ESTIMATING MODE EFFECTS BY THREE METHODS IN THE SWEDISH SURVEY ON INCOME AND LIVING CONDITIONS

The target estimate differences between the web and telephone respondents in the Swedish Survey on Income and Living Conditions can either be due to differences in compositions or measurement, known as selection effects and measurement effects respectively. Three methods are reviewed and used to estimate the selection and measurement effects for 133 target variables of the survey in this thesis. The assumptions of the methods are also discussed in relation to the situation under investigation. The number of significant selection effects and measurement effects varies by method. The two calibration methods that use weights lead to the conclusion that the majority of target variables are affected by selection effects and measurement effects. A minority of target variables are affected by either effect according to the third method that does not use weights. All methods share the conclusion that 23 target variables are affected by measurement effects, and that 33 are not affected. All methods share the conclusion that 28 variables are affected by selection effects, and that 12 variables are not affected. The largest estimated measurement effect is 18.77 %, and the largest selection effect is -14.36 %.