A Nonparametric Approach for Assessing Goodness-of-Fit of IRT Models in a Mixed Format Test

Investigating the fit of a parametric model plays a vital role in validating an item response theory (IRT) model. An area that has received little attention is the assessment of multiple IRT models used in a mixed-format test. The present study extends the nonparametric approach, proposed by Douglas and Cohen (2001), to assess model fit of three IRT models (three- and two-parameter logistic model, and generalized partial credit model) used in a mixed-format test. The statistical properties of the proposed fit statistic were examined and compared to S-X² and PARSCALE’s G². Overall, RISE (Root Integrated Square Error) outperformed the other two fit statistics under the studied conditions in that the Type I error rate was not inflated and the power was acceptable. A further advantage of the nonparametric approach is that it provides a convenient graphical inspection of the misfit.     

Performance of the Generalized S‐X2 Item Fit Index for Polytomous IRT Models

Orlando and Thissen's S‐X ² item fit index has performed better than traditional item fit statistics such as Yen's Q₁ and McKinley and Mill's G² for dichotomous item response theory (IRT) models. This study extends the utility of S‐X ² to polytomous IRT models, including the generalized partial credit model, partial credit model, and rating scale model. The performance of the generalized S‐X ² in assessing item model fit was studied in terms of empirical Type I error rates and power and compared to G². The results suggest that the generalized S‐X ² is promising for polytomous items in educational and psychological testing programs.     

A Comparison of Item Fit Statistics for Mixed IRT Models

In this study we examined procedures for assessing model-data fit of item response theory (IRT) models for mixed format data. The model fit indices used in this study include PARSCALE's G² , Orlando and Thissen's S − X² and S − G² , and Stone's χ²* and G²* . To investigate the relative performance of the fit statistics at the item level, we conducted two simulation studies: Type I error and power studies. We evaluated the performance of the item fit indices for various conditions of test length, sample size, and IRT models. Among the competing measures, the summed score-based indices S − X² and S − G² were found to be the sensible and efficient choice for assessing model fit for mixed format data. These indices performed well, particularly with short tests. The pseudo-observed score indices, χ²* and G²* , showed inflated Type I error rates in some simulation conditions. Consistent with the findings of current literature, the PARSCALE's G² index was rarely useful, although it provided reasonable results for long tests.     

An Assessment of the Nonparametric Approach for Evaluating the Fit of Item Response Models

As item response theory has been more widely applied, investigating the fit of a parametric model becomes an important part of the measurement process. There is a lack of promising solutions to the detection of model misfit in IRT. Douglas and Cohen introduced a general nonparametric approach, RISE (Root Integrated Squared Error), for detecting model misfit. The purposes of this study were to extend the use of RISE to more general and comprehensive applications by manipulating a variety of factors (e.g., test length, sample size, IRT models, ability distribution). The results from the simulation study demonstrated that RISE outperformed G² and S‐X² in that it controlled Type I error rates and provided adequate power under the studied conditions. In the empirical study, RISE detected reasonable numbers of misfitting items compared to G² and S‐X², and RISE gave a much clearer picture of the location and magnitude of misfit for each misfitting item. In addition, there was no practical consequence to classification before and after replacement of misfitting items detected by three fit statistics.     

Effect of Sample Size Ratio and Model Misfit When Using the Difficulty Parameter Differences Procedure to Detect DIF

This study examined the effect of sample size ratio and model misfit on the Type I error rates and power of the Difficulty Parameter Differences procedure using Winsteps. A unidimensional 30-item test with responses from 130,000 examinees was simulated and four independent variables were manipulated: sample size ratio (20/100/250/500/1000); model fit/misfit (1 PL and 3PLc =. 15 models); impact (no difference/mean differences/variance differences/mean and variance differences); and percentage of items with uniform and nonuniform DIF (0%/10%/20%). In general, the results indicate the importance of ensuring model fit to achieve greater control of Type I error and adequate statistical power. The manipulated variables produced inflated Type I error rates, which were well controlled when a measure of DIF magnitude was applied. Sample size ratio also had an effect on the power of the procedure. The paper discusses the practical implications of these results.     

