IMPLICATIONS OF CRITERION-REFERENCED MEASUREMENT1,2

During the past several years measurement and instructional specialists have distinguished between norm-referenced and criterion-referenced approaches to measurement. More traditional, a norm-referenced measure is used to identify an individual's performance in relation to the performance of others on the same measure. A criterion-referenced test is used to identify an individual's status with respect to an established standard of performance. This discussion examines the implications of these two approaches to measurement, particularly criterion-referenced measurement, with respect to variability, item construction, reliability, validity, item analysis, reporting, and interpretation.     

Measurement Models,Estimation, and the Study of Change

The study of change is based on the idea that the score or index at each measurement occasion has the same meaning and metric across time. In tests or scales with multiple items, such as those common in the social sciences, there are multiple ways to create such scores. Some options include using raw or sum scores (i.e., sum of item responses or linear transformation thereof), using Rasch-scaled scores provided by the test developers, fitting item response models to the observed item responses and estimating ability or aptitude, and jointly estimating the item response and growth models. We illustrate that this choice can have an impact on the substantive conclusions drawn from the change analysis using longitudinal data from the Applied Problems subtest of the Woodcock–Johnson Psycho-Educational Battery–Revised collected as part of the National Institute of Child Health and Human Development's Study of Early Child Care. Assumptions of the different measurement models, their benefits and limitations, and recommendations are discussed.     

A Nonparametric Approach to Estimate Classification Accuracy and Consistency

When cut scores for classifications occur on the total score scale, popular methods for estimating classification accuracy (CA) and classification consistency (CC) require assumptions about a parametric form of the test scores or about a parametric response model, such as item response theory (IRT). This article develops an approach to estimate CA and CC nonparametrically by replacing the role of the parametric IRT model in Lee's classification indices with a modified version of Ramsay's kernel‐smoothed item response functions. The performance of the nonparametric CA and CC indices are tested in simulation studies in various conditions with different generating IRT models, test lengths, and ability distributions. The nonparametric approach to CA often outperforms Lee's method and Livingston and Lewis's method, showing robustness to nonnormality in the simulated ability. The nonparametric CC index performs similarly to Lee's method and outperforms Livingston and Lewis's method when the ability distributions are nonnormal.     

Testing the Equivalence Among Different Item Response Formats in Personality Measurement: A Structural Equation Modeling Approach

This study discusses a procedure for testing the equivalence among different item response formats used in personality and attitude measurement. The procedure is based on the assumption that latent response variables underlie the observed item responses (underlying variables approach) and uses a nested series of confirmatory factor analysis models derived from Joreskog's (1971) method for estimating the dissatenuated correlation. The different stages of the procedure are illustrated using real data.     

Person‐Fit Statistics for Joint Models for Accuracy and Speed

Response accuracy and response time data can be analyzed with a joint model to measure ability and speed of working, while accounting for relationships between item and person characteristics. In this study, person‐fit statistics are proposed for joint models to detect aberrant response accuracy and/or response time patterns. The person‐fit tests take the correlation between ability and speed into account, as well as the correlation between item characteristics. They are posited as Bayesian significance tests, which have the advantage that the extremeness of a test statistic value is quantified by a posterior probability. The person‐fit tests can be computed as by‐products of a Markov chain Monte Carlo algorithm. Simulation studies were conducted in order to evaluate their performance. For all person‐fit tests, the simulation studies showed good detection rates in identifying aberrant patterns. A real data example is given to illustrate the person‐fit statistics for the evaluation of the joint model.     

Detecting Careless Responding in Survey Data Using Stochastic Gradient Boosting

Careless responding is a bias in survey responses that disregards the actual item content, constituting a threat to the factor structure, reliability, and validity of psychological measurements. Different approaches have been proposed to detect aberrant responses such as probing questions that directly assess test-taking behavior (e.g., bogus items), auxiliary or paradata (e.g., response times), or data-driven statistical techniques (e.g., Mahalanobis distance). In the present study, gradient boosted trees, a state-of-the-art machine learning technique, are introduced to identify careless respondents. The performance of the approach was compared with established techniques previously described in the literature (e.g., statistical outlier methods, consistency analyses, and response pattern functions) using simulated data and empirical data from a web-based study, in which diligent versus careless response behavior was experimentally induced. In the simulation study, gradient boosting machines outperformed traditional detection mechanisms in flagging aberrant responses. However, this advantage did not transfer to the empirical study. In terms of precision, the results of both traditional and the novel detection mechanisms were unsatisfactory, although the latter incorporated response times as additional information. The comparison between the results of the simulation and the online study showed that responses in real-world settings seem to be much more erratic than can be expected from the simulation studies. We critically discuss the generalizability of currently available detection methods and provide an outlook on future research on the detection of aberrant response patterns in survey research.     

