Methods for addressing missing data and sample misrepresentation

Abstract

Missing data is an important issue in sampling, which creates complications as it leads to loss of precision and irrational inference. Therefore, finding adequate techniques to address the missing data problem is important. Deletion is the most commonly used technique but this may lead to strong biases. The application of imputation methods using these statistical techniques, both for item non-response and unit non-response, will be applied to two different sets of data. The first is a political survey which is subject to item non-response. Thus, imputation techniques are applied both for nominal and ordinal response. The second data consists of the financial characteristics of individual companies over a number of years. It is a time series dataset where period ranges from year 2008 to 2014, and each company has several instruments capturing both assets and liabilities. This data has unit non-response since information for some companies is present for some years but not in others. Another problem faced by data analysts in surveys and datasets is misrepresentation. This issue was eliminated in the political dataset by applying weighting.