With the introduction of the Internet as a new data collection mode, traditional survey methodology is challenged fundamentally. The increasing popularity of web surveys created a demand for appropriate web survey methodology, thereby triggering a heated debate about the quality and reliability of web surveys for scientific use. The most obvious potential drawback of web surveys is that they may not be 114 representative of the population of interest because the sub-population with Internet access is quite specific.

In this context, propensity score adjustment (PSA) has been proposed to statistically surmount inherent problems, particularly in non-probability-based web surveys. In this procedure, a parallel probabilitybased reference survey is used to estimate the propensities of being in the web sample based on a vector of covariates (socio-demographic and ‘webographic’ (lifestyle) variables) measured in both samples. It is critical for the method to choose appropriate covariates. In the scientific community, however, this method has not been applied traditionally in the field of surveys. There has been a minimal amount of evidence for its applicability. The implications particularly for survey methodology still need to be studied more extensively.

Against this background, the paper will explore the efficiency of PSA in adjusting biases arising from nonrandomized sample selection. For that purpose, the un-weighted and weighted results from the Dutch volunteer web sample of the WageIndicator Survey 2009 will be compared with data from the Dutch LISS Panel that have been collected parallel. The advantage of this reference survey is that it provides a proper probability sample stemming from the same resource (e.g. the same questionnaire). Survey mode effects can be excluded as both questionnaires are completed individually on the computer. Finally, the application will also examine the sensitivity of the results, particularly with regard to changes in the specification of the propensity score.