Use of Paradata in Non-Response Bias Adjustment: Application of Sequential Propensity Weighting on Dutch Reluctant Respondent Sample in the European Social Survey

This paper explores the use of European Social Survey (ESS) Round 5 paradata Z-variables (eg. type of dwelling and neighborhood characteristics, and characteristics of contact procedure variables) to adjust survey estimates for non-response. By taking reluctant respondents - as opposed to cooperative respondents, as proxies for all types of non-respondents - this paper obtains calibrated propensity weights in sequential steps. Two types of propensity scores (i.e. focus is on ‘contactability’ and ‘survey cooperation’) are obtained through logistic regression models and multiplied with each other in order to obtain nonresponse weights for reluctant respondents based on sub-group stratification methods. ESS Round 5 Dutch sample is used as a case study because of high number of reluctant respondents sample (N=460). This paper illustrates sequential steps to transform reluctant respondents sample that are representative of all types of non-respondents through sequential propensity weights. The effectiveness of propensity weights are studied through level of standardized bias, level of correlations between Z-R and Z-Y variables as well as effects of multiple regression models on attitudinal variables tested between unweighted and weighted samples. These evaluative items are also compared with the effects of post-stratification weights based on key socio-demographic variables (eg. age, gender, region). The paper addresses implications of data quality in paradata, fieldwork design on refusal conversion activities and use of reluctant respondent sample  for non-response adjustments in cross-national research.