This paper describes a plausible values imputation approach for deriving population estimates on several language proficiency domains. The approach harnessed a multi-dimensional item response analysis combining student responses, rater judgements and student background variables. The target population comprised lower grade primary school students enrolled in the Hong Kong schooling system. The raters were local teachers of English employed within the sampled target schools. The consideration of student background variables in conjunction with student responses and rater judgements facilitated the process of "conditioning" (Reckase, 2002). The primary objective of this research was to impute plausible values for students where no data was provided or where rater data was deemed suspect. By necessity, a secondary objective of this study was to establish rules for justly excluding particular data on the basis of questionable validity. Surveys such as TIMSS, PISA and NAEP have used such "plausible value" methodologies to account for incomplete test designs and person non-response (Beaton & Johnson, 1990; Yamamoto & Kulick, 2000; Adams & Wu, 2002). The point of difference in this study was the use of item response theory (in particular plausible values imputation) to identify and correct for invalid rater judgements in a large-scale educational survey.
[Copyright of Australian Association for Research in Education (AARE) at http://www.aare.edu.au]