Abstract

A lot of studies are showing an increased tendency to use more than one mode of administration to collect data for a particular analysis ([10];[1];[16];[9]). Thus, understand if different data collection methods influence answers becomes a concern. The problem of self selection into different interview modes demands attention especially when the assignment to one or another collection method is not randomly controlled and respondents might self select in one data collection method over another. This paper shows an empirical case concerning the evalua- tion of two different data collection method: the CAWI (Computer Assisted Web Interview) method and the CATI (Computer Assisted Telephone Interview) method. If self-selection exists both mode of data collection and characteristics of the re- spondents influence answers. Hence, the mode effect may be confounded. In order to estimate an unbiased mode effect, this paper proposes a data driven multivariate approach to monitor self-selection that allows to disentangle interview’s modes ef- fects on answers from the effect of self-selection. We will work through the use of the monitoring system with an empirical case. In particular, we will use AlmaLaurea data and compare results of our approach to PS adjustment method that AlmaLaurea usually applies to control data quality as documented in various reports and analysis conducted by the AlmaLaurea Consortium .