If the tools to gather information affect data quality: violence against women survey case

National victimization surveys, and among them women's safety surveys, are recognized as an important tool for gaining insight into many aspects of specific crimes that cannot be measured solely on the basis of administrative data such as police records. The survey on "Women's safety", conducted in many countries around the world, aims to provide the main indicators of the prevalence of violence against women by perpetrators, inside and outside women's household. The Italian Women Safety Survey, published for the first time in 2006 and again in 2014, provides prevalence indicators at national and subnational level. In dealing with such delicate topics, both the training of the interviewers and the choice of the technique or mode used to collect the information are strategic to guarantee reliable information. The mode usually used for data collection are: face-to-face, telephone, mail or web. In Italy, the main mode used for this survey is Computer Assisted Telephone Interviewing (CATI), administered by continuously trained interviewers, before and during the survey. In the 2014 release, when the sample was designed to also provide the measure of violence against foreign women (VAW), the Computer Assisted Personal Interviewing (CAPI) mode was used for this target subpopulation. As a consequence, in the VAW survey the effect on estimates of the data collection mode is confounded with nationality. The present work aims to investigate the presence of an interviewer effect in the latest edition of the Italian survey on Women's Safety. In addition, to verify whether the chosen mode influences the results in the VAW survey, we used data from the Citizen's Safety Survey, published between the end of 2015 and the beginning of 2016, in which the CATI and CAPI interviews were made to all citizens regardless of their nationality. To study the interviewer effect in both surveys, multilevel models, and specifically multilevel models with heterogeneous variance, are applied to the indicators of violence against women in common between the two surveys, also to adjust for the mode effect.