Evaluation of Two New Smoothing Methods in Equating: The Cubic B-Spline Presmoothing Method and the Direct Presmoothing Method

This article considers two new smoothing methods in equipercentile equating , the cubic B-spline presmoothing method and the direct presmoothing method. Using a simulation study , these two methods are compared with established methods , the beta-4 method , the polynomial loglinear method , and the cubic spline postsmoothing method , under three sample sizes (300 , 1,000 , and 3,000) and for three test content areas (ITBS Maps and Diagrams , ITBS Reference and Materials , and ITBS Capitalization). Ten thousand random samples were simulated from population distributions , and the standard error , bias , and RMSE statistics were calculated. The cubic B-spline presmoothing method performed well in reducing total error of equating , whereas the direct presmoothing method appeared to need some modification for it to be as accurate as other smoothing methods.     

Accuracy of Random Groups Equating with Very Small Samples

This study investigated the effectiveness of equating with very small samples using the random groups design. Of particular interest was equating accuracy at specific scores where performance standards might be set. Two sets of simulations were carried out, one in which the two forms were identical and one in which they differed by a tenth of a standard deviation in overall difficulty. These forms were equated using mean equating, linear equating, unsmoothed equipercentile equating, and equipercentile equating using two through six moments of log-linear presmoothing with samples of 25, 50, 75, 100, 150, and 200. The results indicated that identity equating was preferable to any equating method when samples were as small as 25. For samples of 50 and above, the choice of an equating method over identity equating depended on the location of the passing score relative to examinee performance. If passing scores were located below the mean, where data were sparser, mean equating produced the smallest percentage of misclassified examinees. For passing scores near the mean, all methods produced similar results with linear equating being the most accurate. For passing scores above the mean, equipercentile equating with 2- and 3-moment presmoothing were the best equating methods. Higher levels of presmoothing did not improve the results.     

A New Statistic to Assess Fitness of Cubic-Spline Postsmoothing

In equating, smoothing techniques are frequently used to diminish sampling error. There are typically two types of smoothing: presmoothing and postsmoothing. For polynomial log-linear presmoothing, an optimum smoothing degree can be determined statistically based on the Akaike information criterion or Chi-square difference criterion. For cubic-spline postsmoothing, visual inspection has been an important tool in choosing such optimum degrees in operational settings. This study introduces a new statistic for assessing the fitness of the cubic-spline postsmoothing method, which accommodates three conditions: (1) one standard error band, (2) deviation from unsmoothed equivalents, and (3) smoothness. A principal advantage of the new statistic proposed in this study is that an optimum degree of smoothing can be selected automatically by giving consistent amount of attention to deviation and smoothness across multiple equatings, whereas visual inspection may not be consistent.     

Futility1 of Log‐Linear Smoothing when Equating with Unrepresentative Small Samples

The impact of log‐linear presmoothing on the accuracy of small sample chained equipercentile equating was evaluated under two conditions . In the first condition the small samples differed randomly in ability from the target population. In the second condition the small samples were systematically different from the target population. Results showed that equating with small samples (e.g., N < 25 or 50) using either raw or smoothed score distributions led to considerable large random equating error (although smoothing reduced random equating error). Moreover, when the small samples were not representative of the target population, the amount of equating bias also was quite large. It is concluded that although presmoothing can reduce random equating error, it is not likely to reduce equating bias caused by using an unrepresentative sample. Other alternatives to the small sample equating problem (e.g., the SiGNET design) which focus more on improving data collection are discussed.     

Small‐Sample Equating Using a Synthetic Linking Function

This study addressed the sampling error and linking bias that occur with small samples in a nonequivalent groups anchor test design. We proposed a linking method called the synthetic function, which is a weighted average of the identity function and a traditional equating function (in this case, the chained linear equating function). Specifically, we compared the synthetic, identity, and chained linear functions for various‐sized samples from two types of national assessments. One design used a highly reliable test and an external anchor, and the other used a relatively low‐reliability test and an internal anchor. The results from each of these methods were compared to the criterion equating function derived from the total samples with respect to linking bias and error. The study indicated that the synthetic functions might be a better choice than the chained linear equating method when samples are not large and, as a result, unrepresentative.     

Practical Application of a Synthetic Linking Function on Small-Sample Equating

The synthetic function is a weighted average of the identity (the linking function for forms that are known to be completely parallel) and a traditional equating method. The purpose of the present study was to investigate the benefits of the synthetic function on small-sample equating using various real data sets gathered from different administrations of tests from a licensure testing program. We investigated the chained linear, Tucker, Levine, and mean equating methods, along with the identity and the synthetic functions with small samples (N = 19 to 70). The synthetic function did not perform as well as did other linear equating methods because test forms differed markedly in difficulty; thus, the use of the identity function produced substantial bias. The effectiveness of the synthetic function depended on the forms' similarity in difficulty.     

