COMPARATIVE ANALYSIS OF THE RELIABILITY OF MISSING SCORES IMPUTATION TECHNIQUES

Abstract

Examination boards commonly face problems of missing marks during examinations. Examinations data may go
missing due to various reasons including hectic logistics and adverse conditions under which the examination was
written. In educational assessments, a number of methods have been used to address this problem but their
reliability has not been fully explored. This study sought to compare the reliability of five techniques used by
members of the Southern Africa Association for Educational Assessment (Regression Analysis, Criterion Mean
Method, Same Percentile Position, Z Score Method, Standard Mark Calculation) and another technique used in
the United Kingdom (Absolute Standard Deviation Method) in order to recommend a more valid, reliable and
fairer technique. The study used Botswana General Certificate of Secondary Education (BGCSE) data from
Botswana Examinations Council (BEC) and National School Certificate (NSC) data from Umalusi1
. Scores wererandomly selected, deleted and then estimated using each of the specified techniques. Predicted scores were
compared to actual scores using Paired Sample T-test, RMSE and Cohen`s D statistic. The results revealed that
Criterion Mean Method (CMM) was superior since precision of its estimated scores was higher. Despite good
performance displayed by this method, the study identified limitations which could hinder its full potential. The
study developed an improved version of the CMM, tested its performance against the original version and the
improved version of CMM is recommended for estimating missing scores.