Seventy three items of IPA (science) test used in the 2006 University of Gadjah Mada under‐graduate admission procedure (Utul UM UGM) were analyzed to reveal the effect of eliminating items based on corrected item‐total correlation coefficient on reliability of the test. Computations of reliability estimates for four different lengths of test (n=11.287) showed that reducing less discriminating items did not lower reliability coefficients as predicted by Spearman‐Brown prophecy. In fact, length of test was practically irrelevant to reliability coefficient when the test consisted of highly discriminating items.
                        
                        
                        
                        
                            
                                Copyrights © 2009