Statistical Assessment of Estimated Transformations in Observed‐Score Equating

Equating methods make use of an appropriate transformation function to map the scores of one test form into the scale of another so that scores are comparable and can be used interchangeably. The equating literature shows that the ways of judging the success of an equating (i.e., the score transformation) might differ depending on the adopted framework. Rather than targeting different parts of the equating process and aiming to evaluate the process from different aspects, this article views the equating transformation as a standard statistical estimator and discusses how this estimator should be assessed in an equating framework. For the kernel equating framework, a numerical illustration shows the potentials of viewing the equating transformation as a statistical estimator as opposed to assessing it using equating‐specific criteria. A discussion on how this approach can be used to compare other equating estimators from different frameworks is also included.     

A Comparison of the Kernel Equating Method with Traditional Equating Methods Using SAT® Data

This study applied kernel equating (KE) in two scenarios: equating to a very similar population and equating to a very different population, referred to as a distant population, using SAT^® data. The KE results were compared to the results obtained from analogous traditional equating methods in both scenarios. The results indicate that KE results are comparable to the results of other methods. Further, the results show that when the two populations taking the two tests are similar on the anchor score distributions, different equating methods yield the same or very similar results, even though they have different assumptions.     

A Comparison of IRT Equating and Beta 4 Equating

Four equating methods (3PL true score equating, 3PL observed score equating, beta 4 true score equating, and beta 4 observed score equating) were compared using four equating criteria: first-order equity (FOE), second-order equity (SOE), conditional-mean-squared-error (CMSE) difference, and the equipercentile equating property. True score equating more closely achieved estimated FOE than observed score equating when the true score distribution was estimated using the psychometric model that was used in the equating. Observed score equating more closely achieved estimated SOE, estimated CMSE difference, and the equipercentile equating property than true score equating. Among the four equating methods, 3PL observed score equating most closely achieved estimated SOE and had the smallest estimated CMSE difference, and beta 4 observed score equating was the method that most closely met the equipercentile equating property.     

IRT Equating Methods

The purpose of this instructional module is to provide the basis for understanding the process of score equating through the use of item response theory (IRT). A context is provided for addressing the merits of IRT equating methods. The mechanics of IRT equating and the need to place parameter estimates from separate calibration runs on the same scale are discussed. Some procedures for placing parameter estimates on a common scale are presented. In addition, IRT true-score equating is discussed in some detail. A discussion of the practical advantages derived from IRT equating is offered at the end of the module.     

Comparison of Item Preequating and Random Groups Equating Using IRT and Equipercentile Methods

An item-preequating design and a random groups design were used to equate forms of the American College Testing (ACT) Assessment Mathematics Test. Equipercentile and 3-parameter logistic model item-response theory (IRT) procedures were used for both designs. Both pretest methods produced inadequate equating results, and the IRT item preequating method resulted in more equating error than had no equating been conducted. Although neither of the item preequating methods performed well, the results from the equipercentile preequating method were more consistent with those from the random groups method than were the results from the IRT item pretest method. Item context and position effects were likely responsible, at least in part, for the inadequate results for item preequating. Such effects need to be either controlled or modeled, and the design further researched before the item preequating design can be recommended for operational use.     

Bi-Factor MIRT Observed-Score Equating for Mixed-Format Tests

ABSTRACT

The main purposes of this study were to develop bi-factor multidimensional item response theory (BF-MIRT) observed-score equating procedures for mixed-format tests and to investigate relative appropriateness of the proposed procedures. Using data from a large-scale testing program, three types of pseudo data sets were formulated: matched samples, pseudo forms, and simulated data sets. Very minor within-format residual dependence in mixed-format tests was found after controlling for the influence of the primary general factor. The unidimensional IRT and BF-MIRT equating methods produced similar equating results for the data used in this study. When a BF-MIRT model is implemented, we recommend the use of observed-score equating instead of true-score equating because the latter requires an arbitrary approximation or reduction process to relate true scores on test forms.     

The Chain and Post-Stratification Methods for Observed-Score Equating: Their Relationship to Population Invariance

The Non-Equivalent-groups Anchor Test (NEAT) design has been in wide use since at least the early 1940s. It involves two populations of test takers, P and Q, and makes use of an anchor test to link them. Two linking methods used for NEAT designs are those (a) based on chain equating and (b) that use the anchor test to post-stratify the distributions of the two operational test scores to a common population (i.e., Tucker equating and frequency estimation). We show that, under different sets of assumptions, both methods are observed score equating methods and we give conditions under which the methods give identical results. In addition, we develop analogues of the Dorans and Holland (2000) RMSD measures of population invariance of equating methods for the NEAT design for both chain and post-stratification equating methods.     

