Measurement and adjustment of non-response bias based on non-response surveys: the case of Belgium and Norway in the European Social Survey Round 3

Abstract

In earlier rounds of the European Social Survey, non-response bias was studied by using population statistics and call record information (para data). In the third round, a new feature was introduced: two kinds of non-respondent surveys were set up using a short list of questions which were designed to study non-response bias. In Belgium, a very short questionnaire was offered to all refusals at the doorstep (doorstep questions survey, DQS). In Norway and two other countries, somewhat longer versions of the basic questionnaire were offered to all main survey non-respondents and to samples of respondents (non-response survey, NRS). Logistic regression models were applied in order to estimate response propensities. This paper shows that propensity score weighting adjustment of non-response bias, on the basis of key socio-demographic and attitudinal variables, is effective for most demographic and non-demographic variables in both Belgium and Norway. Application of the weighting procedure balances the samples of cooperative respondents and non-respondents according to the key variables studied since systematic differences between cooperative respondents and non-respondents have disappeared.