In this paper, we analyzed effects of defectives replies included in semantic differential data (SD data) on outcomes of analysis of the data. We prepared several data with different ratios of defective replies, through adding defective replies generated by a computer using random numbers to SD data obtained by an experiment. Applying Principal Component Analysis, we found that the proportion of variance of first principal axis declines rapidly and specific features of objects described by PCA become vague in proportion to the ratio of defective replies.