Evaluating Equating Results: Percent Relative Error for Chained Kernel Equating

This article presents a method for evaluating equating results. Within the kernel equating framework, the percent relative error (PRE) for chained equipercentile equating was computed under the nonequivalent groups with anchor test (NEAT) design. The method was applied to two data sets to obtain the PRE, which can be used to measure equating effectiveness. The study compared the PRE results for chained and poststratification equating. The results indicated that the chained method transformed the new form score distribution to the reference form scale more effectively than the poststratification method. In addition, the study found that in chained equating, the population weight had impact on score distributions over the target population but not on the equating and PRE results.     

An Approach to Evaluating the Missing Data Assumptions of the Chain and Post-stratification Equating Methods for the NEAT Design

Two important types of observed score equating (OSE) methods for the non-equivalent groups with Anchor Test (NEAT) design are chain equating (CE) and post-stratification equating (PSE). CE and PSE reflect two distinctly different ways of using the information provided by the anchor test for computing OSE functions. Both types of methods include linear and nonlinear equating functions. In practical situations, it is known that the PSE and CE methods will give different results when the two groups of examinees differ on the anchor test. However, given that both types of methods are justified as OSE methods by making different assumptions about the missing data in the NEAT design, it is difficult to conclude which, if either, of the two is more correct in a particular situation. This study compares the predictions of the PSE and CE assumptions for the missing data using a special data set for which the usually missing data are available. Our results indicate that in an equating setting where the linking function is decidedly non-linear and CE and PSE ought to be different, both sets of predictions are quite similar but those for CE are slightly more accurate .     

Equating With Miditests Using IRT

The equating performance of two internal anchor test structures—miditests and minitests—is studied for four IRT equating methods using simulated data. Originally proposed by Sinharay and Holland, miditests are anchors that have the same mean difficulty as the overall test but less variance in item difficulties. Four popular IRT equating methods were tested, and both the means and SDs of the true ability of the group to be equated were varied. We evaluate equating accuracy marginally and conditional on true ability. Our results suggest miditests perform about as well as traditional minitests for most conditions. Findings are discussed in terms of comparability to the typical minitest design and the trade‐off between accuracy and flexibility in test construction.     

The Effect of Repeaters on Equating

Test equating might be affected by including in the equating analyses examinees who have taken the test previously. This study evaluated the effect of including such repeaters on Medical College Admission Test (MCAT) equating using a population invariance approach. Three-parameter logistic (3-PL) item response theory (IRT) true score and traditional equipercentile equating methods were used under the random groups equating design. This study also examined whether or not population sensitivity of equating by repeater status varies depending on other background variables (gender and ethnicity). The results indicated that there was some evidence of repeaters' effect on equating with varying amounts of such effect by gender.     

Evaluating Equating Accuracy and Assumptions for Groups That Differ in Performance

Accurate equating results are essential when comparing examinee scores across exam forms. Previous research indicates that equating results may not be accurate when group differences are large. This study compared the equating results of frequency estimation, chained equipercentile, item response theory (IRT) true‐score, and IRT observed‐score equating methods. Using mixed‐format test data, equating results were evaluated for group differences ranging from 0 to .75 standard deviations. As group differences increased, equating results became increasingly biased and dissimilar across equating methods. Results suggest that the size of group differences, the likelihood that equating assumptions are violated, and the equating error associated with an equating method should be taken into consideration when choosing an equating method.     

A Comparison of Chained Linear and Poststratification Linear Equating Under Different Testing Conditions

In this study I compared results of chained linear, Tucker, and Levine-observed score equatings under conditions where the new and old forms samples were similar in ability and also when they were different in ability. The length of the anchor test was also varied to examine its effect on the three different equating methods. The three equating methods were compared to a criterion equating to obtain estimates of random equating error, bias, and root mean squared error (RMSE). Results showed that, for most studied conditions, chained linear equating produced fairly good equating results in terms of low bias and RMSE. Levine equating also produced low bias and RMSE in some conditions. Although the Tucker method always produced the lowest random equating error, it produced a larger bias and RMSE than either of the other equating methods. As noted in the literature, these results also suggest that either chained linear or Levine equating be used when new and old form samples differ on ability and/or when the anchor-to-total correlation is not very high. Finally, by testing the missing data assumptions of the three equating methods, this study also shows empirically why an equating method is more or less accurate under certain conditions .