Detecting Rater Biases in Sparse Rater-Mediated Assessment Networks

Human ratings take time: A hierarchical facets model for the joint analysis of ratings and rating times

Performance assessments increasingly utilize onscreen or internet-based technology to collect human ratings. One of the benefits of onscreen ratings is the automatic recording of rating times along with the ratings. Considering rating times as an additional data source can provide a more detailed picture of the rating process and improve the psychometric quality of the assessment outcomes. However, currently available models for analyzing performance assessments do not incorporate rating times. The present research aims to fill this gap and advance a joint modeling approach, the "hierarchical facets model for ratings and rating times" (HFM-RT). The model includes two examinee parameters (ability and time intensity) and three rater parameters (severity, centrality, and speed). The HFM-RT successfully recovered examinee and rater parameters in a simulation study and yielded superior reliability indices. A real-data analysis of English essay ratings collected in a high-stakes assessment context revealed that raters exhibited considerably different speed measures, spent more time on high-quality than low-quality essays, and tended to rate essays faster with increasing severity. However, due to the significant heterogeneity of examinees' writing proficiency, the improvement in the assessment's reliability using the HFM-RT was not salient in the real-data example. This discussion focuses on the advantages of accounting for rating times as a source of information in rating quality studies and highlights perspectives from the HFM-RT for future research on rater cognition.

Modeling Rating Order Effects Under Item Response Theory Models for Rater-Mediated Assessments

Rater effects are commonly observed in rater-mediated assessments. By using item response theory (IRT) modeling, raters can be treated as independent factors that function as instruments for measuring ratees. Most rater effects are static and can be addressed appropriately within an IRT framework, and a few models have been developed for dynamic rater effects. Operational rating projects often require human raters to continuously and repeatedly score ratees over a certain period, imposing a burden on the cognitive processing abilities and attention spans of raters that stems from judgment fatigue and thus affects the rating quality observed during the rating period. As a result, ratees' scores may be influenced by the order in which they are graded by raters in a rating sequence, and the rating order effect should be considered in new IRT models. In this study, two types of many-faceted (MF)-IRT models are developed to account for such dynamic rater effects, which assume that rater severity can drift systematically or stochastically. The results obtained from two simulation studies indicate that the parameters of the newly developed models can be estimated satisfactorily using Bayesian estimation and that disregarding the rating order effect produces biased model structure and ratee proficiency parameter estimations. A creativity assessment is outlined to demonstrate the application of the new models and to investigate the consequences of failing to detect the possible rating order effect in a real rater-mediated evaluation.

The Effects of Rating Designs on Rater Classification Accuracy and Rater Measurement Precision in Large-Scale Mixed-Format Assessments

In standalone performance assessments, researchers have explored the influence of different rating designs on the sensitivity of latent trait model indicators to different rater effects as well as the impacts of different rating designs on student achievement estimates. However, the literature provides little guidance on the degree to which different rating designs might affect rater classification accuracy (severe/lenient) and rater measurement precision in both standalone performance assessments and mixed-format assessments. Using results from an analysis of National Assessment of Educational Progress (NAEP) data, we conducted simulation studies to systematically explore the impacts of different rating designs on rater measurement precision and rater classification accuracy (severe/lenient) in mixed-format assessments. The results suggest that the complete rating design produced the highest rater classification accuracy and greatest rater measurement precision, followed by the multiple-choice (MC) + spiral link design and the MC link design. Considering that complete rating designs are not practical in most testing situations, the MC + spiral link design may be a useful choice because it balances cost and performance. We consider the implications of our findings for research and practice.

A mixture Rasch facets model for rater's illusory halo effects

A rater's overall impression of a ratee's essay (or other assessment) can influence ratings on multiple criteria to yield excessively similar ratings (halo effect). However, existing analytic methods fail to identify whether similar ratings stem from homogeneous criteria (true halo) or rater bias (illusory halo). Hence, we introduce and test a mixture Rasch facets model for halo effects (MRFM-H) that distinguishes true halo versus illusory halo effects to classify normal versus halo raters. In a simulation study, when raters assessed enough ratees, MRFM-H accurately identified halo raters. Also, more rating criteria increased classification accuracy. A simpler model ignored halo effects and biased the parameters for evaluation criteria and for rater severity but not for ratee assessments. MRFM-H application to three empirical datasets showed that (a) experienced raters were subject to illusory halo effects, (b) illusory halo effects were less likely with greater numbers of criteria, and (c) more informative survey responses were more distinguishable from less informative responses.

Detecting Differential Rater Functioning in Severity and Centrality: The Dual DRF Facets Model

Performance assessments heavily rely on human ratings. These ratings are typically subject to various forms of error and bias, threatening the assessment outcomes' validity and fairness. Differential rater functioning (DRF) is a special kind of threat to fairness manifesting itself in unwanted interactions between raters and performance- or construct-irrelevant factors (e.g., examinee gender, rater experience, or time of rating). Most DRF studies have focused on whether raters show differential severity toward known groups of examinees. This study expands the DRF framework and investigates the more complex case of dual DRF effects, where DRF is simultaneously present in rater severity and centrality. Adopting a facets modeling approach, we propose the dual DRF model (DDRFM) for detecting and measuring these effects. In two simulation studies, we found that dual DRF effects (a) negatively affected measurement quality and (b) can reliably be detected and compensated under the DDRFM. Using sample data from a large-scale writing assessment (N = 1,323), we demonstrate the practical measurement consequences of the dual DRF effects. Findings have implications for researchers and practitioners assessing the psychometric quality of ratings.