Issues in data collection: missing data and the 2001 New Zealand census

Abstract

Missing data plagues all surveys, and to a degree the New Zealand Census suffers from the same malaise. While it is not a high level of missingness, it is present. If not correctly dealt with; just deleting cases with missing data will lead to biased conclusions, particularly if the missingness mechanism is NMAR. Some missing data may be inevitable; sometimes a respondent may be incapable of answering a question. This is usually MAR. If however the respondent refuses to answer a question because of say having a high income, then the results of the income question will be biased. Over time there have been a growing number of people employing avoidance tactics so as not to be classified as a refusal, but to make enumeration just too difficult. Anecdotal evidence among enumerators shows that this accounts for about 5% of respondents.