Multiplicity Control in Structural Equation Modeling

Abstract

Researchers conducting structural equation modeling analyses rarely, if ever, control for the inflated probability of Type I errors when evaluating the statistical significance of multiple parameters in a model. In this study, the Type I error control, power and true model rates of famsilywise and false discovery rate controlling procedures were compared with rates when no multiplicity control was imposed. The results indicate that Type I error rates become severely inflated with no multiplicity control, but also that familywise error controlling procedures were extremely conservative and had very little power for detecting true relations. False discovery rate controlling procedures provided a compromise between no multiplicity control and strict familywise error control and with large sample sizes provided a high probability of making correct inferences regarding all the parameters in the model.     

The Impact of Noninvariant Intercepts in Latent Means Models

Latent means methods such as multiple-indicator multiple-cause (MIMIC) and structured means modeling (SMM) allow researchers to determine whether or not a significant difference exists between groups' factor means. Strong invariance is typically recommended when interpreting latent mean differences. The extent of the impact of noninvariant intercepts on conclusions made when implementing both MIMIC and SMM methods was the main purpose of this study. The impact of intercept noninvariance on Type I error rates, power, and two model fit indices when using MIMIC and SMM approaches under various conditions were examined. Type I error and power were adversely affected by intercept noninvariance. Although the fit indices did not detect small misspecifications in the form of noninvariant intercepts, one did perform more optimally.     

Evaluating the Wald Test for Item‐Level Comparison of Saturated and Reduced Models in Cognitive Diagnosis

This article used the Wald test to evaluate the item‐level fit of a saturated cognitive diagnosis model (CDM) relative to the fits of the reduced models it subsumes. A simulation study was carried out to examine the Type I error and power of the Wald test in the context of the G‐DINA model. Results show that when the sample size is small and a larger number of attributes are required, the Type I error rate of the Wald test for the DINA and DINO models can be higher than the nominal significance levels, while the Type I error rate of the A‐CDM is closer to the nominal significance levels. However, with larger sample sizes, the Type I error rates for the three models are closer to the nominal significance levels. In addition, the Wald test has excellent statistical power to detect when the true underlying model is none of the reduced models examined even for relatively small sample sizes. The performance of the Wald test was also examined with real data. With an increasing number of CDMs from which to choose, this article provides an important contribution toward advancing the use of CDMs in practical educational settings.     

Mediated Effects with the Parallel Process Latent Growth Model: An Evaluation of Methods for Testing Mediation in the Presence of Nonnormal Data

This Monte Carlo simulation study investigated the impact of nonnormality on estimating and testing mediated effects with the parallel process latent growth model and 3 popular methods for testing the mediated effect (i.e., Sobel’s test, the asymmetric confidence limits, and the bias-corrected bootstrap). It was found that nonnormality had little effect on the estimates of the mediated effect, standard errors, empirical Type I error, and power rates in most conditions. In terms of empirical Type I error and power rates, the bias-corrected bootstrap performed best. Sobel’s test produced very conservative Type I error rates when the estimated mediated effect and standard error had a relationship, but when the relationship was weak or did not exist, the Type I error was closer to the nominal .05 value.     

Random Permutation Testing Applied to Measurement Invariance Testing with Ordered-Categorical Indicators

We describe and evaluate a random permutation test of measurement invariance with ordered-categorical data. To calculate a p-value for the observed (?)χ², an empirical reference distribution is built by repeatedly shuffling the grouping variable, then saving the χ² from a configural model, or the ?χ² between configural and scalar-invariance models, fitted to each permuted dataset. The current gold standard in this context is a robust mean- and variance-adjusted ?χ² test proposed by Satorra (2000), which yields inflated Type I errors, particularly when thresholds are asymmetric, unless samples sizes are quite large (Bandalos, 2014; Sass et al., 2014). In a Monte Carlo simulation, we compare permutation to three implementations of Satorra’s robust χ² across a variety of conditions evaluating configural and scalar invariance. Results suggest permutation can better control Type I error rates while providing comparable power under conditions that the standard robust test yields inflated errors.     

