The world of survey research is rapidly changing, moving from face-to-face (CAPI) and telephone interviews (CATI) to Internet interviewing. Several methods are used to attempt to survey the general public by Internet, including the use of volunteer and RDD samples. However, these methods may not ful l the demands on coverage, sampling and response posed by scienti c researchers. People in volunteer samples are often a rather selective part of the general public. In addition, because of a vast increase of mobile-only households and of declining response rates, RDD samples do not cover the population anymore.

A newly established Internet panel in the Netherlands uses a dierent design. A true probability sample of households drawn from a population register by Statistics Netherlands is contacted with a telephone or in-person interview, asking respondents to join the panel. The panel provides a computer and internet connection to those households that cannot otherwise participate (about 85% of the households have Internet access).

In the presentation an overview of this new panel concept is given. An experiment was carried out to determine the optimal recruitment strategy for the panel. The factors to be optimised were: contact mode, incentive amount, timing of the incentive, content of the information letter, and timing of the panel participation request. The experimental design took into account the naturally varying factor of whether or not the households involved had a known xed landline. The highest response rate was found with an incentive of 10 euro. The incentives of 20 euro and 50 euro did not substantially increase response rates beyond those seen at the 10 euro level. All incentives were found to have much stronger eects on response rates when they were enclosed in the announcement letter than when they were paid later. Another important factor was the contact mode that was used. Contact was made either by CATI or CAPI. While the contact rate was somewhat higher with CATI, the CAPI interviewers were more successful in obtaining panel participation. The contact rate was substantially lower in the subpopulation of households without a known xed landline, even when controlling for the eect of contact mode. The response rates were not aected by the content of the information letter. Response rates were also independent of whether the nature of the panel study was explained before or after the recruitment-interview was completed.

The panel design is evaluated by comparing the composition of the recruited panel to population statistics, to traditional face-to-face scienti c studies and to commercial access panels.