33: Comparison of Hot-Deck and Neural-Network Imputation

Nonresponse in travel surveys is a matter of major concern. Response rates among travel surveys in the U.S. typically vary between twenty five and forty percent although cases as low as five percent have been observed (Zimowski et al., 1997, p. i). The observed trend is also toward a reduced response rate over time (Stopher and Metcalf, 1996, p. 14). In Europe, an international survey among thirteen countries revealed generally higher response rates than in the U.S., but six of the thirteen surveys reviewed still reported response rates below fifty one percent (de Heer and Moritz, 1997, p. 12). The major concern with nonresponse is that if non-respondents are different to respondents, using the respondents to represent the population introduces a bias. The reduced sample size also increases the variance of estimates of the mean from the sample. With high rates of nonresponse, these effects can be substantial.