Monday, 6 October 2003

Researchers from Linköpings Universitet in Sweden have been working with data from both web surveys and telephone interviews in Denmark, Norway, Sweden and Finland provided by Zapera to estimate the usefulness of various weighting methods applied to data collected via the Internet.

The results of the study was presented last week at the annual conference of the American Association for Public Opinion Research in Nashville, USA.

The results show that in the Nordic countries, a weighting procedure called propensity weighting (where the data are weighted according to the likelihood of each respondent being an internet user) does not improve on standard weighting schemes using demographic variables for web surveys.

Zapera has previously shown that we produce very accurate public opinion surveys using internet-based data collection. We believe that propensity weighting is unlikely to improve on that accuracy for two reasons:

First of all, the vast majority of the Nordic populations are internet users, therefore there is little point in exploring the differences between the online and offline populations (with the possible exception of the oldest segments of the population).

Secondly, within each country, the Nordic populations are much more homogenous than for example the US population where propensity weighting is applied with some success. Because of the homogeneity, there are few defining differences in the population that can be explained by access to the internet - and consequently it is difficult to use these differences to weight the collected data.

The team at Linköpings Universitet consists of Maria Varedian and Gösta Forsman at the Department of Statistics.