Utilizing Fisher’s-Z Transformation for Item Selection

Abstract

The previous work has shown that item selection method based on the use of corrected item-total correlation larger than .3 as the criterion provided the least errors of including items with low corrected item-total correlation in the population and excluding items with high corrected item-total correlation in the population. However, such method did not address the fact that corrected item-total correlation fluctuated across samples. Therefore, in smaller samples, the method provided larger errors. The current article proposed a new method for item selection that took into account the fluctuations of corrected item-total correlation across samples. The method was a significant test of correlation coefficient with the null hypothesis stating that the corrected item-total correlation was larger than or equal to .3. Four simulations were conducted to evaluate the proposed method and its modification. The results showed that the method was performed very well in reducing errors of including items with low corrected item-total correlation even in smaller sample sizes. However, the errors of excluding items with high corrected item-total correlation were large, particularly in small sample size. The large exclusion error was due to the lack of power to reject the null hypothesis when sample size was small. In larger samples, the proposed method and its modification and the method used criterion of corrected item total correlation larger than .3 performed equally well.