Imputation procedures for quality of life questionnaires with missing items

• Main Authors: CHAO HSIEN LEE, 李昭憲
• Other Authors: 黃景祥
• Format: Others
• Language: en_US
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/91716309751756970208

碩士 === 國立臺灣大學 === 流行病學研究所 === 88 === Abstract Our aim of this study was to provide a better and reasonable imputation procedure for quality of life questionnaires with missing items. For estimating missing items in questionnaires and expecting to reduce and control the bias due to informative missing, we use the generalized logit model with those explanatory covariates, which were obtained from the other available items related to the missing items and the subject’s characteristics. The data from the WHOQOL-Taiwan Group for the development of the Taiwan Chinese version of the World Health Organization quality of life measure was utilized to explain this imputation procedure. That study period in Taiwan was from October 1997 to January 1998 and obtained 1066 effective questionnaires (two missing questionnaires). Taking the Q10 item in Taiwan version WHOQOL-BREF questionnaire from that study as an example, we compared our imputation procedure with WHOQOL’s procedure in above data. The results showed that WHOQOL’s procedure was not a suitable way and it did not consider that the missing items might be informative missing; besides they also showed that our procedure could get better and reasonable imputations and it provided a better improvement than WHOQOL’s procedure. The advantages of this procedure in this paper are that it will not reduce data set for statistical analyses and will take into account informative missing; and it is easy to implement and practice; and the software for this method is available in statistical analysis packages. The disadvantage may be that if the probability of the response in some one question is too small in study, then it is difficult to estimate this response by using this procedure. However, this procedure provides a suitable and reasonable way for the imputations of missing items in effective questionnaires.