Predicting and Applying Differential Response Rates in Address-based Sampling for a Household Travel Survey

In recent years, the primary sampling method used for household travel surveys has shifted steadily from random digit dial telephone contact to address-based sampling based on an initial contact by mail. One advantage in the use of address-based sampling is that researchers can control the geographic distribution of households that are invited to participate in the survey and can design a sampling plan that is based on publicly available data about the demographic and geographic distribution of households in the region. This paper describes how response rates from a previous survey were used to anticipate how response rates for a household travel survey for the Puget Sound Regional Council in Washington State would vary across census block groups as a function of the percentage of low-income households in those block groups. The anticipated response rates were used to vary the number of households invited to participate from each block group to obtain a more representative sample. An analysis of the resulting Puget Sound survey data estimated a model for predicting final response rates as a function of published population statistics for the block group level. In addition to income, block group factors found to be significant predictors were household size distribution, owner versus renter fraction, vehicle ownership distribution, and commute mode shares for the nonauto modes. The transferability of the method was tested with data from a subsequent household travel survey done in Anchorage, Alaska. A model based on Puget Sound data only was used to predict response rates in the Anchorage data with success.