Do talk money -- Reducing income nonresponse in surveys

Missing data, in particular nonresponse, are a constant cause of sorrow in surveys. Questions that are especially prone to item nonresponse are questions about monetary values, like income. Over the last decades, several methods to reduce nonresponse were developed and tested. Evidently, there is not one solution that fits all. Being members of the research team of the international OeNB Euro Survey, we will use an experiment, testing a new approach to reduce income nonresponse. The main reason to test new approaches is that the current approach, which will be our control treatment, works well in some country samples but barely in others. Thus, we restrict the experiment to four countries --Bosnia and Herzegovina, Bulgaria, Hungary and Poland -- in which nonresponse is particularly high and current approaches to reduce it have not been fruitful.

In the survey, respondents are initially asked to provide exact amounts on, for example, monthly household income. Currently, if they are then not willing or not able to give an exact amount, i.e. nonresponse occurs, they are offered to choose their income from a list of very granular income brackets. These brackets can amount to over 30 in some countries. We want to test a treatment in which -- instead of presenting respondents these granular brackets after their first income refusal -- we simply ask them if their income falls into the first, second or third income tercile, which is predefined by us. The idea is that (i) for many research projects pursued with the OeNB Euro Survey, being able to categorize people in low, medium or high income is sufficient and (ii) respondents are more likely to answer questions that are cognitively less demanding and less privacy-invading.