The Best of Both Worlds: Factor Analysis of Dichotomous Data Using Item Response Theory and Structural Equation Modeling

Both structural equation modeling (SEM) and item response theory (IRT) can be used for factor analysis of dichotomous item responses. In this case, the measurement models of both approaches are formally equivalent. They were refined within and across different disciplines, and make complementary contributions to central measurement problems encountered in almost all empirical social science research fields. In this article (a) fundamental formal similiarities between IRT and SEM models are pointed out. It will be demonstrated how both types of models can be used in combination to analyze (b) the dimensional structure and (c) the measurement invariance of survey item responses. All analyses are conducted with Mplus, which allows an integrated application of both approaches in a unified, general latent variable modeling framework. The aim is to promote a diffusion of useful measurement techniques and skills from different disciplines into empirical social research.     

A Mixture Group Bifactor Model for Binary Responses

The presence of nuisance dimensionality is a potential threat to the accuracy of results for tests calibrated using a measurement model such as a factor analytic model or an item response theory model. This article describes a mixture group bifactor model to account for the nuisance dimensionality due to a testlet structure as well as the dimensionality due to differences in patterns of responses. The model can be used for testing whether or not an item functions differently across latent groups in addition to investigating the differential effect of local dependency among items within a testlet. An example is presented comparing test speededness results from a conventional factor mixture model, which ignores the testlet structure, with results from the mixture group bifactor model. Results suggested the 2 models treated the data somewhat differently. Analysis of the item response patterns indicated that the 2-class mixture bifactor model tended to categorize omissions as indicating speededness. With the mixture group bifactor model, more local dependency was present in the speeded than in the nonspeeded class. Evidence from a simulation study indicated the Bayesian estimation method used in this study for the mixture group bifactor model can successfully recover generated model parameters for 1- to 3-group models for tests containing testlets.     

An NCME Instructional Module on Polytomous Item Response Theory Models

A polytomous item is one for which the responses are scored according to three or more categories. Given the increasing use of polytomous items in assessment practices, item response theory (IRT) models specialized for polytomous items are becoming increasingly common. The purpose of this ITEMS module is to provide an accessible overview of polytomous IRT models. The module presents commonly encountered polytomous IRT models, describes their properties, and contrasts their defining principles and assumptions. After completing this module, the reader should have a sound understating of what a polytomous IRT model is, the manner in which the equations of the models are generated from the model's underlying step functions, how widely used polytomous IRT models differ with respect to their definitional properties, and how to interpret the parameters of polytomous IRT models.     

Correlated Errors in True Score Models and Their Effect on Coefficient Alpha

It is well known that coefficient alpha is an estimate of reliability if its underlying assumptions are met and that it is a lower-bound estimate if the assumption of essential tau equivalency is violated. Very little literature addresses the assumption of uncorrelated errors among items and the effect of violating this assumption on alpha. True score models are proposed that can account for correlated errors. These models allow random measurement errors on earlier items to affect directly or indirectly scores on later items. Coefficient alpha may yield spuriously high estimates of reliability if these true score models reflect item responding. In practice, it is important to differentiate these models from models in which the errors are correlated because 1 or more factors have been left unspecified. If the latter model is an accurate representation of item responding, the assumption of essential tau equivalency is violated and alpha is a lower-bound estimate of reliability.     

