Item non-response when measuring elder abuse: influence of methodological choices.

BACKGROUND Efficient actions to fight elder abuse are highly dependent on reliable dimensions of the phenomenon. Accurate measures are nevertheless difficult to achieve owing to the sensitivity of the topic. Different research endeavours indicate varying prevalence rates, which are explained by different research designs and definitions used, but little is known about measurement errors such as item non-responses and how outcomes are affected by modes of administration. METHODS A multi-national study was developed to measure domestic abuse against home-dwelling older women (aged >60 years) in Europe. The measurement instrument covered six forms of abuse, adapted from the Conflict Tactics Scale. 2880 individuals were interviewed by three different data collection methods (i.e. postal, face-to-face, telephone). RESULTS Principal component analysis of missing values of 34 indicators of abuse showed various patterns of item non-response. Moreover, principal component analysis indicated several response patterns across different types of data collection. A binary logistic regression explained that item non-response and abuse prevalence is influenced by individual characteristics (social status, vulnerability), method effects such as content (sensitivity), the order of the questions (forms of abuse), by type of data collection and the presence of assistance in survey completion. CONCLUSION The discussion revolves around how these findings could help improving measuring elder abuse. Advantages and disadvantages of the questionnaire and type of data collection methods are discussed in relation to three potential types of response errors: item positioning effect, acquiescence and social desirability.