted likelihood (WL) estimators are commonly used to obtain proficiency level estimates with pre-calibrated item parameters. Both estimators have the same asymptotic standard error (ASE) that can be easily derived from the expected information function of the test. However, the accuracy of this asymptotic formula is uncertain with short tests when only a few items are administered. The purpose of this paper is to compare the ASE of these estimators to their exact values, evaluated at the proficiency level estimates. The exact SE is computed by generating the full exact sample distribution of the estimators, so its practical feasibility is limited to small tests (except under the Rasch model). A simulation study was conducted to compare the ASE and the exact SE of the ML and WL estimators, to the “true” SE (i.e., computed as the exact SE with the true proficiency levels). It is concluded that with small tests, the exact SEs are less biased and return smaller root mean squared error values than the asymptotic SEs, while as expected the two estimators return similar results with longer tests. [less ▲] Detailed reference viewed: 29 (5 ULg)Using R for conducting psychometric research Magis, David Scientific conference (2013, October 31) R is a statistical environment that permits to perform many various analyses, from basic descriptive statistics to advanced and complex modeling through high-level graphical features. In addition, it is ... [more ▼] R is a statistical environment that permits to perform many various analyses, from basic descriptive statistics to advanced and complex modeling through high-level graphical features. In addition, it is open-source software with thousands of users worldwide and with (almost) daily improvements by the development of additional packages. The purpose of this talk is twofold: (a) to provide a broad overview of R (in its basic form), RStudio (software for optimized display of R components) and R Commander (an add-on package with menus and toolboxes for R); and (b) to describe and illustrate the functioning of several R packages related to psychometrics in general and item response theory in particular. The talk will be at a theoretical low-level, focusing rather on applications and illustrations with examples. Topics to be covered may include (depending on time constraints): calibration of item response models, ability estimation, model checking, differential item functioning, computerized adaptive testing. Some information about the creation and compilation of R packages will also be given. References: R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/. [less ▲] Detailed reference viewed: 75 (5 ULg)Asymptotic distribution of robust estimators of ability and applications Magis, David Scientific conference (2013, October 31) In item response theory (IRT), ability estimation can be seriously affected by abnormal responses occurring from e.g., cheating, inattention, lack of time, guessing, tiredness, stress… Those phenomena may ... [more ▼] In item response theory (IRT), ability estimation can be seriously affected by abnormal responses occurring from e.g., cheating, inattention, lack of time, guessing, tiredness, stress… Those phenomena may influence the ability estimation process tremendously. One the one hand, person fit indices were developed as post-hoc approaches to identify abnormal responses patterns as a whole (e.g., Meijer & Sijtsma, 2001). On the other hand, getting uncontaminated ability estimates would also be a challenging issue. Robust estimators were proposed in the IRT framework to lessen the impact of abnormal responses onto the estimation process (Mislevy & Bock, 1982; Schuster & Yuan, 2011; Wainer & Wright, 1980). Yet, these estimators are still rarely used in practice, mostly because very little is known about their statistical properties. The purpose of this talk is to briefly present these robust ability estimators, and to derive their asymptotic distribution under mild regularity conditions. In particular, a simple formula for the asymptotic standard error (ASE) of these estimators is obtained (Magis, in press). Results of a simulation study that involves both presence and absence of cheating in the data generation process will be outlined. References: Magis, D. (in press). On the asymptotic standard error of a class of robust estimators of ability in dichotomous item response models. British Journal of Mathematical and Statistical Psychology. Meijer, R., & Sijtsma, K. (2001). Methodology review: Evaluating person fit. Applied Psychological Measurement, 25, 107-135. doi: 10.1177/01466210122031957 Mislevy, R. J., & Bock, R. D. (1982). Biweight estimates of latent ability. Educational and Psychological Measurement, 42, 725-737. doi: 10.1177/001316448204200302 Schuster, C., & Yuan, K.