Handling incomplete quality-of-life data

Incomplete data sets are often encountered in the analysis of quality-of-li fe (QOL) data. The incompleteness arises from two major sources, namely, mi ssing responses and artificial quantification of response categories. Shen and Lai (1998a) propose using Optimal Scaling (OS) to tackle the problem. T he OS method based on numerical iterative approach attempts to restore the continuous property of the measurements and provide estimates for missing r esponses. However, the OS leads to convergence problem when there are many missing values in the data set; and it incorporates no mechanisms to provid e the standard errors of the mean estimates when missing values are filled. Hot-deck imputation is therefore suggested. This paper presents a simulati on study to show that the random hot-deck imputation yields reasonable esti mates for the population mean and generally preserves the distribution of t he population. In addition, when applying the random hot-deck imputation, v alid estimates for the standard error of the mean estimate can be obtained using the variance formula due to Lai (1998). With hot-deck imputation taki ng care of the missing responses and OS quantifying the response categories , it is postulated that the problem of data incompleteness can be more sati sfactorily handled. By applying the proposed techniques to real survey data , this paper also presents the change of the QOL of Hong Kong residents in the last decade leading to the turning point of the metropolis in 1997.