The telephone survey methodology conference II and the resulting book, Advances in Telephone Survey Methodology, have shown problems and solutions to the use of random-digit-dialing (RDD) and telephone to conduct surveys.
Others have used growing challenges with RDD, both cost and coverage related, as arguments for the use of entirely new approaches to current studies.
Three companies have made such research public. Nielsen, Knowledge Networks, and in a recent study, Gallup have all presented some results suggesting at least higher response rates using mail through address-based designs. All three, however, have one major common feature in common: they use sampling to mail packets in order to recruit sample members into probability based panels.
It is not surprising, therefore, that all find higher response rates using an address based sampling frame - one could expect that using the same mode for contact and enrollment would perform better compared to using different modes and more contacts in another approach. More surprising is that for panels that require mailing of a recruitment package, mail has not been the mode of initial contact sooner.
Two questions remain. The main question is what does this mean for cross sectional surveys? A guess would be "not much." Further research into the conditions that make each sampling and each data collection design desirable is needed.
Second, we need more than response rates to judge designs. For example, a higher response rate in a mail survey compared to a comparable telephone survey may produce higher bias - whether due to literacy levels, reasons for cooperation, or something else. These are empirical questions that can be addressed and are critical to such major design decisions.
It is when there is a combination between a seemingly superior sampling frame and a needed method of data collection that require different modes that the tradeoffs become uncertain. This is one such experiment that Karol Krotki and I at RTI are currently conducting and look forward to sharing results from.