-H. (2011). Robust estimation of latent ability in item response models. Journal of Educational and Behavioral Statistics, 36, 720-735. doi: 10.3102/1076998610396890 Wainer, H., & Wright, B. D. (1980). Robust estimation of ability in the Rasch model. Psychometrika, 45, 373-391. doi: 10.1007/BF02293910 [less ▲] Detailed reference viewed: 46 (2 ULg)Asymptotic standard errors of robust estimators of ability in dichotomous item response models Magis, David Conference (2013, October 10) In item response theory, the classical estimators of ability are highly sensitive to response disturbances and can return strongly biased estimates of the true underlying ability level. Robust methods ... [more ▼] In item response theory, the classical estimators of ability are highly sensitive to response disturbances and can return strongly biased estimates of the true underlying ability level. Robust methods were introduced to lessen the impact of such aberrant responses onto the estimation process. The computation of asymptotic (i.e., large sample) standard errors (ASE) for these robust estimators, however, has not been fully considered yet. This paper focuses on a broad class of robust ability estimators, de ned by an appropriate selection of the weight function and the residual measure, for which the ASE is derived from the theory of estimating equations. The maximum likelihood (ML) and the robust estimators, together with their estimated ASE, are then compared through a simulation study. It is concluded that both the estimators and their ASE perform similarly in absence of response disturbances, while the robust estimator and its estimated ASE are less biased and outperform their ML counterparts in presence of response disturbances with large impact on the item response process. [less ▲] Detailed reference viewed: 8 (3 ULg)A small overview of available computer software to support computerized adaptive testing Magis, David Conference (2013, August 27) Computerized adaptive testing (CAT) is becoming a central tool for testing and assessment. It offers many advantages over fixed (“paper-and-pencil”) methods, such as individualized assessment, reduction ... [more ▼] Computerized adaptive testing (CAT) is becoming a central tool for testing and assessment. It offers many advantages over fixed (“paper-and-pencil”) methods, such as individualized assessment, reduction of fraud, and straightforward estimation of proficiency levels. CAT has been studied for decades and remains an up-to-date research field in psychometrics and educational science. Practical CAT administration, however, is less frequently considered in such studies. Assigning CAT to respondents requires both the sufficient availability of computer machines, and the use of a powerful and easy-to-use CAT software. With the fast increase of computer resources at moderate cost, the availability of computer machines is becoming a less central, yet important, issue in the practical assessment of CAT tests. The choice of an accurate CAT software, on the other hand, should be guided by its flexibility, its underlying statistical modeling, and its user-friendly potential. According to the type of research or data analysis, some CAT software might be preferred to another. It is therefore important for the researcher or the clinician to know about the current availability of such software, in line with current research and practice in the CAT framework. Moreover, these software should allow enough flexibility to incorporate updates and new theoretical developments, such as e.g., new rules for next item selection. This talk proposes a simple and user-oriented presentation of several CAT software that are currently available. The software to be presented are: the Firestar software (Choi, 2009), the R package catR (Magis & Raîche, 2012), the R package catIrt (Nydick, 2012) and the CAT web-platform Concerto (Kosinski & Rust, 2011). The first three are non-commercial software, while Concerto is a web interface between end users (willing to develop computerized assessment tests) and catR (as underlying routine software). Both R packages are written to be most useful for researchers, without end-user interface, and are therefore less appealing for applied researchers who are not familiar with R. Yet, they offer flexible solutions by means of many options to optimize the design of the test and generate many response patterns for further analyses. Also, they can be easily integrated as sub-routines for more sophisticated CAT software. Firestar provides a user interface and makes all necessary computations with underlying R code. This talk aims at focusing on freely available CAT software. For this reason, only the four aforementioned programs will be presented, although it exists other, commercial CAT software such as e.g., the CATSim software (Assessment Systems Corporation, 