Citation

Armoogum J, Madre JL. Transp. Res. Circular 2000; (E-C008): 19 p..

Copyright

(Copyright © 2000, U.S. National Academy of Sciences Transportation Research Board)

DOI

unavailable

PMID

unavailable

Abstract

All surveys contain incomplete data, even if measures are implemented in order to avoid them before and during the data collection. The problems with nonresponse are that estimators are less accurate (the sample size is lower) and that it creates bias (generally, respondents behave differently than nonrespondents). Different types of nonresponse may occur: unit nonresponse refers to the failure of a unit in the sample frame to participate in the survey (noncoverage, not at home, unable to answer, refusal...); and item nonresponse occurs (a) when most of the questions for a unit are answered, but no answer is given for some questions, or (b) when the answer is clearly wrong and must be deleted, or (c) due to under-reporting (questions that are sensitive or for which the respondent does not have the information, boring document to fill out...). As it appears in the definition, the difference between unit and item nonresponse is that we have less information about a unit nonrespondent than for an item nonrespondent. Indeed, for a selected unit, the only information that we can get without error is geographical, and even if the sample is picked from the census, the figures are not necessarily up to date. We know much more about the units that do not reply to some questions (from the responses given to the other questions). This workshop resource paper focuses on the correction of item nonresponse. First, it gives a general presentation of the theory on this subject, then it talks about the existing solution and illustrates the presentation with examples drawn from the last two National Personal Transportation Surveys conducted in France.