Information Functions of Rank-2PL Models for Forced-Choice Questionnaires

This paper presents the item and test information functions of the Rank two-parameter logistic models (Rank-2PLM) for items with two (pair) and three (triplet) statements in forced-choice questionnaires. The Rank-2PLM model for pairs is the MUPP-2PLM (Multi-Unidimensional Pairwise Preference) and, for triplets, is the Triplet-2PLM. Fisher's information and directional information are described, and the test information for Maximum Likelihood (ML), Maximum A Posterior (MAP), and Expected A Posterior (EAP) trait score estimates is distinguished. Expected item/test information indexes at various levels are proposed and plotted to provide diagnostic information on items and tests. The expected test information indexes for EAP scores may be difficult to compute due to a typical test's vast number of item response patterns. The relationships of item/test information with discrimination parameters of statements, standard error, and reliability estimates of trait score estimates are discussed and demonstrated using real data. Practical suggestions for checking the various expected item/test information indexes and plots are provided.     

Detecting Differential Item Discrimination (DID) and the Consequences of Ignoring DID in Multilevel Item Response Models

Cross‐level invariance in a multilevel item response model can be investigated by testing whether the within‐level item discriminations are equal to the between‐level item discriminations. Testing the cross‐level invariance assumption is important to understand constructs in multilevel data. However, in most multilevel item response model applications, the cross‐level invariance is assumed without testing of the cross‐level invariance assumption. In this study, the detection methods of differential item discrimination (DID) over levels and the consequences of ignoring DID are illustrated and discussed with the use of multilevel item response models. Simulation results showed that the likelihood ratio test (LRT) performed well in detecting global DID at the test level when some portion of the items exhibited DID. At the item level, the Akaike information criterion (AIC), the sample‐size adjusted Bayesian information criterion (saBIC), LRT, and Wald test showed a satisfactory rejection rate (>.8) when some portion of the items exhibited DID and the items had lower intraclass correlations (or higher DID magnitudes). When DID was ignored, the accuracy of the item discrimination estimates and standard errors was mainly problematic. Implications of the findings and limitations are discussed.     

多维项目反应理论在数学素养测验中的应用

学生的数学素养具有多维结构,素养导向的数学学业成就测评需要提供被试在各维度上的表现信息,而不仅是一个单一的总分。以PISA数学素养结构为理论模型,以多维项目反应理论(MIRT)为测量模型,利用R语言的MIRT程序包处理和分析某地区8年级数学素养测评题目数据,研究数学素养的多维测量方法。结果表明:MIRT兼具单维项目反应理论和因子分析的优点,利用其可对测试的结构效度和测试题目质量进行分析,以及对被试进行多维能力认知诊断。     

Minimizing the Influence of Item Parameter Estimation Errors in Test Development: A Comparison of Three Selection Procedures

In test development, item response theory (IRT) is a method to determine the amount of information that each item (i.e., item information function) and combination of items (i.e., test information function) provide in the estimation of an examinee's ability. Studies investigating the effects of item parameter estimation errors over a range of ability have demonstrated an overestimation of information when the most discriminating items are selected (i.e., item selection based on maximum information). In the present study, the authors examined the influence of item parameter estimation errors across 3 item selection methods—maximum no target, maximum target, and theta maximum—using the 2- and 3-parameter logistic IRT models. Tests created with the maximum no target and maximum target item selection procedures consistently overestimated the test information function. Conversely, tests created using the theta maximum item selection procedure yielded more consistent estimates of the test information function and, at times, underestimated the test information function. Implications for test development are discussed.     

On the Issue of Item Selection in Computerized Adaptive Testing With Response Times

Many standardized tests are now administered via computer rather than paper‐and‐pencil format. The computer‐based delivery mode brings with it certain advantages. One advantage is the ability to adapt the difficulty level of the test to the ability level of the test taker in what has been termed computerized adaptive testing (CAT). A second advantage is the ability to record not only the test taker's response to each item (i.e., question), but also the amount of time the test taker spends considering and answering each item. Combining these two advantages, various methods were explored for utilizing response time data in selecting appropriate items for an individual test taker. Four strategies for incorporating response time data were evaluated, and the precision of the final test‐taker score was assessed by comparing it to a benchmark value that did not take response time information into account. While differences in measurement precision and testing times were expected, results showed that the strategies did not differ much with respect to measurement precision but that there were differences with regard to the total testing time.     