2012). The different CAT software are briefly presented and their advantages and drawbacks, flexibility and usefulness are compared, mostly from the point of view of the applied researcher and clinician. The following criteria were retained for objective comparison: (a) their main goal of application; (b) the type of data and IRT modeling they can deal with; (c) the type of users they are focusing on; (d) their operating options; (e) their availability and flexibility for further improvements. A small demonstration of the R package catR will be proposed optionally, depending on time limitation. References Assessment Systems Corporation (2012). CATSim: Comprehensive simulation of computerized adaptive testing. St. Paul, MN. URL: http://www.assess.com/. Choi , S. W. (2009). Firestar: Computerized adaptive testing simulation program for polytomous item response theory models. Applied Psychological Measurement, 33, 644-645. Kosinski, M., & Rust, J. (2011). The development of Concerto: An open source online adaptive testing platform. Paper presented at the International Association for Computerized and Adaptive Testing (IACAT), Pacific Grove, CA. Magis, D., & Raîche, G. (2012). Random generation of response patterns under computerized adaptive testing with the R package catR. Journal of Statistical Software, 48, 1-31. Nydick, S. W. (2012). catIrt: An R package for simulating IRT-based computerized adaptive tests. R package version 0.3-0. [less ▲] Detailed reference viewed: 256 (4 ULg)Equivalence of weighted likelihood and Jeffreys modal estimation of proficiency under polytomous item response models Magis, David Conference (2013, July 23) This talk focuses on two proficiency level estimators in item response theory (IRT) framework: the weighted likelihood estimator (WLE) and the Jeffreys modal estimator (JME), that is, the usual Bayes ... [more ▼] This talk focuses on two proficiency level estimators in item response theory (IRT) framework: the weighted likelihood estimator (WLE) and the Jeffreys modal estimator (JME), that is, the usual Bayes modal estimator with Jeffreys’ non-informative prior. With dichotomously scored items, the WLE and the JME are completely equivalent under the two-parameter logistic model, while remarkable relationships were established under the three-parameter logistic model. The purpose of this talk is to extend such comparison to polytomously scored items. It is shown that both WLE and JME are also equivalent for two broad classes of polytomous IRT models, including, among others, the (modified) graded response model, the (generalized) partial credit model, the rating scale model and the nominal response model. Parallelisms with dichotomously scored items are drawn. An example from a real data set is used to illustrate this finding. [less ▲] Detailed reference viewed: 31 (4 ULg)Proposition de nouveaux indices de détection de patrons de réponses inappropriés dans le contexte des enquêtes et des épreuves d’évaluation des apprentissages ; ; Magis, David Conference (2013, June 07) Il n’est pas rare de voir des étudiants répondre de façon inappropriée à une épreuve d’évaluation comportant des items à réponses choisies. Par exemple, certains individus peuvent tricher alors que ... [more ▼] Il n’est pas rare de voir des étudiants répondre de façon inappropriée à une épreuve d’évaluation comportant des items à réponses choisies. Par exemple, certains individus peuvent tricher alors que d’autres peuvent tenter de se sous-classer intentionnellement à un examen. Plusieurs approches ont été développées pour faire la détection de ce type d’étudiants. À ce jour, l’approche la plus prometteuse est l’utilisation d’indice de détection de type person‐fit (Meijer et Sijtsma, 2001). L’indice de détection lz (Drasgow, Levine et Williams, 1985) est fort probablement le plus utilisé et le plus connu d’entre tous. Malheureusement, cet indice est fortement affecté par le fait que l’habileté des étudiants est estimée et non pas réelle ; ce qui peut biaiser son calcul (Molenaar et Hoijtink, 1990). Pour pallier ce problème, Snijders (2001) a proposé une correction qui permet de diminuer considérablement le biais associé à la moyenne et à la variance de l’indice lz. Dans le cadre de notre projet doctoral, nous nous inspirerons de la suggestion de Snijders (2001) pour corriger deux autres indices de détection de patrons de réponses inappropriés : l’infit mean square (u) et l’outfit mean square (w). À cette fin, nous utiliserons une approche monte carliste afin d’investiguer plus en détails l’erreur de type 1 et la puissance de ces autres indices. Nos résultats préliminaires, que nous présenterons lors de cette communication, montrent que ces autres indices corrigés semblent eux aussi plus efficaces que leur version traditionnelle sans correction. [less ▲] Detailed reference viewed: 