When individual items from a multi-item scale are missing there are problems in calculating scores for the summated scale. In such cases methods have been developed to impute the most likely values for these items. Such methods are no substitute for real observations but merely a device to facilitate analysis. The objective of imputation is to replace the missing data by estimated values which preserve the relationships between items, and which reflect as far as possible the most likely true value. If properly carried out, imputation should reduce the bias that can arise by ignoring non-response. By filling in the gaps in the data, it also restores balance to the data and permits simpler analyses. There are several approaches that can be used for imputation but the final choice is likely to depend on the particular context.
Was this article helpful?