Abstract

Traditionally, K-MMSE, a psychological assessment tool used for dementia screening, has been used to evaluate cognitive ability based on summative scores. However, aspects such as whether the relationship between cognitive ability and test scores is assumed to be linear or non-linear and whether item-weight is considered or not may lead to different test scores and different classification. Use of the total score as a test score requires linear relationship and item unweighting, but many psychological tests do not meet these assumptions. The current study examined similarity and classification agreements among test scores derived from different scoring methods, by using K-MMSE data sourced from 6,548 middle-aged and older adults. The Pearson correlation coefficients were high between the scores based on the classical test theory with linearity assumption and between the scores based on the item response theory with nonlinearity assumption. The unweighted scores of total and partial credit model were completely consistent in their classification despite the inconsistency in linearity assumption, but the weighted scores from factor analysis and generalized partial credit model, had the lowest classification agreement. We also found that the greater the asymmetry in the distribution of the total score, the lower the similarity of test scores and classification agreement based on different scoring methods. Lastly, it was emphasized that the selection of appropriate scoring methods should be consistent with the objectives of the test.