19 (2 ULg)Introduction to the R software Magis, David Scientific conference (2013, May 08) The R software (R Core Development Team, 2012) is an open-source statistical software that allows data handling, statistical analyses and model fitting, , and graphical representations, among others. It ... [more ▼] The R software (R Core Development Team, 2012) is an open-source statistical software that allows data handling, statistical analyses and model fitting, , and graphical representations, among others. It is very flexible and has lots of pre-installed statistical methods. It is working under all operating systems, including Windows, Linux/UNIX and MacOS. The R community is worldwide and proposes free exchanges of shared R packages through the CRAN (comprehensive R archive network). However, the user needs some practice to become familiar with the software, as it does not have easy to use interface, yet. The purpose of this workshop is to illustrate some aspects of this software with applied purposes. A data set from the field of clinical psychology will be considered throughout the workshop as an illustrative example. Data loading in R, data manipulation, summary and simple statistics, graphics, basic (t-tests, ANOVA, …) and advanced (factor analysis, generalized linear modeling, item response theory, …) statistical analyses will be described and illustrated. Live demonstrations will be run and participants will be encouraged to practice during the workshop. Participants are required to bring their own laptops, preferably with R already installed (technical assistance will be provided before the workshop to help participants in installing R if necessary). The workshop will be mostly Windows users oriented. The illustrative data set will also be available for participants. Useful references and links: 1) The R software website: http://www.r-project.org 2) The Use R! series of Springer books in general, and more precisely: Zuur, A. F., Ieno, E. N., & Meesters, E. H. W. G. (2009). A beginner’s guide to R. New York: Springer. [less ▲] Detailed reference viewed: 77 (22 ULg)Modèles polytomiques issus de la théorie de la réponse à l’item ; ; Magis, David Conference (2013, May 06) Les items qui composent les échelles de mesure en éducation et en sciences humaines sont fréquemment associées à plus de deux choix de réponses. Il s'agit alors d'items à réponses polytomiques. Plusieurs ... [more ▼] Les items qui composent les échelles de mesure en éducation et en sciences humaines sont fréquemment associées à plus de deux choix de réponses. Il s'agit alors d'items à réponses polytomiques. Plusieurs modélisations issues de la théorie de la réponse à l'item ont été proposées pour calibrer de telles échelles de mesures. Lorsque les choix de réponses ne sont pas ordonnés, les modèles à réponses nominales peuvent être utilisés. Lorsque ces choix sont ordonnés, les modèles gradués, à crédit partiel ou à appréciation peuvent être appliqués. Cette communication a pour objectif de présenter ces différentes modélisations et d'identifier des solutions logicielles pour effectuer la calibration. La présentation sera effectuée à l'aide d'un exemple auquel les différentes modélisations seront appliquées. Il sera ainsi possible de comparer les paramètres d'items et de personnes obtenus à l'aide de chacune des modélisations retenues. [less ▲] Detailed reference viewed: 68 (4 ULg)Application of lasso penalization to differential item functioning detection Magis, David Scientific conference (2013, February 26) Identification of differential item functioning (DIF) in dichotomously scored items is often performed item by item. This approach increases the risk of false discovery errors (Type I error rate) as all ... [more ▼] Identification of differential item functioning (DIF) in dichotomously scored items is often performed item by item. This approach increases the risk of false discovery errors (Type I error rate) as all items other than the tested one are assumed to be free of DIF. Some ad-hoc procedures, such as item purification and alpha level adjustment for multiple comparisons, have been studied in this context. The purpose of this talk is to focus on a different approach based on penalized likelihood estimation of a look-alike IRT model. Specifically, a Rasch model is being introduced with item-group interaction terms (i.e. DIF effects). Rather than obtaining pointwise estimates of the interaction parameters, which may be impossible because of high collinearity effects, the DIF effects are estimated with a lasso penalty term. Several criteria for optimally selecting the lasso tuning parameter are discussed, including cross-validation, AIC, BIC, and variants of these criteria. Preliminary results of a simulation study are presented and discussed. [less ▲] Detailed reference viewed: 79 (2 ULg) |
||