Combining samples to offset nonresponse and respondent biases

Abstract

What if probabilistic-based sampling does not result in a representative sample? How can researchers overcome low respondent engagement and hypothetical choices that are perceived as being socially desirable? These questions are relevant regardless of the way primary data collection is conducted.A statistically sound sampling strategy still depends on individuals volunteering their participation. Even with extrinsic rewards, there is no guarantee the respondent will contribute an honest effort. This research reports on the data collection for a study investigating the acceptance of electric vehicles (EV) in Australia. Complementing the Western Australian Electric Vehicle Trial, this research focuses on household preferences and attitudes towards EV. The data set represents the last stage of data collection with four surveys (initially delivered to trial participants and later aimed at the broader community).An initial sample showed high interest in EV and environmentally friendly technologies, but higher education levels and higher socioeconomic status households were overrepresented. To compensate for the bias, a second sample was collected from an online panel (PureProfile). Although neither sample is representative of the population, the results from the pooled data are deemed more appropriate for understanding drivers of EV uptake in Western Australia and informing policy making accordingly.