Şans Başarısı Düzeltme ve Kayıp Veri Yöntemlerine Göre Test ve Madde İstatistiklerinin İncelenmesi

Abstract

The aim of this study is to examine test and item statistics according to chance success correction formulas and methods of coping with missing data. For this purpose, data sets of 500, 1000 and 1500 people consisting of 20 items were produced. Test and item statistics of these data sets were calculated. Chance success correction formulas were applied to the produced data sets, and then test and item statistics were calculated. New datasets with 5%, 10% and 20% loss rates were produced from the originally produced full datasets in accordance with the completely random loss (MCAR) mechanism. The test and item statistics of these missing data sets were calculated. Then, test and item statistics of the completed data sets were calculated by using listwise deletion (LD), regression imputation (RI) and expectation maximization (EM) methods, which are methods of dealing with missing data. The test and item statistics obtained from the data sets that have full, chance success correction formulas, have missing data and methods of coping with missing data have been examined. As a result of the research, it was found that chance success decreased test reliability, test variance and item discrimination index; It was understood that it increased the item difficulty index, the average difficulty of the test and the test average. The most important measure to be taken against chance success is to choose the test to be applied with the feature that will be least affected by chance success and to reduce the probability of finding the right answer in the items by chance. As a result of applying methods to deal with missing data, it was understood that the most consistent results with full data sets were obtained with the expectation maximization (EM) method. It can be recommended to use the expectation maximization method, especially in data sets where the rate of missing data increases.