multifaceted Rasch measurement (MFRM)

The application of Multifaceted Rasch Measurement (MFRM) in rating test takers’ oral language proficiency has been investigated in some previous studies (e.g., Winke, Gass, & Myford, 2012). However, little research so far has ever documented the effect of test takers’ genders on their oral performances and few studies have investigated the relationship between the impact of raters’ gender on the awarded scores to male and female test takers. Thus, this study aimed to address the above-mentioned issue. Twenty English as a Foreign Language (EFL) teachers rated the oral performances of 300 test takers. The outcomes demonstrated that test takers’ gender differences did not have any significant role in their performance differences when they were rated by the raters of the same or opposite gender. The findings also reiterated that raters of different genders did not demonstrate bias in rating test takers of the opposite or same gender. Moreover, no significant difference was observed regarding male and female raters’ biases towards the rating scale categories. The outcomes of the study showed that both male and female raters assign fairly similar scores to test takers. This suggests no evidence based on which either male or female raters must be excluded from the rating process. The findings imply that there is no need to worry about the impact of gender for a more valid and reliable assessment.

Performance testing including the use of rating scales has become widespread in the evaluation of second/foreign oral language assessment. However, no study has used Multifaceted Rasch Measurement (MFRM) including the facets of test takers’ ability, raters’ severity, group expertise, and scale category, in one study. 20 EFL teachers scored the speaking performance of 200 test-takers prior and subsequent to a rater training program using an analytic rating scale consisting of fluency, grammar, vocabulary, intelligibility, cohesion, and comprehension categories. The outcome demonstrated that the categories were at different levels of difficulty even after the training program. However, this outcome by no means indicated the uselessness of the training program since data analysis reflected the constructive influence of training in providing enough consistency in raters’ rating of each category of the rating scale at the post-training phase. Such an outcome indicated that raters could discriminate the various categories of the rating scale. The outcomes also indicated that MFRM can result in enhancement in rater training and functionality validation of the rating scale descriptors. The training helped raters use the descriptors of the rating scale more efficiently of its various band descriptors resulting in a reduced halo effect. The findings conveyed that stakeholders had better establish training programs to assist raters in better use of the rating scale categories of various levels of difficulty in an appropriate way. Further research could be done to make a comparative analysis between the outcome of this study and the one using a holistic rating scale in oral assessment.

Due to subjectivity in oral assessment, much concentration has been put on obtaining a satisfactory measure of consistency among raters. However, obtaining consistency might not result in valid decisions. One matter that is at the core of both reliability and validity in oral performance is rater training. Recently, Multifaceted Rasch Measurement (MFRM) has been adopted to address the problem of rater bias and inconsistency; however, no research has incorporated the facets of test takers’ ability, raters’ severity, task difficulty, group expertise, scale criterion category, and test version together in a piece of research along with their two-sided impacts. Moreover, little research has investigated how long rater training effects last. Consequently, this study explored the influence of the training program and feedback by having 20 raters score the oral production, as measured by the CEP (Community English Program) test, produced by 300 test takers in three phases, i.e., before, immediately after and long after the training program. The results indicated that training can lead to higher degrees of interrater reliability and diminished measures of severity/leniency, and biasedness. However, it won't lead the raters into total unanimity, except for making them more self-consistent. Although rater training might result in higher internal consistency among raters, it cannot eradicate individual differences. That is, experienced raters, due to their idiosyncratic characteristics, did not benefit as much as inexperienced ones. This study also showed that the outcome of training might not endure in long run after training; thus, it requires ongoing training letting raters regain consistency.