Do we have to mix modes in probability-based online panel research to obtain more accurate results?

Online probability-based panels often apply two or more data collection modes to cover both the online and offline populations with the aim of obtaining results that are more representative of the population of interest. This study used such a panel to investigate how necessary it is, from the coverage error standpoint, to include the offline population by mixing modes in online panel survey research. This study evaluated the problem from three different perspectives: undercoverage bias, bias related to survey item topics and variable characteristics, and accuracy of online-only samples relative to nationally representative benchmarks. The results indicated that attitudinal, behavioral, and factual differences between the online and offline populations in Australia are, on average, minor. This means that, considering that survey research commonly includes a relatively low proportion of the offline population, survey estimates would not be significantly affected if probability-based panels did not mix modes and instead were online only, for the majority of topics. The benchmarking analysis showed that mixing the online mode with the offline mode did not improve the average accuracy of estimates relative to nationally representative benchmarks. Based on these findings, it is argued that other online panels should study this issue from different perspectives using the approaches proposed in this paper. There might also be an argument for (temporarily) excluding the offline population in probability-based online panel research in particular country contexts as this might have practical implications.