Possible biases of temporal and regional comparisons How to handle item non-response?

Abstract

Abstract The main question of this paper is how to handle missing data in spatial and temporal analysis. We argue here, that in the case of spatial and temporal data, there may be several unique aspects that can lead to different response structures between datasets. In the empirical analysis, we have investigated the data of seven waves and 14 countries of ESS (N = 185 049) regarding the volume on item non-response per wave and country in the case of anti-regime attitudes. The ESS data has showed great deviation in the volume of missing answers of anti-regime attitudes. In the case of Western and Scandinavian countries higher rates, and in the case of Eastern and Mediterranean countries lower rates of valid answers have been measured. Based on the multi-level analysis countries and waves have been responsible for around 3 percent of the difference in non-response. At the respondent level, a lower response rate was more typical in the case of low social status, female respondents and older people. We have treated non-responses with different methods such us complete case analyis or nearest neighbour imputation. The difference was not extremely high between the methods, but overall the imputed anti-regime index was higher than the basic version. Missing data is an everyday component of our analysis. The non-treatment of non-response is also some kind of treatment. This non-intentional treatment often comes from the default setting of the statistical programmes used. If we do not want to be led by software we have to develop clear protocols for how to handle missing data. This could be an important foundation of reliable and valid data analysis. Keywords: non-response, imputation, ESS, multilevel analysis, anti-regime attitude