Performance of Person-Fit Statistics Under Model Misspecification

In educational and psychological measurement, a person-fit statistic (PFS) is designed to identify aberrant response patterns. For parametric PFSs, valid inference depends on several assumptions, one of which is that the item response theory (IRT) model is correctly specified. Previous studies have used empirical data sets to explore the effects of model misspecification on PFSs. We further this line of research by using a simulation study, which allows us to explore issues that may be of interest to practitioners. Results show that, depending on the generating and analysis item models, Type I error rates at fixed values of the latent variable may be greatly inflated, even when the aggregate rates are relatively accurate. Results also show that misspecification is most likely to affect PFSs for examinees with extreme latent variable scores. Two empirical data analyses are used to illustrate the importance of model specification.     

AN EVALUATION MODEL FOR MASTERY TESTING1

Noting the desirability of the current shift toward mastery testing and criterion-referenced test procedures, an evaluation model is presented which should be useful and practical for such purposes. This model is based on the assumptions that the learning of fundamental skills can be considered all or none, that each item response on a single skill test represents an unbiased sample of the examinee's true mastery status, that measurement error occurring on the test (as estimated from the average interitem correlation) can be of only one type (α or β) for each examinee, and that through practical and theoretical considerations of evaluation error costs and item error characteristics, an optimal mastery criterion can be calculated. Each of these assumptions is discussed and the resultant mastery criteria algorithm is described along with an example from the IPI math program.     

Assessing the Dimensionality of NAEP Reading Data

The reading data from the 1983–84 National Assessment of Educational Progress survey were scaled using a unidimensional item response theory model. To determine whether the responses to the reading items were consistent with unidimensionality, the full-information factor analysis method developed by Bock and associates (1985) and Rosenbaum's (1984) test of unidimensionality, conditional (local) independence, and monotonicity were applied. Full-information factor analysis involves the assumption of a particular item response function; the number of latent variables required to obtain a reasonable fit to the data is then determined. The Rosenbaum method provides a test of the more general hypothesis that the data can be represented by a model characterized by unidimensionality, conditional independence, and monotonicity. Results of both methods indicated that the reading items could be regarded as measures of a single dimension. Simulation studies were conducted to investigate the impact of balanced incomplete block (BIB) spiraling, used in NAEP to assign items to students, on methods of dimensionality assessment. In general, conclusions about dimensionality were the same for BIB-spiraled data as for complete data.     

The Use of Hierarchical Generalized Linear Model for Item Dimensionality Assessment

To assess item dimensionality, the following two approaches are described and compared: hierarchical generalized linear model (HGLM) and multidimensional item response theory (MIRT) model. Two generating models are used to simulate dichotomous responses to a 17-item test: the unidimensional and compensatory two-dimensional (C2D) models. For C2D data, seven items are modeled to load on the first and second factors, θ₁ and θ₂, with the remaining 10 items modeled unidimensionally emulating a mathematics test with seven items requiring an additional reading ability dimension. For both types of generated data, the multidimensionality of item responses is investigated using HGLM and MIRT. Comparison of HGLM and MIRT's results are possible through a transformation of items' difficulty estimates into probabilities of a correct response for a hypothetical examinee at the mean on θ and θ₂. HGLM and MIRT performed similarly. The benefits of HGLM for item dimensionality analyses are discussed.     

The Hierarchy Consistency Index: Evaluating Person Fit for Cognitive Diagnostic Assessment

In this article, we introduce a person‐fit statistic called the hierarchy consistency index (HCI) to help detect misfitting item response vectors for tests developed and analyzed based on a cognitive model. The HCI ranges from ?1.0 to 1.0, with values close to ?1.0 indicating that students respond unexpectedly or differently from the responses expected under a given cognitive model. A simulation study was conducted to evaluate the power of the HCI in detecting different types of misfitting item response vectors. Simulation results revealed that the detection rate of the HCI was a function of type of misfit, item discriminating power, and test length. The best detection rates were achieved when the HCI was applied to tests that consisted of a large number of highly discriminating items. In addition, whether a misfitting item response vector can be correctly identified depends, to a large degree, on the number of misfits of the item response vector relative to the cognitive model. When misfitting response behavior only affects a small number of item responses, the resulting item response vector will not be substantially different from the expectations under the cognitive model and consequently may not be statistically identified as misfitting. As an item response vector deviates further from the model expectations, misfits are more easily identified and consequently higher detection rates of the HCI are expected.