Title: Evaluating the validity and reliability of multi item scales after multiple imputation
Authors: Oya Kalaycioglu - Bolu Abant Izzet Baysal University, Bolu (Turkey) [presenting]
Abstract: Various multiple imputation (MI) methods for handling missing items in multi-item scales were evaluated based on the real data collected from a questionnaire consists of 39 Likert type items and four subscales. For different MI strategies at item, subscale and scale levels, the mean of the sub-scale scores were compared with simulation studies using the bias and coverage as the performance parameters. Additionally, commonly used measures to ensure validity and reliability of the multi-item scales were assessed. MI of each item separately outperformed in terms of bias and coverage of the mean sub-scale scores, as well as the validity and reliability measures. When the number of incomplete items was too large overfitting problems occurred with this method, therefore two different techniques were proposed to reduce the number of predictors in the imputation model. First, the predictors in the imputation model were selected with forward selection approach and second, rather than using item scores, sub-scale scores were used as predictors when imputing an item. All methods were most sensitive when missing data were not at random.