An Investigation of the Power of the Likelihood Ratio Goodness-of-Fit Statistic in Detecting Differential Item Functioning

The purpose of this study was to investigate the power and Type I error rate of the likelihood ratio goodness-of-fit (LR) statistic in detecting differential item functioning (DIF) under Samejima's (1969, 1972) graded response model. A multiple-replication Monte Carlo study was utilized in which DIF was modeled in simulated data sets which were then calibrated with MULTILOG (Thissen, 1991) using hierarchically nested item response models. In addition, the power and Type I error rate of the Mantel (1963) approach for detecting DIF in ordered response categories were investigated using the same simulated data, for comparative purposes. The power of both the Mantel and LR procedures was affected by sample size, as expected. The LR procedure lacked the power to consistently detect DIF when it existed in reference/focal groups with sample sizes as small as 500/500. The Mantel procedure maintained control of its Type I error rate and was more powerful than the LR procedure when the comparison group ability distributions were identical and there was a constant DIF pattern. On the other hand, the Mantel procedure lost control of its Type I error rate, whereas the LR procedure did not, when the comparison groups differed in mean ability; and the LR procedure demonstrated a profound power advantage over the Mantel procedure under conditions of balanced DIF in which the comparison group ability distributions were identical. The choice and subsequent use of any procedure requires a thorough understanding of the power and Type I error rates of the procedure under varying conditions of DIF pattern, comparison group ability distributions.–or as a surrogate, observed score distributions–and item characteristics.     

Evaluating Model Fit With Ordered Categorical Data Within a Measurement Invariance Framework: A Comparison of Estimators

A paucity of research has compared estimation methods within a measurement invariance (MI) framework and determined if research conclusions using normal-theory maximum likelihood (ML) generalizes to the robust ML (MLR) and weighted least squares means and variance adjusted (WLSMV) estimators. Using ordered categorical data, this simulation study aimed to address these queries by investigating 342 conditions. When testing for metric and scalar invariance, Δχ² results revealed that Type I error rates varied across estimators (ML, MLR, and WLSMV) with symmetric and asymmetric data. The Δχ² power varied substantially based on the estimator selected, type of noninvariant indicator, number of noninvariant indicators, and sample size. Although some the changes in approximate fit indexes (ΔAFI) are relatively sample size independent, researchers who use the ΔAFI with WLSMV should use caution, as these statistics do not perform well with misspecified models. As a supplemental analysis, our results evaluate and suggest cutoff values based on previous research.     

Hierarchical Logistic Regression: Accounting for Multilevel Data in DIF Detection

The purpose of this study was to examine the performance of differential item functioning (DIF) assessment in the presence of a multilevel structure that often underlies data from large-scale testing programs. Analyses were conducted using logistic regression (LR), a popular, flexible, and effective tool for DIF detection. Data were simulated using a hierarchical framework, such as might be seen when examinees are clustered in schools, for example. Both standard and hierarchical LR (accounting for multilevel data) approaches to DIF detection were employed. Results highlight the differences in DIF detection rates when the analytic strategy matches the data structure. Specifically, when the grouping variable was within clusters, LR and HLR performed similarly in terms of Type I error control and power. However, when the grouping variable was between clusters, LR failed to maintain the nominal Type I error rate of .05. HLR was able to maintain this rate. However, power for HLR tended to be low under many conditions in the between cluster variable case.     

Improving the Root Mean Square Error of Approximation for Nonnormal Conditions in Structural Equation Modeling

In this study, the authors investigated incorporating adjusted model fit information into the root mean square error of approximation (RMSEA) fit index. Through Monte Carlo simulation, the usefulness of this adjusted index was evaluated for assessing model adequacy in structural equation modeling when the multivariate normality assumption underlying maximum likelihood estimation is violated. Adjustment to the RMSEA was considered in 2 forms: a rescaling adjustment via the Satorra-Bentler rescaled goodness-of-fit statistic and a bootstrap adjustment via the Bollen and Stine adjusted model p value. Both properly specified and misspecifed models were examined. The adjusted RMSEA was evaluated in terms of the average index value across study conditions and with respect to model rejection rates under tests of exact fit, close fit, and not-close fit.     

Using Fit Indexes to Select a Covariance Model for Longitudinal Data

This study investigated the performance of fit indexes in selecting a covariance structure for longitudinal data. Data were simulated to follow a compound symmetry, first-order autoregressive, first-order moving average, or random-coefficients covariance structure. We examined the ability of the likelihood ratio test (LRT), root mean square error of approximation (RMSEA), comparative fit index (CFI), and Tucker–Lewis Index (TLI) to reject misspecified models with varying degrees of misspecification. With a sample size of 20, RMSEA, CFI, and TLI are high in both Type I and Type II error rates, whereas LRT has a high Type II error rate. With a sample size of 100, these indexes generally have satisfactory performance, but CFI and TLI are affected by a confounding effect of their baseline model. Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC) have high success rates in identifying the true model when sample size is 100. A comparison with the mixed model approach indicates that separately modeling the means and covariance structures in structural equation modeling dramatically improves the success rate of AIC and BIC.     

Goodness of Fit in Item Factor Analysis: Effect of the Number of Response Alternatives

The power of the chi-square test statistic used in structural equation modeling decreases as the absolute value of excess kurtosis of the observed data increases. Excess kurtosis is more likely the smaller the number of item response categories. As a result, fit is likely to improve as the number of item response categories decreases, regardless of the true underlying factor structure or χ²-based fit index used to examine model fit. Equivalently, given a target value of approximate fit (e.g., root mean square error of approximation ≤ .05) a model with more factors is needed to reach it as the number of categories increases. This is true regardless of whether the data are treated as continuous (common factor analysis) or as discrete (ordinal factor analysis). We recommend using a large number of response alternatives (≥ 5) to increase the power to detect incorrect substantive models.     

Missing Data and Multiple Imputation in the Context of Multivariate Analysis of Variance

Multivariate analysis of variance (MANOVA) is widely used in educational research to compare means on multiple dependent variables across groups. Researchers faced with the problem of missing data often use multiple imputation of values in place of the missing observations. This study compares the performance of 2 methods for combining p values in the context of a MANOVA, with the typical default for dealing with missing data: listwise deletion. When data are missing at random, the new methods maintained the nominal Type I error rate and had power comparable to the complete data condition. When 40% of the data were missing completely at random, the Type I error rates for the new methods were inflated, but not for lower percents.     

A Study Of The Relationship Between Grade And Age And Variability

Power and stability of Type I error rates are investigated for the Box-Scheffé test of homogeneity of variance with varying subsample sizes under conditions of normality and nonnormality. The test is shown to be robust to violation of the normality assumption when sampling is from a leptokurtic population. Subsample sizes which produce maximum power are given for small, intermediate, and large sample situations. Suggestions for selecting subsample sizes which will produce maximum power for a given n are provided. A formula for estimating power in the equal n case is shown to give results agreeing with empirical results.     

The Effects of Type I Error Rate and Power of the ANCOVA F Test and Selected Alternatives Under Nonnormality and Variance Heterogeneity

The authors sought to identify through Monte Carlo simulations those conditions for which analysis of covariance (ANCOVA) does not maintain adequate Type I error rates and power. The conditions that were manipulated included assumptions of normality and variance homogeneity, sample size, number of treatment groups, and strength of the covariate-dependent variable relationship. Alternative tests studied were Quade's procedure, Puri and Sen's solution, Burnett and Barr's rank difference scores, Conover and Iman's rank transformation test, Hettmansperger's procedure, and the Puri-Sen-Harwell-Serlin test. For balanced designs, the ANCOVA F test was robust and was often the most powerful test through all sample-size designs and distributional configurations. With unbalanced designs, with variance heterogeneity, and when the largest treatment-group variance was matched with the largest group sample size, the nonparametric alternatives generally outperformed the ANCOVA test. When sample size and variance ratio were inversely coupled, all tests became very liberal; no test maintained adequate control over Type I error.